diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -10,7 +10,7 @@
   "log_history": [
     {
       "epoch": 0.0010471204188481676,
-      "grad_norm": 1.5250242434220578,
+      "grad_norm": 1.5250827885007505,
       "learning_rate": 1.7421602787456448e-07,
       "logits/chosen": -11.625,
       "logits/rejected": -12.0,
@@ -25,4851 +25,4851 @@
     },
     {
       "epoch": 0.010471204188481676,
-      "grad_norm": 1.389122809127924,
+      "grad_norm": 1.3886428611401354,
       "learning_rate": 1.7421602787456445e-06,
       "logits/chosen": -11.375,
       "logits/rejected": -11.125,
       "logps/chosen": -442.0,
       "logps/rejected": -516.0,
-      "loss": 0.6912,
-      "rewards/accuracies": 0.1805555522441864,
-      "rewards/chosen": 0.000766754150390625,
-      "rewards/margins": 0.0012969970703125,
-      "rewards/rejected": -0.00052642822265625,
+      "loss": 0.6919,
+      "rewards/accuracies": 0.1388888955116272,
+      "rewards/chosen": 0.00063323974609375,
+      "rewards/margins": -0.000640869140625,
+      "rewards/rejected": 0.00127410888671875,
       "step": 10
     },
     {
       "epoch": 0.020942408376963352,
-      "grad_norm": 1.409590573779433,
+      "grad_norm": 1.4135446538732952,
       "learning_rate": 3.484320557491289e-06,
       "logits/chosen": -11.125,
       "logits/rejected": -11.125,
       "logps/chosen": -406.0,
       "logps/rejected": -418.0,
-      "loss": 0.6913,
-      "rewards/accuracies": 0.23125000298023224,
-      "rewards/chosen": 9.34600830078125e-05,
-      "rewards/margins": 0.00122833251953125,
-      "rewards/rejected": -0.00113677978515625,
+      "loss": 0.6917,
+      "rewards/accuracies": 0.22499999403953552,
+      "rewards/chosen": -0.000347137451171875,
+      "rewards/margins": -0.000194549560546875,
+      "rewards/rejected": -0.00015735626220703125,
       "step": 20
     },
     {
       "epoch": 0.031413612565445025,
-      "grad_norm": 1.4705607714647213,
+      "grad_norm": 1.4738194128623547,
       "learning_rate": 5.226480836236934e-06,
       "logits/chosen": -11.375,
       "logits/rejected": -11.375,
-      "logps/chosen": -422.0,
+      "logps/chosen": -424.0,
       "logps/rejected": -458.0,
-      "loss": 0.691,
-      "rewards/accuracies": 0.3687500059604645,
-      "rewards/chosen": -0.0028533935546875,
-      "rewards/margins": 0.002838134765625,
-      "rewards/rejected": -0.005706787109375,
+      "loss": 0.6915,
+      "rewards/accuracies": 0.3499999940395355,
+      "rewards/chosen": -0.005462646484375,
+      "rewards/margins": 0.0006103515625,
+      "rewards/rejected": -0.006072998046875,
       "step": 30
     },
     {
       "epoch": 0.041884816753926704,
-      "grad_norm": 1.6460039306987837,
+      "grad_norm": 1.6442204689692357,
       "learning_rate": 6.968641114982578e-06,
       "logits/chosen": -11.875,
       "logits/rejected": -11.6875,
       "logps/chosen": -436.0,
       "logps/rejected": -456.0,
-      "loss": 0.6875,
-      "rewards/accuracies": 0.48124998807907104,
-      "rewards/chosen": -0.05712890625,
-      "rewards/margins": 0.01202392578125,
-      "rewards/rejected": -0.06884765625,
+      "loss": 0.6883,
+      "rewards/accuracies": 0.46875,
+      "rewards/chosen": -0.05615234375,
+      "rewards/margins": 0.00970458984375,
+      "rewards/rejected": -0.06591796875,
       "step": 40
     },
     {
       "epoch": 0.05235602094240838,
-      "grad_norm": 1.8680864585896733,
+      "grad_norm": 1.8721117495707351,
       "learning_rate": 8.710801393728225e-06,
       "logits/chosen": -12.5625,
       "logits/rejected": -12.5625,
       "logps/chosen": -448.0,
       "logps/rejected": -512.0,
-      "loss": 0.681,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": -0.189453125,
-      "rewards/margins": 0.03466796875,
-      "rewards/rejected": -0.224609375,
+      "loss": 0.6803,
+      "rewards/accuracies": 0.518750011920929,
+      "rewards/chosen": -0.1826171875,
+      "rewards/margins": 0.03271484375,
+      "rewards/rejected": -0.21484375,
       "step": 50
     },
     {
       "epoch": 0.06282722513089005,
-      "grad_norm": 1.8887715548630017,
+      "grad_norm": 1.891671200712439,
       "learning_rate": 1.0452961672473868e-05,
-      "logits/chosen": -11.9375,
+      "logits/chosen": -12.0,
       "logits/rejected": -11.875,
       "logps/chosen": -368.0,
-      "logps/rejected": -434.0,
-      "loss": 0.6676,
+      "logps/rejected": -432.0,
+      "loss": 0.667,
       "rewards/accuracies": 0.606249988079071,
-      "rewards/chosen": -0.1865234375,
-      "rewards/margins": 0.07470703125,
-      "rewards/rejected": -0.26171875,
+      "rewards/chosen": -0.1845703125,
+      "rewards/margins": 0.07568359375,
+      "rewards/rejected": -0.259765625,
       "step": 60
     },
     {
       "epoch": 0.07329842931937172,
-      "grad_norm": 1.8422098676961731,
+      "grad_norm": 1.8378201787564292,
       "learning_rate": 1.2195121951219513e-05,
-      "logits/chosen": -11.6875,
-      "logits/rejected": -11.6875,
+      "logits/chosen": -11.75,
+      "logits/rejected": -11.75,
       "logps/chosen": -474.0,
       "logps/rejected": -544.0,
-      "loss": 0.6638,
+      "loss": 0.6636,
       "rewards/accuracies": 0.5687500238418579,
-      "rewards/chosen": -0.306640625,
-      "rewards/margins": 0.10595703125,
+      "rewards/chosen": -0.30859375,
+      "rewards/margins": 0.10546875,
       "rewards/rejected": -0.412109375,
       "step": 70
     },
     {
       "epoch": 0.08376963350785341,
-      "grad_norm": 1.8975797746738434,
+      "grad_norm": 1.8909882060624779,
       "learning_rate": 1.3937282229965156e-05,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -11.25,
-      "logps/chosen": -452.0,
-      "logps/rejected": -560.0,
-      "loss": 0.6384,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -0.54296875,
-      "rewards/margins": 0.2158203125,
-      "rewards/rejected": -0.7578125,
+      "logits/chosen": -11.375,
+      "logits/rejected": -11.3125,
+      "logps/chosen": -450.0,
+      "logps/rejected": -556.0,
+      "loss": 0.6378,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -0.5234375,
+      "rewards/margins": 0.2177734375,
+      "rewards/rejected": -0.7421875,
       "step": 80
     },
     {
       "epoch": 0.08376963350785341,
       "eval_logits/chosen": -10.125,
-      "eval_logits/rejected": -9.8125,
-      "eval_logps/chosen": -390.0,
-      "eval_logps/rejected": -368.0,
-      "eval_loss": 0.6898750066757202,
-      "eval_rewards/accuracies": 0.556640625,
-      "eval_rewards/chosen": -0.7109375,
-      "eval_rewards/margins": 0.08544921875,
-      "eval_rewards/rejected": -0.796875,
-      "eval_runtime": 48.2248,
-      "eval_samples_per_second": 41.472,
-      "eval_steps_per_second": 0.664,
+      "eval_logits/rejected": -9.875,
+      "eval_logps/chosen": -386.0,
+      "eval_logps/rejected": -366.0,
