{ "measurement": { "model.layers.0.self_attn": [ { "accuracy": 0.9625965436234286, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9678215637410942, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.971806756269775, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9788545766649278, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9793922634501206, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9804003188190492, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9807515333553678, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9817895339801908, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9860740630936465, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9865958478399798, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9860498310605946, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9916444860006633, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9866391550994625, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9921946843939, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9930559393435129, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952436929176512, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9932846493014184, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9972721428095707, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9977074259720546, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.0.block_sparse_moe": [ { "accuracy": 0.9732868446429309, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9766402233784136, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.977124547585845, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9774567779447687, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9942990934329206, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.995622608700375, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959337343216726, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981134930337583, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983753152804351, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983684061321823, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985281605714638, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988690854270795, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9989460280109932, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991471209321293, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9992753936147845, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9993338510371119, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999389956755821, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.1.self_attn": [ { "accuracy": 0.9895521741066324, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9912984470015784, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9922750083925692, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951121252410016, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954221291026395, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955535021665702, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969244618305789, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9970591035824711, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9971195172003814, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9972227932456391, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9977360106820828, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997885128058873, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979572993776712, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981066563708928, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988435565763595, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990196620392932, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991567040392271, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994551963977037, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997391684921336, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.1.block_sparse_moe": [ { "accuracy": 0.9606681853336723, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9690563533651201, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9716872700343007, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.971689635887742, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9950581813341399, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956954962659725, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957136103521885, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982572483192933, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980539085165794, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.997830076342277, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985574609814457, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987847245096131, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990572743347577, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999146358847563, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9992532480146589, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992555242765535, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.99932952800928, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.2.self_attn": [ { "accuracy": 0.9986747719690596, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9987987262307128, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989271079968219, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993000586370104, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993383318130662, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993694328249579, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995464443743817, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.99958203178616, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996455565445725, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996579680308443, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999683125766368, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997002814596175, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997177764560451, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997387893243669, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9998386488751796, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9998633401011926, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9998802342000234, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999293612308975, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999963483329147, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.2.block_sparse_moe": [ { "accuracy": 0.9979903219316743, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9980614850371096, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.99827089643498, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983176760632839, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9990990189774158, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999171777845875, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992437136472865, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995244977640381, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995745113395204, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995497288243649, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996126801147751, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9997469194457057, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9998012355706458, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9998408313427047, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9998845646960514, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9998984296362323, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999649780693289, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.3.self_attn": [ { "accuracy": 0.9980552396430683, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982324439548201, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984646002627223, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989178762434198, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990825920034886, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991089924848271, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993600720588706, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993999113288928, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.99947097459134, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999493417484796, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995404018394, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995869249036523, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995957126421241, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996441812721681, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997698300101451, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9998156084471511, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9998243813055236, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998970429465102, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999491290766999, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.3.block_sparse_moe": [ { "accuracy": 0.9949940554184937, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9951768029407647, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9954277557369909, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9955065701965635, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.996869871233541, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971985704277789, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973091479072249, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982463020561753, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974509553407858, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984454006618378, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986044355992865, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999031003684075, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9989609552951726, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991088610580559, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9996810297276457, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9997085465392478, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997588502956205, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.4.self_attn": [ { "accuracy": 0.9972967651770696, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974843856670257, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978666345368287, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984683408524448, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986429782982572, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987212882475242, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991061093659435, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991476010273848, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992384723583726, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992638457777886, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993073860471333, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993946049484963, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994226516956635, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994813970728214, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996725771006344, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997383863274559, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997483704133437, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998616065958662, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999313762954875, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.4.block_sparse_moe": [ { "accuracy": 0.9969436725368723, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9970454031319701, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9973492554609517, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974166090833023, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9986001023023977, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987041072515574, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988047370693254, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9992938385786194, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9993353497120552, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9992980579906878, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999379041672096, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996415941116682, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996942032195971, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999804734386424, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.999816640867242, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9998347732444496, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999457535944346, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.5.self_attn": [ { "accuracy": 0.9968676134625352, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9971090792923382, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997606657457685, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982041733547751, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984755627382057, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985135724143076, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989455554229942, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990122396021003, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991143460624459, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991572305137341, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992556137394307, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.999316501621108, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993611851854152, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994227619471004, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996411589979145, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997053885553645, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997237921006193, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998455937851471, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999174550868798, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.5.block_sparse_moe": [ { "accuracy": 0.9963081814348698, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9964341371393713, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9968168411256844, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9969029752169981, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982650408974701, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984082522936851, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985329727029526, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991251688671151, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991995830113363, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991288485317981, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992370501338308, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995573530491368, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996215665320817, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9997626481628302, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9997722145290984, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999794342588039, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999330910493307, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.6.self_attn": [ { "accuracy": 0.9966330609259832, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9969035338267291, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974110292785457, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997983497671636, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984302648073552, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984498670091853, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989573567903502, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989953546074072, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990690652722198, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991156129062323, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992111954243706, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992699093050569, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993484446772767, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.999398202368818, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996335227287522, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996929599621677, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9997214389755085, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998302916976268, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999177880048304, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.6.block_sparse_moe": [ { "accuracy": 0.995240713616735, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9954056816518699, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9959276241470912, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9960424146418901, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977633773753616, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979550267116314, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981233946439859, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9988643339908633, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9989623321177389, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9988013056817612, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990204623747175, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994134214709782, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995058899481879, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996640307462689, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9997067323351266, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9997373901715036, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999126610097224, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.7.self_attn": [ { "accuracy": 