+      "eval_loss": 0.6867812275886536,
+      "eval_rewards/accuracies": 0.568359375,
+      "eval_rewards/chosen": -0.67578125,
+      "eval_rewards/margins": 0.091796875,
+      "eval_rewards/rejected": -0.765625,
+      "eval_runtime": 60.3411,
+      "eval_samples_per_second": 33.145,
+      "eval_steps_per_second": 0.53,
       "step": 80
     },
     {
       "epoch": 0.09424083769633508,
-      "grad_norm": 2.1829683859160904,
+      "grad_norm": 2.2191250224336923,
       "learning_rate": 1.56794425087108e-05,
       "logits/chosen": -10.4375,
       "logits/rejected": -10.4375,
-      "logps/chosen": -452.0,
-      "logps/rejected": -490.0,
-      "loss": 0.626,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": -0.81640625,
-      "rewards/margins": 0.16796875,
-      "rewards/rejected": -0.984375,
+      "logps/chosen": -454.0,
+      "logps/rejected": -492.0,
+      "loss": 0.6263,
+      "rewards/accuracies": 0.581250011920929,
+      "rewards/chosen": -0.83203125,
+      "rewards/margins": 0.1669921875,
+      "rewards/rejected": -0.99609375,
       "step": 90
     },
     {
       "epoch": 0.10471204188481675,
-      "grad_norm": 2.150149674051317,
+      "grad_norm": 2.0747518663361872,
       "learning_rate": 1.742160278745645e-05,
-      "logits/chosen": -11.5,
-      "logits/rejected": -11.6875,
+      "logits/chosen": -11.4375,
+      "logits/rejected": -11.5625,
       "logps/chosen": -592.0,
       "logps/rejected": -624.0,
-      "loss": 0.6542,
-      "rewards/accuracies": 0.581250011920929,
-      "rewards/chosen": -1.1953125,
-      "rewards/margins": 0.130859375,
-      "rewards/rejected": -1.328125,
+      "loss": 0.6543,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -1.1875,
+      "rewards/margins": 0.1298828125,
+      "rewards/rejected": -1.3203125,
       "step": 100
     },
     {
       "epoch": 0.11518324607329843,
-      "grad_norm": 1.8807721043038954,
+      "grad_norm": 1.9055511646839332,
       "learning_rate": 1.9163763066202093e-05,
-      "logits/chosen": -11.625,
-      "logits/rejected": -11.5625,
-      "logps/chosen": -512.0,
-      "logps/rejected": -572.0,
-      "loss": 0.6472,
-      "rewards/accuracies": 0.65625,
-      "rewards/chosen": -0.83203125,
-      "rewards/margins": 0.2041015625,
-      "rewards/rejected": -1.0390625,
+      "logits/chosen": -11.25,
+      "logits/rejected": -11.1875,
+      "logps/chosen": -508.0,
+      "logps/rejected": -568.0,
+      "loss": 0.6475,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.7890625,
+      "rewards/margins": 0.20703125,
+      "rewards/rejected": -0.99609375,
       "step": 110
     },
     {
       "epoch": 0.1256544502617801,
-      "grad_norm": 2.5043080843651477,
+      "grad_norm": 2.4619703878135337,
       "learning_rate": 2.0905923344947736e-05,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -11.25,
-      "logps/chosen": -498.0,
-      "logps/rejected": -576.0,
-      "loss": 0.6302,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -1.1953125,
-      "rewards/margins": 0.2578125,
-      "rewards/rejected": -1.453125,
+      "logits/chosen": -12.125,
+      "logits/rejected": -12.0,
+      "logps/chosen": -504.0,
+      "logps/rejected": -584.0,
+      "loss": 0.6309,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -1.2578125,
+      "rewards/margins": 0.259765625,
+      "rewards/rejected": -1.515625,
       "step": 120
     },
     {
       "epoch": 0.13612565445026178,
-      "grad_norm": 2.1173336892174524,
+      "grad_norm": 2.141339067230429,
       "learning_rate": 2.264808362369338e-05,
       "logits/chosen": -13.1875,
       "logits/rejected": -13.0625,
-      "logps/chosen": -528.0,
-      "logps/rejected": -560.0,
-      "loss": 0.6514,
+      "logps/chosen": -532.0,
+      "logps/rejected": -564.0,
+      "loss": 0.6493,
       "rewards/accuracies": 0.59375,
-      "rewards/chosen": -1.1875,
-      "rewards/margins": 0.1708984375,
-      "rewards/rejected": -1.359375,
+      "rewards/chosen": -1.21875,
+      "rewards/margins": 0.1787109375,
+      "rewards/rejected": -1.3984375,
       "step": 130
     },
     {
       "epoch": 0.14659685863874344,
-      "grad_norm": 2.494073608537398,
+      "grad_norm": 2.5246514217822797,
       "learning_rate": 2.4390243902439026e-05,
-      "logits/chosen": -13.125,
-      "logits/rejected": -13.0,
-      "logps/chosen": -548.0,
-      "logps/rejected": -684.0,
-      "loss": 0.616,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -0.94140625,
-      "rewards/margins": 0.3359375,
-      "rewards/rejected": -1.28125,
+      "logits/chosen": -13.5,
+      "logits/rejected": -13.3125,
+      "logps/chosen": -576.0,
+      "logps/rejected": -708.0,
+      "loss": 0.6144,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": -1.1875,
+      "rewards/margins": 0.341796875,
+      "rewards/rejected": -1.53125,
       "step": 140
     },
     {
       "epoch": 0.15706806282722513,
-      "grad_norm": 2.3573458900401065,
+      "grad_norm": 2.290292066383781,
       "learning_rate": 2.6132404181184672e-05,
-      "logits/chosen": -11.375,
-      "logits/rejected": -11.375,
-      "logps/chosen": -568.0,
+      "logits/chosen": -11.625,
+      "logits/rejected": -11.5625,
+      "logps/chosen": -572.0,
       "logps/rejected": -588.0,
-      "loss": 0.6389,
-      "rewards/accuracies": 0.643750011920929,
-      "rewards/chosen": -1.328125,
-      "rewards/margins": 0.1904296875,
-      "rewards/rejected": -1.5234375,
+      "loss": 0.6442,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -1.359375,
+      "rewards/margins": 0.1826171875,
+      "rewards/rejected": -1.5390625,
       "step": 150
     },
     {
       "epoch": 0.16753926701570682,
-      "grad_norm": 1.9398704546256205,
+      "grad_norm": 1.927538142019672,
       "learning_rate": 2.7874564459930312e-05,
-      "logits/chosen": -11.1875,
-      "logits/rejected": -10.875,
-      "logps/chosen": -536.0,
-      "logps/rejected": -584.0,
-      "loss": 0.6225,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": -1.34375,
-      "rewards/margins": 0.2080078125,
-      "rewards/rejected": -1.546875,
+      "logits/chosen": -10.75,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -500.0,
+      "logps/rejected": -548.0,
+      "loss": 0.6219,
+      "rewards/accuracies": 0.606249988079071,
+      "rewards/chosen": -0.9921875,
+      "rewards/margins": 0.1845703125,
+      "rewards/rejected": -1.1796875,
       "step": 160
     },
     {
       "epoch": 0.16753926701570682,
-      "eval_logits/chosen": -9.6875,
-      "eval_logits/rejected": -9.25,
-      "eval_logps/chosen": -466.0,
-      "eval_logps/rejected": -450.0,
-      "eval_loss": 0.7089999914169312,
-      "eval_rewards/accuracies": 0.55078125,
-      "eval_rewards/chosen": -1.484375,
-      "eval_rewards/margins": 0.1279296875,
-      "eval_rewards/rejected": -1.609375,
-      "eval_runtime": 46.5518,
-      "eval_samples_per_second": 42.963,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -10.5,
+      "eval_logits/rejected": -10.125,
+      "eval_logps/chosen": -410.0,
+      "eval_logps/rejected": -394.0,
+      "eval_loss": 0.694937527179718,
+      "eval_rewards/accuracies": 0.59765625,
+      "eval_rewards/chosen": -0.91015625,
+      "eval_rewards/margins": 0.140625,
+      "eval_rewards/rejected": -1.0546875,
+      "eval_runtime": 58.2686,
+      "eval_samples_per_second": 34.324,
+      "eval_steps_per_second": 0.549,
       "step": 160
     },
     {
       "epoch": 0.17801047120418848,
-      "grad_norm": 1.9994737406017982,
+      "grad_norm": 1.9703167548795155,
       "learning_rate": 2.9616724738675962e-05,
-      "logits/chosen": -9.9375,
-      "logits/rejected": -10.0,
-      "logps/chosen": -552.0,
-      "logps/rejected": -648.0,
-      "loss": 0.6085,
-      "rewards/accuracies": 0.581250011920929,
-      "rewards/chosen": -1.1953125,
-      "rewards/margins": 0.28125,
-      "rewards/rejected": -1.4765625,
+      "logits/chosen": -11.3125,
+      "logits/rejected": -11.3125,
+      "logps/chosen": -520.0,
+      "logps/rejected": -620.0,
+      "loss": 0.5929,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -0.89453125,
+      "rewards/margins": 0.3125,
+      "rewards/rejected": -1.203125,
       "step": 170
     },
     {
       "epoch": 0.18848167539267016,
-      "grad_norm": 2.1426934356554783,
+      "grad_norm": 2.4182468580601575,
       "learning_rate": 3.13588850174216e-05,
-      "logits/chosen": -7.96875,
-      "logits/rejected": -7.84375,
-      "logps/chosen": -616.0,
-      "logps/rejected": -676.0,
-      "loss": 0.615,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -1.6640625,
-      "rewards/margins": 0.365234375,
-      "rewards/rejected": -2.03125,
+      "logits/chosen": -11.75,
+      "logits/rejected": -11.75,
+      "logps/chosen": -628.0,
+      "logps/rejected": -696.0,
+      "loss": 0.6053,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -1.78125,
+      "rewards/margins": 0.44140625,
+      "rewards/rejected": -2.21875,
       "step": 180
     },
     {
       "epoch": 0.19895287958115182,
-      "grad_norm": 2.0716089888301203,
+      "grad_norm": 2.3100195311518936,
       "learning_rate": 3.310104529616725e-05,
-      "logits/chosen": -8.625,
-      "logits/rejected": -8.6875,
-      "logps/chosen": -548.0,
-      "logps/rejected": -612.0,
-      "loss": 0.6277,
-      "rewards/accuracies": 0.643750011920929,
-      "rewards/chosen": -1.4140625,
-      "rewards/margins": 0.263671875,
-      "rewards/rejected": -1.6796875,
+      "logits/chosen": -13.0,
+      "logits/rejected": -12.875,
+      "logps/chosen": -592.0,
+      "logps/rejected": -664.0,
+      "loss": 0.6219,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -1.890625,
+      "rewards/margins": 0.298828125,
+      "rewards/rejected": -2.1875,
       "step": 190
     },
     {
       "epoch": 0.2094240837696335,
-      "grad_norm": 2.2722032809767105,
+      "grad_norm": 2.3163984145832868,
       "learning_rate": 3.48432055749129e-05,
-      "logits/chosen": -8.375,
-      "logits/rejected": -8.5,
-      "logps/chosen": -588.0,
-      "logps/rejected": -680.0,
-      "loss": 0.6299,
+      "logits/chosen": -15.625,
+      "logits/rejected": -15.5625,
+      "logps/chosen": -584.0,
+      "logps/rejected": -668.0,
+      "loss": 0.6381,
       "rewards/accuracies": 0.6312500238418579,
-      "rewards/chosen": -1.2734375,
-      "rewards/margins": 0.326171875,
-      "rewards/rejected": -1.6015625,
+      "rewards/chosen": -1.2265625,
+      "rewards/margins": 0.28515625,
+      "rewards/rejected": -1.5078125,
       "step": 200
     },
     {
       "epoch": 0.2198952879581152,
-      "grad_norm": 2.2296408151400393,
+      "grad_norm": 2.2124811855136515,
       "learning_rate": 3.6585365853658535e-05,
-      "logits/chosen": -6.9375,
-      "logits/rejected": -7.03125,
+      "logits/chosen": -12.75,
+      "logits/rejected": -12.625,
       "logps/chosen": -494.0,
-      "logps/rejected": -616.0,
-      "loss": 0.6421,
-      "rewards/accuracies": 0.6187499761581421,
+      "logps/rejected": -612.0,
+      "loss": 0.6408,
+      "rewards/accuracies": 0.574999988079071,
       "rewards/chosen": -1.25,
-      "rewards/margins": 0.3828125,
-      "rewards/rejected": -1.6328125,
+      "rewards/margins": 0.359375,
+      "rewards/rejected": -1.609375,
       "step": 210
     },
     {
       "epoch": 0.23036649214659685,
-      "grad_norm": 2.130611899424163,
+      "grad_norm": 2.1866800062497505,
       "learning_rate": 3.8327526132404185e-05,
-      "logits/chosen": -11.5625,
-      "logits/rejected": -11.1875,
-      "logps/chosen": -556.0,
-      "logps/rejected": -604.0,
-      "loss": 0.6289,
-      "rewards/accuracies": 0.5687500238418579,
-      "rewards/chosen": -1.2421875,
-      "rewards/margins": 0.2001953125,
-      "rewards/rejected": -1.4453125,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -13.25,
+      "logps/chosen": -540.0,
+      "logps/rejected": -596.0,
+      "loss": 0.6138,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -1.0703125,
+      "rewards/margins": 0.279296875,
+      "rewards/rejected": -1.34375,
       "step": 220
     },
     {
       "epoch": 0.24083769633507854,
-      "grad_norm": 1.8520601361217366,
+      "grad_norm": 1.9395551683491863,
       "learning_rate": 4.006968641114983e-05,
-      "logits/chosen": -9.875,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -552.0,
-      "logps/rejected": -612.0,
-      "loss": 0.6373,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -1.6640625,
-      "rewards/margins": 0.26953125,
-      "rewards/rejected": -1.9375,
+      "logits/chosen": -10.9375,
+      "logits/rejected": -10.9375,
+      "logps/chosen": -560.0,
+      "logps/rejected": -624.0,
+      "loss": 0.6312,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -1.75,
+      "rewards/margins": 0.306640625,
+      "rewards/rejected": -2.0625,
       "step": 230
     },
     {
       "epoch": 0.2513089005235602,
-      "grad_norm": 2.1016756286711233,
+      "grad_norm": 2.2975204261649655,
       "learning_rate": 4.181184668989547e-05,
-      "logits/chosen": -9.5625,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -560.0,
-      "logps/rejected": -652.0,
-      "loss": 0.6067,
+      "logits/chosen": -10.6875,
+      "logits/rejected": -10.5,
+      "logps/chosen": -588.0,
+      "logps/rejected": -680.0,
+      "loss": 0.6151,
       "rewards/accuracies": 0.643750011920929,
-      "rewards/chosen": -1.765625,
-      "rewards/margins": 0.34765625,
-      "rewards/rejected": -2.109375,
+      "rewards/chosen": -2.046875,
+      "rewards/margins": 0.359375,
+      "rewards/rejected": -2.40625,
       "step": 240
     },
     {
       "epoch": 0.2513089005235602,
-      "eval_logits/chosen": -7.53125,
-      "eval_logits/rejected": -6.75,