0.9961220144392237, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9963543079670911, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968403121333962, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976507991147963, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980657207718315, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981215599568953, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987131300272027, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987646138546743, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988513575241852, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989048907576178, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990443593889818, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991102981332102, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991785735075706, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992409805571171, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995367727257728, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.999611815085365, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996580368223429, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997849040297105, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998997678002955, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.7.block_sparse_moe": [ { "accuracy": 0.9943126133569565, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9945061836073077, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9951588198983747, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9953032049325932, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9973452297149619, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975563511313674, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977669201754523, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.998667718637367, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9987756475916525, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.998667402879188, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988342336575059, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999323467774574, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994201998771658, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996392219756288, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9996524598291695, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996908570495802, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9999032393205102, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.8.self_attn": [ { "accuracy": 0.9954388511342633, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9956934546659651, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962596992202299, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973158640480649, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977114632431614, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997767567156667, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985478778712844, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986127329236631, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987208437125542, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987654328854804, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988741066649949, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989367107168379, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990399841580687, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990985804807248, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.999459833870213, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9995497279221461, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9996186237625998, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997596176184934, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998925895836835, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.8.block_sparse_moe": [ { "accuracy": 0.993631106948382, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9938440163815884, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9946001968714163, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9947691987097067, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.996986719314009, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972205798104895, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974681487196664, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984110905727568, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.998558060290586, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984481833924196, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986567949220587, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999165126006119, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9993110288602398, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994711265148668, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995164511997964, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.999553936302331, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997205795658183, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.9.self_attn": [ { "accuracy": 0.9947674547959315, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9950276219138974, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956190562561938, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968023610203282, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997292559747094, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973263334127536, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982705879343772, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9983696956677657, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9985300964204056, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986067552113367, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986390030458815, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987464528500201, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988508027439055, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989615267933052, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993523543801673, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994705644938605, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.999544896919121, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997166320053542, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998683035752803, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.9.block_sparse_moe": [ { "accuracy": 0.9925869634914163, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9928375411484587, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9937002430424878, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.993891159807773, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9965482443620107, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9968241762835532, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971099023915533, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982606592943499, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983972891564726, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982627716932544, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984834373340403, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991179339122027, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992417167124682, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995238353240941, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995485052012986, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9996016258343221, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998714373095458, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.10.self_attn": [ { "accuracy": 0.9940475344437322, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9943920918005077, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950225528109035, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963435168059445, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970028355509337, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970784172244174, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980732773087526, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998152698047067, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9982963014749417, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9983940164460555, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984870313783176, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998610365149369, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987148711133111, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988280780993304, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992818847638677, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994056744939084, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994840995319751, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996763892873117, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998560931114633, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.10.block_sparse_moe": [ { "accuracy": 0.9920175737015119, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9922901104253373, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9932711250393799, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9934894514123076, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.99626003525612, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9965556147630865, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9968806527505972, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981123898423424, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.998252061067002, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981229650959569, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983574001878304, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990517923816744, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991795361428032, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994850373217272, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9995121234556121, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995730112775498, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998581552327347, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.11.self_attn": [ { "accuracy": 0.9934805314742813, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9937838628324435, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943646950475675, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995799375930801, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967292731039619, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967909255493922, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979175156239715, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9979961068295923, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9981513488363769, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998257560269466, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9983583368900183, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984767232702947, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998561539751551, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998680782565651, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991980316316538, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993324453626931, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9994497155834994, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996276189975968, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998428093558208, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.11.block_sparse_moe": [ { "accuracy": 0.9909173489891385, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9912189592509285, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.992337570035536, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9925855617038906, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9957265793958581, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9960660822444448, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964252600421835, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978639117049936, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980339741848997, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978553660912439, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981214627807372, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9989129411375248, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990667578408887, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994227636412497, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9994419269159583, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9995077168844689, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998458115767693, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.12.self_attn": [ { "accuracy": 0.9917686909290129, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9921752114437128, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9930443805887511, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948978339885607, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995887960956775, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959483960716936, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974032816121747, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.997449425343228, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9976610530682496, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9977996975404063, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9979313277994821, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980792629373211, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982327033128393, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983721610654979, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990112435403525, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991796164126994, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9993100498855041, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995288230849454, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998052881541968, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.12.block_sparse_moe": [ { "accuracy": 0.9895865148619601, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9899200828942029, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9912518346682191, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9915526664668792, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.99509350408947, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9954823404305467, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959161388197619, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9975455704721083, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977432537860679, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.997537054709698, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978434042698753, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987515107832702, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998928543014795, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9993365859661839, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9993595887575984, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9994394639369076, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999822708905987, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.13.self_attn": [ { "accuracy": 0.9911092506280463, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.991470952248691, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9923026532326874, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9944681933909458, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954670431631568, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954973039168277, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972778833655053, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9972787155144799, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.997524677867707, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9976936330479619, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9976925554464718, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978609574578801, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980274074845702, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981858035469869, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989070211580072, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990914380116585, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9992827379365304, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995016032572541, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997998641372664, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.13.block_sparse_moe": [ { "accuracy": 0.9883661111081509, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9887656568990726, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9902348607337397, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9905575911612495, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9945765979539015, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9950097007431874, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9954830291435907, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.997285023172337, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9975041595937096, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9972767969225779, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976174363707143, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986185262249293, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988159490412248, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992635538204157, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9992900373594908, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9993766280117837, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9998029738957197, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.14.self_attn": [ { "accuracy": 0.9891743807024077, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.989487266756202, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9907550233857412, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9932013899455533, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994471483251178, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9946036815373717, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996772109071332, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9968495767220462, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9970892066944783, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9972656305732304, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9972468934781653, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974335272134723, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976619415254774, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978516163721093, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987169067455014, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989222493849842, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9991527431626118, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994187104503477, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997624427365066, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.14.block_sparse_moe": [ { "accuracy": 0.9868962290745816, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9873413354354469, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9890221976430008, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9893965574814693, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9938627574592829, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9943516237514192, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9948993778395417, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9969184819584418, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9971672618443048, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9969139960471933, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973004930101237, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984330855322218, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986581144621596, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991643833743996, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991947252021515, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992962759508948, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997766657393904, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.15.self_attn": [ { "accuracy": 0.9882417474628279, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.988587953983561, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9896858646779468, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9922046246448237, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9940176101425957, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994122916170837, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965320151777154, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9965681890515905, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.996869345575473, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9970720060549578, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9970427071675658, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972612578176746, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974721461951145, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976802335280043, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986001298560663, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988333610143807, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990940228499791, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993580545828751, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997478688130382, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.15.block_sparse_moe": [ { "accuracy": 0.9856745743829953, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9861758688562795, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9880300668117247, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9884370840399673, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9933144234512982, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9938455147080516, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9944444205489402, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9966445769011778, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9969145820880505, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9966429447828743, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970620555962485, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982967646424904, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998540190884868, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990909773746113, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9991253193400457, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9992340413155034, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997572243860064, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.16.self_attn": [ { "accuracy": 0.9867235258662779, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9874342915994164, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9884821629690889, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910547189953688, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9934211858719784, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9935370492013661, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960923553505716, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9962244026277116, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9964991838634504, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9967319519798222, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9967279650185159, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969730764982829, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971990974031781, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974250318482518, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984493423682196, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987022210476234, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9990005903037902, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992802056627912, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997222937913888, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.16.block_sparse_moe": [ { "accuracy": 0.9842136272866475, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9847673261047978, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9868488828896692, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9873034250677416, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9926322432921121, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9932153404592291, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.993876505275502, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9962922851432507, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9965936516497382, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9963002259181323, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967633245181978, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981236857435617, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983920868100157, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9989967492721533, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9990377484661478, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9991564014512972, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997318040749585, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.17.self_attn": [ { "accuracy": 0.9847091234926331, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9853325244529467, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9868762300741908, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9903858391886675, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9923570366257704, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9924453506491294, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9957137468736619, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9957784249778151, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.996097227681036, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9963107531473628, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9961498587600592, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963960979366675, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967176503391544, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969922561002406, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982042522966176, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984840034494961, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988774804162524, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991867634556624, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999693601612786, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.17.block_sparse_moe": [ { "accuracy": 0.9828835956654266, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9834717590557901, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.985705084878167, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9861929617135933, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9920127308241239, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9926424629152998, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9933568926920232, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9959822333368816, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9963063314115923, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9959948763226796, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964918108644748, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979711579906411, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982568971218394, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9989148268604798, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9989600416677269, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990904951256724, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9997102339096416, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.18.self_attn": [ { "accuracy": 0.9832921965831989, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9841675861542555, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9856679435868404, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9893186045075325, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9916711411022238, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.991798997427778, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950203751826561, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.995170508842229, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9955105377900365, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9958220351132026, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9958415032597259, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961322178316646, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9964160645062006, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966908091752741, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980196450794067, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983447280706299, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987179282531877, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991001479456348, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996526764451502, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.18.block_sparse_moe": [ { "accuracy": 0.9816961297275204, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9823195632841242, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9846987210606274, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.985227120991208, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9914386658684203, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9921165244947923, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9928777469077, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9956747815993271, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9960262642707676, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.995704876945207, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962372767591947, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978224652892861, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981305259524992, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998829951191223, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9988842460014368, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9990236391252103, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996837405740638, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.19.self_attn": [ { "accuracy": 0.9812268797602308, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9825085449571672, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9840973855339383, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9877422952926472, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9907477776961107, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9909214513180288, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99419003139299, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944127869522689, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949426774278676, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9952559863388735, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953780093254816, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995708269336702, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959924222019158, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963181032186472, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977878012608639, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981520876907254, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998513223064181, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989666873650429, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995992804424053, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.19.block_sparse_moe": [ { "accuracy": 0.9803707533466973, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.981047121464814, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9836145585500881, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9841817809072765, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9907862412449169, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9915185859671941, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9923320365824589, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9953562638624326, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.995727716001535, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9953730433174458, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959469387234238, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976529674481993, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997985328020724, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987415720120465, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9987968110608084, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9989445710272872, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996638624908979, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.20.self_attn": [ { "accuracy": 0.9806324012675568, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9816290827860173, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9836590823374296, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9875948193336004, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9904863843081617, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9906390084713501, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9942443099510121, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944280610958997, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9947666881949102, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950806024288269, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9952487375550462, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955451190961819, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959530187464368, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962620520351553, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977827864502998, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981233390587333, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984877089277449, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989301575314082, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995878681159076, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.20.block_sparse_moe": [ { "accuracy": 0.9784751721706829, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9792168057082515, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9820478213460822, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.982679029289437, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9898832169312396, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9906913966528679, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9915860077415249, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9948807165369784, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9952890542090723, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9949180920191697, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9955512848437617, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974234481219595, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977879000484551, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986105495114125, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9986814361947932, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9988441077263193, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996255598893377, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.21.self_attn": [ { "accuracy": 0.97778894801281, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9787377947451252, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.980802324012314, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9854495449383792, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9886755542584548, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9888842984004632, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9929838220490829, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9932049579403707, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.993861645570417, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994234320646348, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944017250948635, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947691651943483, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995170052518676, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955395026573617, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973230375111789, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977609812447458, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982239064932066, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987481890497484, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995155694019846, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.21.block_sparse_moe": [ { "accuracy": 