-      "eval_logps/chosen": -540.0,
-      "eval_logps/rejected": -528.0,
-      "eval_loss": 0.7371875047683716,
-      "eval_rewards/accuracies": 0.5625,
-      "eval_rewards/chosen": -2.1875,
-      "eval_rewards/margins": 0.1865234375,
-      "eval_rewards/rejected": -2.375,
-      "eval_runtime": 46.5548,
-      "eval_samples_per_second": 42.96,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -11.25,
+      "eval_logits/rejected": -10.9375,
+      "eval_logps/chosen": -564.0,
+      "eval_logps/rejected": -552.0,
+      "eval_loss": 0.7636562585830688,
+      "eval_rewards/accuracies": 0.556640625,
+      "eval_rewards/chosen": -2.453125,
+      "eval_rewards/margins": 0.203125,
+      "eval_rewards/rejected": -2.65625,
+      "eval_runtime": 58.29,
+      "eval_samples_per_second": 34.311,
+      "eval_steps_per_second": 0.549,
       "step": 240
     },
     {
       "epoch": 0.2617801047120419,
-      "grad_norm": 2.982206902816664,
+      "grad_norm": 2.6358608824930045,
       "learning_rate": 4.3554006968641115e-05,
-      "logits/chosen": -10.1875,
-      "logits/rejected": -9.875,
-      "logps/chosen": -688.0,
-      "logps/rejected": -748.0,
-      "loss": 0.6188,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": -2.234375,
-      "rewards/margins": 0.400390625,
-      "rewards/rejected": -2.640625,
+      "logits/chosen": -11.875,
+      "logits/rejected": -11.625,
+      "logps/chosen": -656.0,
+      "logps/rejected": -720.0,
+      "loss": 0.609,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -1.9375,
+      "rewards/margins": 0.421875,
+      "rewards/rejected": -2.359375,
       "step": 250
     },
     {
       "epoch": 0.27225130890052357,
-      "grad_norm": 2.961667282448676,
+      "grad_norm": 2.5678912060328893,
       "learning_rate": 4.529616724738676e-05,
-      "logits/chosen": -9.125,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -608.0,
-      "logps/rejected": -728.0,
-      "loss": 0.6,
-      "rewards/accuracies": 0.6312500238418579,
-      "rewards/chosen": -2.203125,
-      "rewards/margins": 0.50390625,
-      "rewards/rejected": -2.703125,
+      "logits/chosen": -12.125,
+      "logits/rejected": -11.8125,
+      "logps/chosen": -552.0,
+      "logps/rejected": -668.0,
+      "loss": 0.6134,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -1.6484375,
+      "rewards/margins": 0.453125,
+      "rewards/rejected": -2.09375,
       "step": 260
     },
     {
       "epoch": 0.28272251308900526,
-      "grad_norm": 1.8985917047516574,
+      "grad_norm": 2.0641919876942465,
       "learning_rate": 4.703832752613241e-05,
-      "logits/chosen": -8.6875,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -584.0,
-      "logps/rejected": -648.0,
-      "loss": 0.6257,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": -2.015625,
-      "rewards/margins": 0.3046875,
-      "rewards/rejected": -2.328125,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -14.3125,
+      "logps/chosen": -536.0,
+      "logps/rejected": -596.0,
+      "loss": 0.6103,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -1.5078125,
+      "rewards/margins": 0.3125,
+      "rewards/rejected": -1.828125,
       "step": 270
     },
     {
       "epoch": 0.2931937172774869,
-      "grad_norm": 1.9036327280293572,
+      "grad_norm": 2.0701160239572,
       "learning_rate": 4.878048780487805e-05,
-      "logits/chosen": -7.15625,
-      "logits/rejected": -7.25,
-      "logps/chosen": -576.0,
-      "logps/rejected": -680.0,
-      "loss": 0.6455,
-      "rewards/accuracies": 0.643750011920929,
-      "rewards/chosen": -2.078125,
-      "rewards/margins": 0.37109375,
-      "rewards/rejected": -2.453125,
+      "logits/chosen": -13.625,
+      "logits/rejected": -13.4375,
+      "logps/chosen": -544.0,
+      "logps/rejected": -660.0,
+      "loss": 0.6134,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.765625,
+      "rewards/margins": 0.48828125,
+      "rewards/rejected": -2.25,
       "step": 280
     },
     {
       "epoch": 0.3036649214659686,
-      "grad_norm": 2.3919446297292826,
+      "grad_norm": 2.5860382671689806,
       "learning_rate": 4.999983293481418e-05,
-      "logits/chosen": -9.125,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -616.0,
-      "logps/rejected": -672.0,
-      "loss": 0.6709,
-      "rewards/accuracies": 0.581250011920929,
-      "rewards/chosen": -2.21875,
-      "rewards/margins": 0.2734375,
-      "rewards/rejected": -2.484375,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -12.8125,
+      "logps/chosen": -660.0,
+      "logps/rejected": -720.0,
+      "loss": 0.6646,
+      "rewards/accuracies": 0.59375,
+      "rewards/chosen": -2.65625,
+      "rewards/margins": 0.328125,
+      "rewards/rejected": -2.96875,
       "step": 290
     },
     {
       "epoch": 0.31413612565445026,
-      "grad_norm": 2.3065774840245457,
+      "grad_norm": 2.055355083372985,
       "learning_rate": 4.9996862949181595e-05,
-      "logits/chosen": -9.25,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -632.0,
-      "logps/rejected": -668.0,
-      "loss": 0.6427,
-      "rewards/accuracies": 0.65625,
-      "rewards/chosen": -2.1875,
-      "rewards/margins": 0.279296875,
-      "rewards/rejected": -2.46875,
+      "logits/chosen": -14.25,
+      "logits/rejected": -14.125,
+      "logps/chosen": -656.0,
+      "logps/rejected": -688.0,
+      "loss": 0.6451,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -2.4375,
+      "rewards/margins": 0.2265625,
+      "rewards/rejected": -2.65625,
       "step": 300
     },
     {
       "epoch": 0.32460732984293195,
-      "grad_norm": 2.53412202251708,
+      "grad_norm": 5.027724244827355,
       "learning_rate": 4.9990180911526896e-05,
-      "logits/chosen": -11.1875,
-      "logits/rejected": -11.0,
-      "logps/chosen": -508.0,
-      "logps/rejected": -576.0,
-      "loss": 0.6411,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -1.359375,
-      "rewards/margins": 0.296875,
-      "rewards/rejected": -1.65625,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -14.1875,
+      "logps/chosen": -560.0,
+      "logps/rejected": -628.0,
+      "loss": 0.6399,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -1.8671875,
+      "rewards/margins": 0.3125,
+      "rewards/rejected": -2.1875,
       "step": 310
     },
     {
       "epoch": 0.33507853403141363,
-      "grad_norm": 2.727666591615435,
+      "grad_norm": 2.360366917235795,
       "learning_rate": 4.997978781413745e-05,
-      "logits/chosen": -9.25,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -640.0,
-      "logps/rejected": -680.0,
-      "loss": 0.63,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -1.984375,
-      "rewards/margins": 0.32421875,
-      "rewards/rejected": -2.3125,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -14.5,
+      "logps/chosen": -696.0,
+      "logps/rejected": -732.0,