0.9763605400527778, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9771703455695196, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.980221947017861, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9809080247130049, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9889208662421688, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9897979893476555, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9907570759961871, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9943979753818559, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9948441650424349, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9944383426152757, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9951241392838327, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971784347986901, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975740687406965, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984815360696399, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9985551564543704, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9987312049794893, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995911433744415, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.22.self_attn": [ { "accuracy": 0.9753568483222472, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9766904945044141, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9792163421056772, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9847782399554393, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9877863113621348, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9879692767893797, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9927874220241057, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9929911446586055, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.993418696128126, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9938376125585484, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9938737924553847, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943113790604433, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948075910368445, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995222565705741, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971429462189247, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975992357909778, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981022419218023, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9986963707916928, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994956337024249, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.22.block_sparse_moe": [ { "accuracy": 0.975573107894314, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9764329676859473, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9796054370603279, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9803064606691662, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.988613655913229, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9895089635821549, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9904793425414123, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9942634738736639, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9947079178068394, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9942815385538301, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9949886383194673, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971010128991973, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975084263623055, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984432058041229, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9985136829567828, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9986907554404368, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995833334641678, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.23.self_attn": [ { "accuracy": 0.9798769171496755, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9807503854267692, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9827229846759062, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.986826954556531, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9895349880435357, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9896893653479454, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9935373395656872, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937466446334791, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943690642371381, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946864276487184, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994831661301616, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951743366025192, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956616454659716, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959956857093953, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975903191615975, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997974656745914, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983342797342246, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988768358537192, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995394547846732, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.23.block_sparse_moe": [ { "accuracy": 0.9753504969963902, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9761808822421651, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.979235195456759, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9799193171784282, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9883603949687982, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9892888772791546, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9902420912281071, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.994146412508072, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9946014875374538, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9941464634367118, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9948742391639634, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970322872916433, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974525927529229, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9984091296883937, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9984769545997935, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.99864896008512, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995691633881315, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.24.self_attn": [ { "accuracy": 0.9790969731678304, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9804222328392299, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9820739571495276, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9861409616117415, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.98957328263082, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.989899068016951, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9936510917338494, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9940927498728821, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9945178547825076, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948780185812595, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949570807113656, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995336554883244, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956767684412434, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960149680693192, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975984285610091, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980103959617401, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984007130444393, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988835814445721, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995569582656332, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.24.block_sparse_moe": [ { "accuracy": 0.9742228875034734, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9750688592658231, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9781355845104707, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9788105278030822, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9877531308141586, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9887454489147977, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9896898430674091, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9938593869661226, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9943418888938859, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9938295959053856, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9946042662976604, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9968656562023649, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973144753725807, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983289385563694, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983912969106122, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985607268938207, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995525296759106, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.25.self_attn": [ { "accuracy": 0.98099081747626, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9817622033388991, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9838280793965647, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9874992032878493, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9903007528107417, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9905715994291792, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9941303681974348, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943161055715265, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948082705224423, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950428571135393, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951915550658381, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955785086280421, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959688143265483, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963206702377647, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977696699972608, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981516236194262, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984654732979834, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989649568543475, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999561028015083, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.25.block_sparse_moe": [ { "accuracy": 0.9733302876549331, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9741973826465639, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9773194224230553, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9780059529370383, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9873316286523875, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9883476869007083, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9893149337486217, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9936374685187873, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9941356736112779, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9936229594934144, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9944153563609641, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967628560016716, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972234077102161, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982721690168163, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9983387788402309, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9985135983657337, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999534300868522, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.26.self_attn": [ { "accuracy": 0.9784405961828796, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9793903255148938, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9820190580855859, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9859449636485231, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9891177436434909, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9892558007452049, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9933816138900032, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9936175944504181, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9942227071028595, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946293254799553, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9947451579531557, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951125883227704, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955429956684575, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959459726723205, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975013991107086, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979517672603068, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998288072584393, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988761540597281, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995163108841026, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.26.block_sparse_moe": [ { "accuracy": 0.9717906635058554, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9726969018382462, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9760030236977496, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.976729590249689, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9866106538837286, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9876709032411638, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9886974498237434, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9932753365290793, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9938041874035997, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9932646582540321, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9940916263179755, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9965841578393194, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970627396649338, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981793411210865, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9982489217562895, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.998438683396671, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995125697322166, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.27.self_attn": [ { "accuracy": 0.9797050898502532, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9804531226522828, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9828884171714124, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9861608400303674, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9895554479750756, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9897415374748801, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9936469515638524, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937466042557437, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943142527910439, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946731556035382, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948616053075775, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952651881980464, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956823155210403, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960427635876266, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975814776139726, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979822297678574, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983079452343334, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988951370147621, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995103818640162, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.27.block_sparse_moe": [ { "accuracy": 0.9706257865519116, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9715661484454023, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.974981070153023, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9757451846411354, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9860711688371865, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9871699143152096, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9882592284365704, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9929856191095161, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9935401247716263, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9929959587997904, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9938563465532896, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964486292845226, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969445500878225, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981015832714835, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981799070649829, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983800772338567, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994868158736561, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.28.self_attn": [ { "accuracy": 0.976489458380169, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9772998076912603, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9803919770607823, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9846157698499921, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9876388452064834, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9880606457836142, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9924051170482447, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9928141636406317, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9934299960609918, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.993805440755463, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994137505571799, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9944873530613748, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950368594329216, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9954562114894783, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972627701875019, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977030697811738, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99808138458818, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987157331871506, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994557059789689, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.28.block_sparse_moe": [ { "accuracy": 0.9691505957590906, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9701549014762828, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9739109905142533, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9747816170320699, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9853328996288934, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9865110155783201, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.987751967445212, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9925731059016758, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9931894980959202, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9926421056182957, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9935463654784191, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962676393456364, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9967915852995295, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979956168803925, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980923322972989, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983198924945962, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994601108835348, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.29.self_attn": [ { "accuracy": 0.9749339761114434, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9760917477113636, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9794238427359807, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9843229170349476, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9874217961424667, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9877170296630969, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9922336122560266, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.992665669562197, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9931880078204957, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9935176701799623, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9936562510525906, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994158292164732, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9948357565107903, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951782102917174, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971283517998496, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975988338732062, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979657386919778, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998669449549389, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999438259392168, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.29.block_sparse_moe": [ { "accuracy": 0.9694152064621449, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.970368828526453, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9740395034712396, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9748963689627616, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9854351485050038, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9865831424188065, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9877930632057158, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9926190533871321, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9932147536524817, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9926773058250546, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9935803120392129, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.99628898893532, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9968044096420176, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980140084302739, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9981026414193606, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9983236333171494, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994621564385996, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.30.self_attn": [ { "accuracy": 0.9805115162345924, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9814347636542822, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9839439843781292, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9875220536058279, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9900907625346199, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9901994759412972, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937516629793927, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.993994664756189, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943964090688449, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946693900457927, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.995010635521459, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953622357846287, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958235326247584, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961491373465642, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976778576199553, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980573274814034, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983312618753285, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989472494466769, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995198953915142, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.30.block_sparse_moe": [ { "accuracy": 0.968262855061575, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9692602786970766, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9729743691063241, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9738465969107652, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9848243843911118, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9860284217682324, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9872499930819398, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9923436906384794, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9929619731234485, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9923707767585782, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9933111892008272, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9961332288987347, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966750440408328, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979324292477318, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9980219546215314, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9982425626673687, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994407023860779, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.31.self_attn": [ { "accuracy": 0.9786951473276866, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9796270698701081, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9825742298148965, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9863282307051122, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9891506279818714, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9893815253469113, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9929777718835363, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9932808350943225, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9938918961163022, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941727979794929, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946621866455596, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9950924198461795, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9955249549085764, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995881934331632, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974659467606168, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979257406973231, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981076329652416, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.998877246131949, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994782126695531, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.31.block_sparse_moe": [ { "accuracy": 0.9674497370265032, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9684669107786918, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9721267716468949, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9729698488586828, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9844378570682908, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9856698079230753, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9868496623821557, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9921578530369228, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9927901490959093, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9921768948337749, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9931372861549455, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9960347347353634, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9965902746189386, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978848994207127, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9979723607692378, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981858791832469, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994317899936026, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.32.self_attn": [ { "accuracy": 0.979842871623604, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9807214215888005, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9831519328596953, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9871252403888655, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9898244608438721, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9899984076374063, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937286801685236, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939771198030365, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994411898531804, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946846057366776, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949265566262367, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952806952458463, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9957470087285497, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996091001801879, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976330948687208, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980279432799618, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998348478365714, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.99888411387311, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995342109667897, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.32.block_sparse_moe": [ { "accuracy": 0.9667622332314127, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9677903139473576, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9714293935777325, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9722772110743743, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9840556418915328, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9853166400356904, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9864851561080861, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9919745350060495, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9926190693421584, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.991985615318347, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9929690537238983, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959421527915095, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9965069894249993, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997839427638897, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9979236995445957, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981327775197937, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994202713439192, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.33.self_attn": [ { "accuracy": 0.9798874289875752, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9807773644202634, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9829530561048734, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9863884688698148, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9897775309601504, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9899936921796516, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9935308017424846, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9937311543617398, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941351286282665, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944845782102723, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949321222221968, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952679372294561, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99570956624015, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960568444497001, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975694832208223, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979902276444543, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982464507767489, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988620159933098, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994828114546468, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.33.block_sparse_moe": [ { "accuracy": 0.966130263123073, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9671781949306789, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9708612099485961, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9717120934081706, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9837428370098534, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9850456545521554, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9862087568464247, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9918223293007988, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9924879802734052, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9918198008825513, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9928331201969597, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.995851842845839, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964395816940325, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977952710193533, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978772471720156, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980826896713361, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994107742677443, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.34.self_attn": [ { "accuracy": 0.9791836305276344, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9799873712717703, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9824896610194915, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9869225981008065, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9894957589965901, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.989758602858178, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.993531221928271, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.993770142723071, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9942769053955808, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9945683586901348, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948652548666456, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995192027822333, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956331796282413, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959666757525778, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9975748836540764, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9979685207179404, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982958413433822, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9988682777467673, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995137613027749, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.34.block_sparse_moe": [ { "accuracy": 0.9667309934371396, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9677466651130664, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.971260682189543, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9720752434805036, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9839991447515786, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9852700615791898, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9863802217143146, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9919588274175399, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9926090343904338, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9919542812399174, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9929456563029242, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959257871990925, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9964944491542778, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978337657184487, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9979143476934711, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9981090400374102, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994175978415449, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.35.self_attn": [ { "accuracy": 0.9853694343841389, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9874433684652966, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9886061000196558, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9907108820778759, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9926237318125603, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9927184415343953, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995633388195481, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9958986939102608, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9962222190990456, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9963750699044842, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.996285287332476, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965530317399258, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971044391124068, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973733680457563, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984214751993453, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998664821476651, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988392818915216, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993025997215514, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996673404461226, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.35.block_sparse_moe": [ { "accuracy": 0.9658274209420932, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9668688042775581, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9704307724574679, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9712494208820557, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9835522188069789, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9848675775880876, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9859771210651257, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9917351663308708, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9924039580173006, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9917305931694022, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9927482267253493, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.995810786629782, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9963967675943614, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977723140771942, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978553352228022, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980510808833826, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994023784391558, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.36.self_attn": [ { "accuracy": 0.9823778276086638, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9832938879140114, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9849649991497005, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9882833458189118, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9907594150618503, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910063136516041, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943781467399707, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946881448359866, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950881827573635, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953133964690527, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953543070801779, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956570659445501, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961897667779244, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965370673269621, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978893350589236, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982524652671265, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985197459065699, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990263974698457, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995776661778922, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.36.block_sparse_moe": [ { "accuracy": 0.9647647743358424, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9659463445606985, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9694698239235502, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9702796725262153, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9832778478923597, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9845703222151649, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9856842106061154, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9915890294176183, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9922524958996004, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9916000796089831, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9926335228435499, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957433158139649, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9963394486909046, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977365070402524, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978210804003634, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980189016389073, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999389856241913, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.37.self_attn": [ { "accuracy": 0.9789652496968445, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9798429867154673, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9818607884410181, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9864480226839843, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9892903034036097, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9895174462152155, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9931502620150384, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9934374446686554, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939082156327602, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9941992280073464, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994635377410113, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9949895501087763, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953136451566886, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9956897005046669, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973784442378306, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997824435220345, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998147134969354, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987760327548083, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994732475821174, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.37.block_sparse_moe": [ { "accuracy": 0.9649798860283274, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9660969500086809, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9696090725299559, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9704259859496042, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.983118540215257, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9844965486248073, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.985602559203184, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9915250905201232, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9922120854454605, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9915217927068865, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9925644647232011, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957037166398215, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9963057584582681, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977162887685393, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.997799981458995, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979937201347447, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993856221661094, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.38.self_attn": [ { "accuracy": 0.9860520237125456, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9866770698051703, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9885957060430787, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910671866842007, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9928852069485736, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9931340642199901, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9952072251808682, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.995564028277601, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9959087526474736, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.996093557216227, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9965071058709567, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967395278355596, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970018459532998, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972636678476671, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983147064677292, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986208971700092, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987049695153377, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992154816955965, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996227102145847, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.38.block_sparse_moe": [ { "accuracy": 0.9655046137539964, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9665625484562234, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9699559369075456, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9707443194561883, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9832957591371316, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9846434162586535, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9856891318616506, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9915943058256648, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9922743211990517, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9915949124107627, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9926367675591456, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957417787150725, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.996339990420414, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977300826239547, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9978201854667675, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9980029170646479, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993890862434024, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.39.self_attn": [ { "accuracy": 0.9829557906640204, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.983982078682043, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9861695071878401, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9894982041349929, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9914888010948504, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9917374087595626, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9945683832108778, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948546783630982, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9952708725253806, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954907837642455, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9958076209540626, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.99609578446191, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965002288931588, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967665917816033, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998065501335077, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983706731030619, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985967364168333, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999064316890374, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995792671244615, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.39.block_sparse_moe": [ { "accuracy": 0.9650062243209073, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.966058173952134, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.96946698902665, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9702625340340953, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9830339076977811, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9844003733688671, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9854541215555448, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9914675383142343, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9921612863330856, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9914619728040538, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9925156552222019, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956747506328515, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962800402672177, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997701923375165, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977869276802889, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979686952245078, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993829614341896, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.40.self_attn": [ { "accuracy": 0.9848339793419367, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9856405308911282, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9873972269216258, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.990218985972828, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9923317602247392, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9925852669893127, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994819521070703, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951322278816646, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9955181284611555, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9957610141167319, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9961986845530766, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996501023311315, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966981770519755, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970140327809771, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981562802751892, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984799707277776, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986039845971391, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991156610041415, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995928994413636, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.40.block_sparse_moe": [ { "accuracy": 0.9645925844578367, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9656579726816792, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9690470375414741, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9698387638812787, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9827834447649749, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9841877512731835, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9852386617444848, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9913475734279736, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9920533278858975, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9913335316814482, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9924129761853501, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956065266711736, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962283169458571, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976609320759675, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977486531409484, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979272450038575, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993688761498657, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.41.self_attn": [ { "accuracy": 0.9822028848391614, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9833439939018143, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9855254516495686, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9889167576145969, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9912365704382721, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9914759858266303, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9944043484957594, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946844109187001, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950603035058042, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953159168117532, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9956744063621092, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959527324689063, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963352412573601, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966285386809001, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997939788233979, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982865027299053, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985144156349921, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990068418433351, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995618554316216, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.41.block_sparse_moe": [ { "accuracy": 0.9646893391679776, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9657483316565815, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9691186097303504, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9699064706893343, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9828266820527221, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9842199800830138, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9852615016953725, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9913672948697287, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9920721769234851, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9913612100246706, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9924267968162894, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956235117769163, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9962382818094307, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976764461331952, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977596245252675, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979376167297902, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993766479065495, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.42.self_attn": [ { "accuracy": 0.9821311084945735, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9833897074782535, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.985178665993245, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9883486653776153, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910746649103729, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9913983513405057, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9942004028952828, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946133302353126, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950322282966226, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9952813905271652, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9955963737233297, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959058178417188, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962000771376648, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965168359003177, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9978782345945212, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982399086706879, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984620458470952, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989895986938407, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995653134859535, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.42.block_sparse_moe": [ { "accuracy": 0.9641206437035611, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9651968274265528, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9685553837273466, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9693441373345099, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9825301430629272, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.983943480410074, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9849839955323229, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9912098369953272, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9919293505678836, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9912100840899113, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9922933312574107, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9955488563466229, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9961709827062135, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997635141780943, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9977228172968975, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9979005072412914, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.999365815777932, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.43.self_attn": [ { "accuracy": 