+      "loss": 0.6607,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -2.546875,
+      "rewards/margins": 0.283203125,
+      "rewards/rejected": -2.828125,
       "step": 320
     },
     {
       "epoch": 0.33507853403141363,
-      "eval_logits/chosen": -5.21875,
-      "eval_logits/rejected": -4.3125,
-      "eval_logps/chosen": -616.0,
-      "eval_logps/rejected": -604.0,
-      "eval_loss": 0.7780937552452087,
-      "eval_rewards/accuracies": 0.572265625,
-      "eval_rewards/chosen": -2.96875,
-      "eval_rewards/margins": 0.205078125,
-      "eval_rewards/rejected": -3.171875,
-      "eval_runtime": 46.5338,
-      "eval_samples_per_second": 42.98,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -14.4375,
+      "eval_logits/rejected": -14.25,
+      "eval_logps/chosen": -592.0,
+      "eval_logps/rejected": -584.0,
+      "eval_loss": 0.7306562662124634,
+      "eval_rewards/accuracies": 0.57421875,
+      "eval_rewards/chosen": -2.734375,
+      "eval_rewards/margins": 0.19921875,
+      "eval_rewards/rejected": -2.9375,
+      "eval_runtime": 58.3061,
+      "eval_samples_per_second": 34.302,
+      "eval_steps_per_second": 0.549,
       "step": 320
     },
     {
       "epoch": 0.34554973821989526,
-      "grad_norm": 2.581501456104908,
+      "grad_norm": 2.703381453509918,
       "learning_rate": 4.996568520039565e-05,
-      "logits/chosen": -6.3125,
-      "logits/rejected": -6.4375,
-      "logps/chosen": -728.0,
-      "logps/rejected": -844.0,
-      "loss": 0.6133,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -3.078125,
-      "rewards/margins": 0.443359375,
-      "rewards/rejected": -3.53125,
+      "logits/chosen": -15.3125,
+      "logits/rejected": -15.25,
+      "logps/chosen": -656.0,
+      "logps/rejected": -784.0,
+      "loss": 0.6112,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -2.375,
+      "rewards/margins": 0.546875,
+      "rewards/rejected": -2.921875,
       "step": 330
     },
     {
       "epoch": 0.35602094240837695,
-      "grad_norm": 2.843422741311636,
+      "grad_norm": 2.0368300453949892,
       "learning_rate": 4.994787516454973e-05,
-      "logits/chosen": -6.90625,
-      "logits/rejected": -7.03125,
-      "logps/chosen": -740.0,
-      "logps/rejected": -816.0,
-      "loss": 0.6251,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 0.31640625,
-      "rewards/rejected": -3.6875,
+      "logits/chosen": -14.875,
+      "logits/rejected": -14.6875,
+      "logps/chosen": -612.0,
+      "logps/rejected": -696.0,
+      "loss": 0.6126,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -2.09375,
+      "rewards/margins": 0.376953125,
+      "rewards/rejected": -2.46875,
       "step": 340
     },
     {
       "epoch": 0.36649214659685864,
-      "grad_norm": 2.315433708975923,
+      "grad_norm": 2.796384314910774,
       "learning_rate": 4.992636035140273e-05,
-      "logits/chosen": -7.46875,
-      "logits/rejected": -7.625,
-      "logps/chosen": -576.0,
-      "logps/rejected": -644.0,
-      "loss": 0.6526,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": -2.09375,
-      "rewards/margins": 0.21875,
-      "rewards/rejected": -2.3125,
+      "logits/chosen": -13.9375,
+      "logits/rejected": -13.875,
+      "logps/chosen": -544.0,
+      "logps/rejected": -620.0,
+      "loss": 0.6441,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -1.7421875,
+      "rewards/margins": 0.310546875,
+      "rewards/rejected": -2.0625,
       "step": 350
     },
     {
       "epoch": 0.3769633507853403,
-      "grad_norm": 2.842855272207156,
+      "grad_norm": 2.604091948627254,
       "learning_rate": 4.99011439559198e-05,
-      "logits/chosen": -8.1875,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -644.0,
-      "logps/rejected": -720.0,
-      "loss": 0.6259,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -2.328125,
-      "rewards/margins": 0.322265625,
-      "rewards/rejected": -2.65625,
+      "logits/chosen": -13.25,
+      "logits/rejected": -13.0,
+      "logps/chosen": -600.0,
+      "logps/rejected": -684.0,
+      "loss": 0.607,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -1.890625,
+      "rewards/margins": 0.375,
+      "rewards/rejected": -2.265625,
       "step": 360
     },
     {
       "epoch": 0.387434554973822,
-      "grad_norm": 2.3109388778838014,
+      "grad_norm": 2.2547076545133584,
       "learning_rate": 4.987222972275369e-05,
-      "logits/chosen": -9.0625,
-      "logits/rejected": -9.125,
-      "logps/chosen": -632.0,
-      "logps/rejected": -712.0,
-      "loss": 0.6241,
-      "rewards/accuracies": 0.6187499761581421,
-      "rewards/chosen": -2.203125,
-      "rewards/margins": 0.388671875,
-      "rewards/rejected": -2.59375,
+      "logits/chosen": -10.9375,
+      "logits/rejected": -10.75,
+      "logps/chosen": -704.0,
+      "logps/rejected": -792.0,
+      "loss": 0.6271,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 0.4296875,
+      "rewards/rejected": -3.359375,
       "step": 370
     },
     {
       "epoch": 0.39790575916230364,
-      "grad_norm": 2.1333836779428665,
+      "grad_norm": 2.1254508862931716,
       "learning_rate": 4.983962194568872e-05,
-      "logits/chosen": -9.75,
-      "logits/rejected": -9.625,
-      "logps/chosen": -632.0,
-      "logps/rejected": -732.0,
-      "loss": 0.6433,
-      "rewards/accuracies": 0.5687500238418579,
-      "rewards/chosen": -2.125,
-      "rewards/margins": 0.2734375,
-      "rewards/rejected": -2.40625,
+      "logits/chosen": -12.5,
+      "logits/rejected": -12.1875,
+      "logps/chosen": -700.0,
+      "logps/rejected": -800.0,
+      "loss": 0.6379,
+      "rewards/accuracies": 0.59375,
+      "rewards/chosen": -2.796875,
+      "rewards/margins": 0.310546875,
+      "rewards/rejected": -3.109375,
       "step": 380
     },
     {
       "epoch": 0.4083769633507853,
-      "grad_norm": 2.1636616183783115,
+      "grad_norm": 1.9574259898250574,
       "learning_rate": 4.9803325467003084e-05,
-      "logits/chosen": -9.5,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -700.0,
-      "logps/rejected": -780.0,
-      "loss": 0.6208,
-      "rewards/accuracies": 0.6187499761581421,
-      "rewards/chosen": -2.6875,
-      "rewards/margins": 0.43359375,
-      "rewards/rejected": -3.125,
+      "logits/chosen": -14.0625,
+      "logits/rejected": -13.75,
+      "logps/chosen": -704.0,
+      "logps/rejected": -784.0,
+      "loss": 0.6195,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -2.734375,
+      "rewards/margins": 0.4140625,
+      "rewards/rejected": -3.15625,
       "step": 390
     },
     {
       "epoch": 0.418848167539267,
-      "grad_norm": 2.2472665316693554,
+      "grad_norm": 1.9709626806319096,
       "learning_rate": 4.9763345676749816e-05,
-      "logits/chosen": -7.25,
-      "logits/rejected": -7.21875,
-      "logps/chosen": -700.0,
-      "logps/rejected": -776.0,
-      "loss": 0.6294,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -3.015625,
-      "rewards/margins": 0.353515625,
-      "rewards/rejected": -3.375,
+      "logits/chosen": -13.625,
+      "logits/rejected": -13.4375,
+      "logps/chosen": -652.0,
+      "logps/rejected": -724.0,
+      "loss": 0.6304,
+      "rewards/accuracies": 0.6187499761581421,
+      "rewards/chosen": -2.546875,
+      "rewards/margins": 0.33203125,
+      "rewards/rejected": -2.875,
       "step": 400
     },
     {
       "epoch": 0.418848167539267,
-      "eval_logits/chosen": -5.75,
-      "eval_logits/rejected": -4.65625,
-      "eval_logps/chosen": -576.0,
-      "eval_logps/rejected": -576.0,
-      "eval_loss": 0.7329531311988831,
-      "eval_rewards/accuracies": 0.59765625,
-      "eval_rewards/chosen": -2.59375,
-      "eval_rewards/margins": 0.27734375,
-      "eval_rewards/rejected": -2.859375,
-      "eval_runtime": 46.5435,
-      "eval_samples_per_second": 42.971,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -13.0,
+      "eval_logits/rejected": -12.5,
+      "eval_logps/chosen": -592.0,
+      "eval_logps/rejected": -588.0,
+      "eval_loss": 0.7129218578338623,
+      "eval_rewards/accuracies": 0.58984375,
+      "eval_rewards/chosen": -2.734375,
+      "eval_rewards/margins": 0.271484375,
+      "eval_rewards/rejected": -3.015625,
+      "eval_runtime": 58.3006,
+      "eval_samples_per_second": 34.305,
+      "eval_steps_per_second": 0.549,
       "step": 400
     },
     {
       "epoch": 0.4293193717277487,
-      "grad_norm": 1.9276746222864911,
+      "grad_norm": 2.067465712936359,
       "learning_rate": 4.9719688511956374e-05,
-      "logits/chosen": -7.09375,
-      "logits/rejected": -7.09375,
-      "logps/chosen": -640.0,
-      "logps/rejected": -752.0,
-      "loss": 0.6344,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -2.109375,
-      "rewards/margins": 0.375,
-      "rewards/rejected": -2.484375,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -13.0625,
+      "logps/chosen": -692.0,
+      "logps/rejected": -812.0,
+      "loss": 0.6424,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -2.640625,
+      "rewards/margins": 0.421875,
+      "rewards/rejected": -3.0625,
       "step": 410
     },
     {
       "epoch": 0.4397905759162304,
-      "grad_norm": 2.4543488152289012,
+      "grad_norm": 2.3409559806248668,
       "learning_rate": 4.967236045574294e-05,
-      "logits/chosen": -6.875,
-      "logits/rejected": -6.53125,
-      "logps/chosen": -692.0,
-      "logps/rejected": -764.0,
-      "loss": 0.6045,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": -2.78125,
-      "rewards/margins": 0.369140625,
-      "rewards/rejected": -3.140625,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -13.375,
+      "logps/chosen": -704.0,
+      "logps/rejected": -788.0,
+      "loss": 0.6016,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 0.486328125,
+      "rewards/rejected": -3.375,
       "step": 420
     },
     {
       "epoch": 0.450261780104712,
-      "grad_norm": 2.544800507924643,
+      "grad_norm": 2.1801568985894804,
       "learning_rate": 4.96213685363597e-05,
-      "logits/chosen": -7.125,
-      "logits/rejected": -7.1875,
-      "logps/chosen": -680.0,
-      "logps/rejected": -792.0,
-      "loss": 0.638,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -2.71875,
-      "rewards/margins": 0.49609375,
-      "rewards/rejected": -3.21875,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -14.0,
+      "logps/chosen": -648.0,
+      "logps/rejected": -760.0,
+      "loss": 0.6308,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -2.421875,
+      "rewards/margins": 0.490234375,
+      "rewards/rejected": -2.90625,
       "step": 430
     },
     {
       "epoch": 0.4607329842931937,
-      "grad_norm": 5.34313153624845,
+      "grad_norm": 3.4232504629304206,
       "learning_rate": 4.956672032614315e-05,
-      "logits/chosen": -7.84375,
-      "logits/rejected": -7.90625,
-      "logps/chosen": -648.0,
-      "logps/rejected": -724.0,
-      "loss": 0.6481,
-      "rewards/accuracies": 0.643750011920929,
-      "rewards/chosen": -2.265625,
-      "rewards/margins": 0.408203125,
-      "rewards/rejected": -2.671875,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -11.9375,
+      "logps/chosen": -652.0,
+      "logps/rejected": -732.0,
+      "loss": 0.621,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -2.296875,
+      "rewards/margins": 0.466796875,
+      "rewards/rejected": -2.75,
       "step": 440
     },
     {
       "epoch": 0.4712041884816754,
-      "grad_norm": 2.4388854172377843,
+      "grad_norm": 2.287505154218043,
       "learning_rate": 4.950842394039157e-05,
-      "logits/chosen": -9.3125,
-      "logits/rejected": -8.875,
-      "logps/chosen": -612.0,
-      "logps/rejected": -704.0,
-      "loss": 0.601,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": -2.265625,
-      "rewards/margins": 0.34765625,
-      "rewards/rejected": -2.609375,
+      "logits/chosen": -12.1875,
+      "logits/rejected": -11.75,
+      "logps/chosen": -576.0,
+      "logps/rejected": -664.0,
+      "loss": 0.6231,
+      "rewards/accuracies": 0.606249988079071,
+      "rewards/chosen": -1.9140625,
+      "rewards/margins": 0.28125,
+      "rewards/rejected": -2.203125,
       "step": 450
     },
     {
       "epoch": 0.4816753926701571,
-      "grad_norm": 2.5611060806590435,
+      "grad_norm": 2.4037034213395656,
       "learning_rate": 4.944648803615994e-05,
-      "logits/chosen": -8.875,
-      "logits/rejected": -8.5625,
-      "logps/chosen": -788.0,
-      "logps/rejected": -888.0,
-      "loss": 0.6067,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -3.71875,
-      "rewards/margins": 0.62109375,
-      "rewards/rejected": -4.34375,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -10.875,
+      "logps/chosen": -636.0,
+      "logps/rejected": -720.0,
+      "loss": 0.6162,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -2.1875,
+      "rewards/margins": 0.462890625,
+      "rewards/rejected": -2.65625,
       "step": 460
     },
     {
       "epoch": 0.49214659685863876,
-      "grad_norm": 2.3216279901898402,
+      "grad_norm": 2.613001518083897,
       "learning_rate": 4.938092181097433e-05,
-      "logits/chosen": -10.5625,
-      "logits/rejected": -10.5,
-      "logps/chosen": -792.0,
-      "logps/rejected": -904.0,
-      "loss": 0.5923,
-      "rewards/accuracies": 0.6187499761581421,
-      "rewards/chosen": -3.90625,
-      "rewards/margins": 0.375,
-      "rewards/rejected": -4.28125,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -10.875,
+      "logps/chosen": -652.0,
+      "logps/rejected": -780.0,
+      "loss": 0.5662,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -2.515625,
+      "rewards/margins": 0.55078125,
+      "rewards/rejected": -3.0625,