0.9879073807969689, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9885907841258144, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9902438268702674, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9920982637344614, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9936681024877256, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9938558764559658, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958925277950537, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9962897842602902, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9965247617127668, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9966830483294631, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9967561195678029, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970213289152047, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974922094628901, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977060434053113, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985852363078218, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9988274416506389, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9989047439640241, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993487536358818, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996730521149737, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.43.block_sparse_moe": [ { "accuracy": 0.9632287789136171, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9643234675259966, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9677064715835609, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9684990822876755, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9820704789538133, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9835334079045999, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9845662750793915, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9909791435164056, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.991722026521242, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9909724995346838, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9920941709638819, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.995428383583203, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9960740961894197, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975713381029078, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9976604340205851, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978356569895128, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993508214186142, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.44.self_attn": [ { "accuracy": 0.9830589907635984, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9841191279456803, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9863190991117766, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9888935603544509, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9914563273226744, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9916584807889242, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.994325489761602, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9944812818675449, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9949557402808415, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950546847264233, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.995726789776726, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960661946893915, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9964571784887659, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9967650310136378, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980034391873663, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983349022529039, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984564068114483, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989940719158192, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995350782456195, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.44.block_sparse_moe": [ { "accuracy": 0.9634563568979502, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9645841752031916, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9679246641891568, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9687071160266274, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9822048696836359, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9836439423655209, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9846561498155719, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9910491299080221, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9917767796861497, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9910456988421318, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9921512818209043, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9954675043638992, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9961016775525519, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975929751281479, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.99768152937146, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9978524442382255, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993592670913418, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.45.self_attn": [ { "accuracy": 0.9823894630627412, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9835072796987859, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9851935626938939, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9882873283012917, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9909023194279718, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9912815163843334, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943913617661517, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9947335669329684, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951290861997557, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954350001031631, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9954488530730534, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9958109033882225, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962976212539759, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966287820869567, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997915997378234, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9982827054520481, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985092986607924, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990388293732203, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995703645721445, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.45.block_sparse_moe": [ { "accuracy": 0.9622780714968318, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9634148891230947, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9668290156282877, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9676261009942544, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.981600133604125, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.98309292698181, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9841253514107513, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.99074223684147, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9914983664884379, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9907411605266756, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9918857809205196, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9953114088918817, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959696909584301, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975085270247961, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9976007726876751, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977748095560329, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993341612877741, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.46.self_attn": [ { "accuracy": 0.9839242150222784, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9846926395732322, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9866546381166891, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9896843166237599, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9914379203446994, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.991773106427373, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9947635074219617, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9951438021625539, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9955329801029477, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9957460765236694, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9957651180019113, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960738547930592, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9966220302556298, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9969041912424329, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998096740822994, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984180330127281, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9986131092790809, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991102238142814, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995894606798078, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.46.block_sparse_moe": [ { "accuracy": 0.9621072152727529, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.963257788435409, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9666236102777093, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9674132224662524, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9814857511634105, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9829890684371716, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9840098793226245, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9906766974171134, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9914400889538229, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9906821648443216, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.991833754398517, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.995279949905939, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9959425284943887, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9974912487546375, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9975840395720872, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9977554873146705, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993292875342855, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.47.self_attn": [ { "accuracy": 0.9841478014560906, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9847966264443178, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9873647859418079, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9897689843599341, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9917520006481362, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9919368753111676, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951010555102441, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9953950571361929, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9957482737587079, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9959581344094324, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9958737442073854, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961613550120475, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968351347396445, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971180272619485, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.998226061659424, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9985512987630883, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9987114966742842, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992513460885292, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9996218739696289, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.47.block_sparse_moe": [ { "accuracy": 0.9601467788023383, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9613649269663974, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9648702008355605, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9656971008761933, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9805347910641056, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9821272284576767, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9831744712826452, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9902060133110928, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9910082158044373, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9902082217220021, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9914211556805592, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9950448764414576, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9957400801787643, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973679367072978, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9974645873605225, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976404029202011, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993011751426629, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.48.self_attn": [ { "accuracy": 0.9811355963251308, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9820919365278984, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9844681338260048, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.988366174389069, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9902544662503427, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9903334776968941, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9939914146977428, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943507035066815, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.994797758598763, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9950847049047681, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.995102551974062, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995430869099341, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9960501455138192, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963778064243103, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977954075255088, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981718633054314, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9984372508846909, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989847150190097, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995462119266832, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.48.block_sparse_moe": [ { "accuracy": 0.9581291243237884, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9594383824028467, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9630028952501322, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.963851183162708, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.979565043198435, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9812269770099145, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9823007515484565, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9896909429348613, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9905258493841087, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9897178917829144, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9909849617826311, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9947953100169176, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9955213048961014, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9972237415954863, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9973366975012284, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9975181800975023, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992562768548898, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.49.self_attn": [ { "accuracy": 0.979685212926645, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9809823417055764, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9836373120840443, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9868493178663286, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9893530617517076, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9894539430296343, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9936224650591612, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9939159168616721, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9943609313402129, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9946680094090927, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9948676550809882, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9951842031196544, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9959388995714682, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9963049097966991, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9977817470330352, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9981530941593902, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9983389189491343, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989413911274536, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9995234099256586, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.49.block_sparse_moe": [ { "accuracy": 0.9546757910988832, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.956299631611297, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9599952817355332, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9609087549738193, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9780043306221303, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.979807079720654, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9809702298064765, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9888810871757174, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9897862654669505, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9889218802073676, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9902876324363445, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9943914045702273, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9951767608964521, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970072300332647, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9971301256752524, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973308545488276, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991936325876802, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.50.self_attn": [ { "accuracy": 0.9743504725200566, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9760822851799036, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.979320595923223, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9835319952843221, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9868650243764645, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9872634405956456, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9925657537226614, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9927159803674409, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9933118147187328, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.993637580816683, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9935372267163506, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9940569043453586, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.995005234095611, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9953798485650239, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9972095251401984, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976800639108804, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9980296448461319, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9987410880612994, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9994461591229896, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.50.block_sparse_moe": [ { "accuracy": 0.9531344101813278, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9543402818472762, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9584258186974024, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9593263728250014, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9778074427066665, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9795588940676105, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9807536389286581, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9887258886408649, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9896665379454038, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9888360049624584, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9902040426325249, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9943337610702178, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9951259522747836, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969332895829881, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9970897163852657, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973019067945547, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.99913383457246, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.51.self_attn": [ { "accuracy": 0.9690277881332134, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9707019190843168, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9751432959578539, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9805351641323221, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9848088651503387, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9849864249362758, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9911719044719479, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914124963167859, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9920053304124036, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9923573967412507, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9923695225121552, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9929161616355965, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937379662878811, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9942517387067997, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965404828422164, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9970944551934832, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976357115726722, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984314987821946, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993456112753943, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.51.block_sparse_moe": [ { "accuracy": 0.9543641139998248, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9558297939794629, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9595772315208849, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9605014074201647, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9777513987531787, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9795997476107196, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9807966442680672, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9887157048490879, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9896309715940764, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9888128104962801, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9901883165155979, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9943177810698551, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9951269448685803, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9969184548520532, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9971060447676695, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9973176447123191, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991789426176662, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.52.self_attn": [ { "accuracy": 0.9678433800214216, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9694013197563196, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9727757996634433, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9775656452892643, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9848284283652902, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9850820276984259, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9910950328615543, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914315676218585, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9920785305531401, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9923806800507009, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9924766703165675, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9928867673305305, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9936909135921221, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9942482126214983, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965455105258642, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.997111235034505, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976936190413606, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984034980910732, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993480808380378, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.52.block_sparse_moe": [ { "accuracy": 0.9528234430441731, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9543118672935587, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9582035844459346, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9592085354226201, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.976814666713931, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9787363520774403, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.980104562042183, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9881089480201665, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9890723443639121, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.988277228058953, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9897391493922394, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9940497594101256, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9948930971471495, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.996753217952669, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9969656849341271, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.997219013571004, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9991033621450984, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.53.self_attn": [ { "accuracy": 0.9648685968039852, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9670851572759842, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9723176306211635, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9785976861848643, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9826800419801944, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9831384486567817, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9899063593658962, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9903565936623827, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9910480896452147, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914846087707893, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914391332207934, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9920787648788016, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9931297045152047, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937123575762502, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9961781229603252, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.996820716089324, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9973441652316404, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9982733651137862, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992651432801626, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.53.block_sparse_moe": [ { "accuracy": 0.9498194690401617, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9514897981364476, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9557364612425628, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9568548253492305, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9755391867733315, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9774993469840602, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9790279137852945, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9874933011290666, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9885395408951139, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9876176581816062, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.989155884145906, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9937210747678029, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9946023646962682, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966180790889714, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9967817759484445, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9970695350110825, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990601138285312, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.54.self_attn": [ { "accuracy": 0.9646947517206794, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9673028555943778, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9724297310764852, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.978179406178625, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.982994856263854, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9833442073217348, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9901041792656639, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9904212245570594, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9913260857574642, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9917226401647847, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914950923258928, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9921485003280012, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9931828521535193, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9937018690459234, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9962390784517323, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9968208847996337, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9974478892491836, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.99829693805843, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9992832698810258, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.54.block_sparse_moe": [ { "accuracy": 0.95103951033793, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9525281677120611, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9563693778687402, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9573947489261627, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9760888161039666, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9780498665120256, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9794690563392482, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9876140042600271, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9886573346723851, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9878470511910947, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.98936269939632, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9938112084484217, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9946970759498838, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9966173962702191, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9968343854416162, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971114737410588, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9990470455530541, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.55.self_attn": [ { "accuracy": 0.9685467731599745, "total_bits": 187058176, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9705001499111715, "total_bits": 192563200, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9746255009974304, "total_bits": 197738240, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9793367833389264, "total_bits": 234305024, "q_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9843059681650055, "total_bits": 276964352, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9844907334466514, "total_bits": 277163264, "q_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64, "3": 64 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.991243941096687, "total_bits": 355869696, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9914467399939895, "total_bits": 356068608, "q_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9913694611260373, "total_bits": 359029248, "q_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 64 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9920250053125385, "total_bits": 364154880, "q_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9924824913767608, "total_bits": 365243648, "q_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9930283447825595, "total_bits": 368204288, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9938025830043969, "total_bits": 373442560, "q_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 64 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 64, "4": 64 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9943635697648125, "total_bits": 378966016, "q_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "5": 32 }, "bits": [ 5 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9965892571490258, "total_bits": 458761216, "q_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9971222042130601, "total_bits": 467046400, "q_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 } }, { "accuracy": 0.9976775591555787, "total_bits": 532030464, "q_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9984275445427844, "total_bits": 552898560, "q_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 32 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "6": 32 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9993178526500216, "total_bits": 708191232, "q_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "k_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "v_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "o_proj": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.layers.55.block_sparse_moe": [ { "accuracy": 0.9591578813759905, "total_bits": 5403947904, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9604559534866559, "total_bits": 5605274496, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "6": 32, "3": 64, "2": 64 }, "bits": [ 6, 3, 2 ], "bits_prop": [ 0.05, 0.2, 0.75 ], "scale_bits": 4 } }, { "accuracy": 0.9633979222884304, "total_bits": 6255026176, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "3": 32 }, "bits": [ 5, 3 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9643714849494005, "total_bits": 7019438080, "w1": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "3": 64, "2": 64 }, "bits": [ 3, 2 ], "bits_prop": [ 0.3, 0.7 ], "scale_bits": 4 }, "w2": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.980747265692212, "total_bits": 7906094592, "w1": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 128, "3": 128 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128, "3": 128 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.982378172982288, "total_bits": 8128905216, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32, "3": 32 }, "bits": [ 8, 4, 3 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.983617174831268, "total_bits": 8739823104, "w1": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32, "3": 32 }, "bits": [ 4, 3 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9897530631309277, "total_bits": 9989670400, "w1": { "group_size": { "4": 128 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 128 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9907836571923996, "total_bits": 10136928256, "w1": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "4": 32 }, "bits": [ 4 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "4": 32 }, "bits": [ 8, 4 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9902422087953279, "total_bits": 10281119232, "w1": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 128, "4": 128 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 128, "4": 128 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9914621277712286, "total_bits": 10503929856, "w1": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "5": 32, "4": 32 }, "bits": [ 5, 4 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "5": 32, "4": 32 }, "bits": [ 8, 5, 4 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9949502851568947, "total_bits": 12656143872, "w1": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128, "5": 128 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128, "5": 128 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9956636329083458, "total_bits": 12878954496, "w1": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 32, "5": 32 }, "bits": [ 6, 5 ], "bits_prop": [ 0.25, 0.75 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 32, "5": 32 }, "bits": [ 8, 6, 5 ], "bits_prop": [ 0.05, 0.1, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9971022841361302, "total_bits": 14664166912, "w1": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "6": 128 }, "bits": [ 6 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 32, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.05, 0.95 ], "scale_bits": 4 } }, { "accuracy": 0.9973727284722325, "total_bits": 15165640704, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.15, 0.85 ], "scale_bits": 4 } }, { "accuracy": 0.9976540774866742, "total_bits": 16524595200, "w1": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128, "6": 128 }, "bits": [ 8, 6 ], "bits_prop": [ 0.1, 0.9 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } }, { "accuracy": 0.9989871072834112, "total_bits": 19410276352, "w1": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w3": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 }, "w2": { "group_size": { "8": 128 }, "bits": [ 8 ], "bits_prop": [ 1 ], "scale_bits": 4 } } ], "model.norm.norm": null, "lm_head.linear": null }, "last_module_idx": 114 }