       "step": 470
     },
     {
       "epoch": 0.5026178010471204,
-      "grad_norm": 3.4840013957576925,
+      "grad_norm": 3.6359277123903864,
       "learning_rate": 4.931173500146607e-05,
-      "logits/chosen": -10.3125,
-      "logits/rejected": -10.0,
-      "logps/chosen": -724.0,
-      "logps/rejected": -848.0,
-      "loss": 0.6199,
+      "logits/chosen": -9.8125,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -640.0,
+      "logps/rejected": -768.0,
+      "loss": 0.623,
       "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 0.6015625,
-      "rewards/rejected": -4.03125,
+      "rewards/chosen": -2.609375,
+      "rewards/margins": 0.609375,
+      "rewards/rejected": -3.21875,
       "step": 480
     },
     {
       "epoch": 0.5026178010471204,
-      "eval_logits/chosen": -8.9375,
-      "eval_logits/rejected": -7.96875,
-      "eval_logps/chosen": -652.0,
-      "eval_logps/rejected": -656.0,
-      "eval_loss": 0.7531874775886536,
-      "eval_rewards/accuracies": 0.58203125,
-      "eval_rewards/chosen": -3.34375,
-      "eval_rewards/margins": 0.330078125,
-      "eval_rewards/rejected": -3.671875,
-      "eval_runtime": 46.5223,
-      "eval_samples_per_second": 42.99,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -9.0,
+      "eval_logits/rejected": -8.0625,
+      "eval_logps/chosen": -572.0,
+      "eval_logps/rejected": -584.0,
+      "eval_loss": 0.7717812657356262,
+      "eval_rewards/accuracies": 0.5859375,
+      "eval_rewards/chosen": -2.546875,
+      "eval_rewards/margins": 0.388671875,
+      "eval_rewards/rejected": -2.9375,
+      "eval_runtime": 58.299,
+      "eval_samples_per_second": 34.306,
+      "eval_steps_per_second": 0.549,
       "step": 480
     },
     {
       "epoch": 0.5130890052356021,
-      "grad_norm": 2.0336288673222653,
+      "grad_norm": 2.1236223892132693,
       "learning_rate": 4.923893788192582e-05,
-      "logits/chosen": -9.8125,
+      "logits/chosen": -9.9375,
       "logits/rejected": -9.5625,
-      "logps/chosen": -728.0,
-      "logps/rejected": -836.0,
-      "loss": 0.6331,
-      "rewards/accuracies": 0.581250011920929,
-      "rewards/chosen": -3.046875,
-      "rewards/margins": 0.4453125,
-      "rewards/rejected": -3.484375,
+      "logps/chosen": -644.0,
+      "logps/rejected": -760.0,
+      "loss": 0.626,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -2.1875,
+      "rewards/margins": 0.54296875,
+      "rewards/rejected": -2.734375,
       "step": 490
     },
     {
       "epoch": 0.5235602094240838,
-      "grad_norm": 1.988892356530348,
+      "grad_norm": 1.9093011532468762,
       "learning_rate": 4.9162541262777906e-05,
-      "logits/chosen": -11.4375,
-      "logits/rejected": -11.3125,
-      "logps/chosen": -656.0,
+      "logits/chosen": -9.75,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -652.0,
       "logps/rejected": -796.0,
-      "loss": 0.6263,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": -2.359375,
-      "rewards/margins": 0.51953125,
-      "rewards/rejected": -2.875,
+      "loss": 0.6209,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": -2.328125,
+      "rewards/margins": 0.55859375,
+      "rewards/rejected": -2.890625,
       "step": 500
     },
     {
       "epoch": 0.5340314136125655,
-      "grad_norm": 2.430050889250553,
+      "grad_norm": 2.803597849589253,
       "learning_rate": 4.90825564889749e-05,
-      "logits/chosen": -11.25,
-      "logits/rejected": -11.125,
-      "logps/chosen": -696.0,
-      "logps/rejected": -704.0,
-      "loss": 0.6478,
-      "rewards/accuracies": 0.6312500238418579,
-      "rewards/chosen": -2.484375,
-      "rewards/margins": 0.255859375,
-      "rewards/rejected": -2.75,
+      "logits/chosen": -8.875,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -776.0,
+      "logps/rejected": -784.0,
+      "loss": 0.656,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 0.28125,
+      "rewards/rejected": -3.578125,
       "step": 510
     },
     {
       "epoch": 0.5445026178010471,
-      "grad_norm": 2.401337651368821,
+      "grad_norm": 2.2586977510601702,
       "learning_rate": 4.899899543831288e-05,
-      "logits/chosen": -10.4375,
-      "logits/rejected": -10.25,
-      "logps/chosen": -716.0,
-      "logps/rejected": -816.0,
-      "loss": 0.5863,
-      "rewards/accuracies": 0.643750011920929,
-      "rewards/chosen": -2.65625,
-      "rewards/margins": 0.57421875,
-      "rewards/rejected": -3.234375,
+      "logits/chosen": -9.5625,
+      "logits/rejected": -9.5,
+      "logps/chosen": -752.0,
+      "logps/rejected": -852.0,
+      "loss": 0.5896,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 0.5859375,
+      "rewards/rejected": -3.59375,
       "step": 520
     },
     {
       "epoch": 0.5549738219895288,
-      "grad_norm": 2.4303123802001974,
+      "grad_norm": 2.469500009146035,
       "learning_rate": 4.891187051966765e-05,
-      "logits/chosen": -9.75,
-      "logits/rejected": -9.625,
-      "logps/chosen": -736.0,
-      "logps/rejected": -780.0,
-      "loss": 0.638,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": -3.046875,
-      "rewards/margins": 0.28515625,
-      "rewards/rejected": -3.328125,
+      "logits/chosen": -9.9375,
+      "logits/rejected": -9.875,
+      "logps/chosen": -724.0,
+      "logps/rejected": -764.0,
+      "loss": 0.6514,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 0.265625,
+      "rewards/rejected": -3.171875,
       "step": 530
     },
     {
       "epoch": 0.5654450261780105,
-      "grad_norm": 2.38792274118369,
+      "grad_norm": 2.0316082276482117,
       "learning_rate": 4.882119467115191e-05,
-      "logits/chosen": -11.5625,
-      "logits/rejected": -11.4375,
-      "logps/chosen": -704.0,
-      "logps/rejected": -796.0,
-      "loss": 0.6003,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": -2.921875,
-      "rewards/margins": 0.478515625,
-      "rewards/rejected": -3.40625,
+      "logits/chosen": -9.8125,
+      "logits/rejected": -9.75,
+      "logps/chosen": -648.0,
+      "logps/rejected": -732.0,
+      "loss": 0.6066,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": -2.359375,
+      "rewards/margins": 0.384765625,
+      "rewards/rejected": -2.75,
       "step": 540
     },
     {
       "epoch": 0.5759162303664922,
-      "grad_norm": 3.113086745652925,
+      "grad_norm": 3.0506796646793557,
       "learning_rate": 4.872698135819402e-05,
-      "logits/chosen": -11.75,
-      "logits/rejected": -11.3125,
-      "logps/chosen": -712.0,
-      "logps/rejected": -776.0,
-      "loss": 0.6182,
-      "rewards/accuracies": 0.643750011920929,
-      "rewards/chosen": -3.015625,
-      "rewards/margins": 0.416015625,
-      "rewards/rejected": -3.4375,
+      "logits/chosen": -11.25,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -652.0,
+      "logps/rejected": -720.0,
+      "loss": 0.6087,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -2.421875,
+      "rewards/margins": 0.4296875,
+      "rewards/rejected": -2.84375,
       "step": 550
     },
     {
       "epoch": 0.5863874345549738,
-      "grad_norm": 2.2064790385674518,
+      "grad_norm": 2.398111255020076,
       "learning_rate": 4.862924457153831e-05,
-      "logits/chosen": -9.6875,
-      "logits/rejected": -9.375,
-      "logps/chosen": -680.0,
-      "logps/rejected": -796.0,
-      "loss": 0.5935,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 0.60546875,
-      "rewards/rejected": -3.9375,
+      "logits/chosen": -12.1875,
+      "logits/rejected": -11.75,
+      "logps/chosen": -636.0,
+      "logps/rejected": -744.0,
+      "loss": 0.6091,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 0.55859375,
+      "rewards/rejected": -3.453125,
       "step": 560
     },
     {
       "epoch": 0.5863874345549738,
-      "eval_logits/chosen": -8.875,
-      "eval_logits/rejected": -7.90625,
-      "eval_logps/chosen": -724.0,
-      "eval_logps/rejected": -728.0,
-      "eval_loss": 0.7359374761581421,
-      "eval_rewards/accuracies": 0.5859375,
-      "eval_rewards/chosen": -4.0625,
-      "eval_rewards/margins": 0.326171875,
-      "eval_rewards/rejected": -4.40625,
-      "eval_runtime": 46.5481,
-      "eval_samples_per_second": 42.966,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -12.75,
+      "eval_logits/rejected": -12.0,
+      "eval_logps/chosen": -652.0,
+      "eval_logps/rejected": -656.0,
+      "eval_loss": 0.7543437480926514,
+      "eval_rewards/accuracies": 0.595703125,
+      "eval_rewards/chosen": -3.328125,
+      "eval_rewards/margins": 0.33203125,
+      "eval_rewards/rejected": -3.65625,
+      "eval_runtime": 58.3077,
+      "eval_samples_per_second": 34.301,
+      "eval_steps_per_second": 0.549,
       "step": 560
     },
     {
       "epoch": 0.5968586387434555,
-      "grad_norm": 2.224105492790409,
+      "grad_norm": 2.2952575031575315,
       "learning_rate": 4.8527998825167484e-05,
-      "logits/chosen": -8.375,
-      "logits/rejected": -8.375,
-      "logps/chosen": -728.0,
-      "logps/rejected": -836.0,
-      "loss": 0.621,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -3.46875,
-      "rewards/margins": 0.50390625,
-      "rewards/rejected": -3.984375,
+      "logits/chosen": -11.25,
+      "logits/rejected": -11.125,
+      "logps/chosen": -680.0,
+      "logps/rejected": -788.0,
+      "loss": 0.6237,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -2.984375,
+      "rewards/margins": 0.494140625,
+      "rewards/rejected": -3.484375,
       "step": 570
     },
     {
       "epoch": 0.6073298429319371,
-      "grad_norm": 1.9086663902018601,
+      "grad_norm": 2.3134695661524,
       "learning_rate": 4.842325915414728e-05,
-      "logits/chosen": -11.0625,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -624.0,
-      "logps/rejected": -732.0,
-      "loss": 0.6105,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -2.484375,
-      "rewards/margins": 0.51953125,
-      "rewards/rejected": -3.0,
+      "logits/chosen": -13.875,
+      "logits/rejected": -13.3125,
+      "logps/chosen": -636.0,
+      "logps/rejected": -748.0,
+      "loss": 0.6161,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -2.578125,
+      "rewards/margins": 0.56640625,
+      "rewards/rejected": -3.140625,
       "step": 580
     },
     {
       "epoch": 0.6178010471204188,
-      "grad_norm": 2.844222159597312,
+      "grad_norm": 2.5700564209735943,
       "learning_rate": 4.831504111239373e-05,
-      "logits/chosen": -13.0,
-      "logits/rejected": -12.6875,
-      "logps/chosen": -716.0,
-      "logps/rejected": -832.0,
-      "loss": 0.6152,
-      "rewards/accuracies": 0.606249988079071,
-      "rewards/chosen": -2.90625,
-      "rewards/margins": 0.412109375,
-      "rewards/rejected": -3.3125,
+      "logits/chosen": -12.875,
+      "logits/rejected": -12.5625,
+      "logps/chosen": -760.0,
+      "logps/rejected": -868.0,
+      "loss": 0.6197,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 0.361328125,
+      "rewards/rejected": -3.703125,
       "step": 590
     },
     {
       "epoch": 0.6282722513089005,
-      "grad_norm": 2.732838087673599,
+      "grad_norm": 2.823426405766663,
       "learning_rate": 4.8203360770363425e-05,
-      "logits/chosen": -14.4375,
-      "logits/rejected": -14.1875,
-      "logps/chosen": -756.0,
-      "logps/rejected": -800.0,
-      "loss": 0.6024,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 0.390625,
-      "rewards/rejected": -3.734375,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -14.4375,
+      "logps/chosen": -736.0,
+      "logps/rejected": -780.0,
+      "loss": 0.5912,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 0.353515625,
+      "rewards/rejected": -3.515625,
       "step": 600
     },
     {
       "epoch": 0.6387434554973822,
-      "grad_norm": 1.948112836388519,
+      "grad_norm": 2.2032054884051915,
       "learning_rate": 4.808823471266702e-05,
-      "logits/chosen": -14.875,
-      "logits/rejected": -14.8125,
-      "logps/chosen": -628.0,
-      "logps/rejected": -692.0,
-      "loss": 0.6165,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -2.21875,
-      "rewards/margins": 0.44140625,
-      "rewards/rejected": -2.671875,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -14.25,
+      "logps/chosen": -664.0,
+      "logps/rejected": -732.0,
+      "loss": 0.616,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -2.578125,
+      "rewards/margins": 0.48828125,
+      "rewards/rejected": -3.078125,
       "step": 610
     },
     {
       "epoch": 0.6492146596858639,
-      "grad_norm": 2.5647741224819116,
+      "grad_norm": 2.278949905844074,
       "learning_rate": 4.796968003560639e-05,
-      "logits/chosen": -12.9375,
-      "logits/rejected": -12.5,
-      "logps/chosen": -676.0,
-      "logps/rejected": -800.0,
-      "loss": 0.6224,
-      "rewards/accuracies": 0.6312500238418579,
-      "rewards/chosen": -2.59375,
-      "rewards/margins": 0.482421875,
-      "rewards/rejected": -3.078125,
+      "logits/chosen": -12.75,
+      "logits/rejected": -12.375,
+      "logps/chosen": -716.0,
+      "logps/rejected": -848.0,
+      "loss": 0.6108,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 0.55078125,
+      "rewards/rejected": -3.5625,
       "step": 620
     },
     {
       "epoch": 0.6596858638743456,
-      "grad_norm": 1.9426038158980632,
+      "grad_norm": 2.010737238386616,
       "learning_rate": 4.7847714344635864e-05,
-      "logits/chosen": -11.25,
-      "logits/rejected": -11.0,
-      "logps/chosen": -700.0,
-      "logps/rejected": -728.0,
-      "loss": 0.6332,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": -3.125,
-      "rewards/margins": 0.345703125,
-      "rewards/rejected": -3.46875,
+      "logits/chosen": -12.9375,
+      "logits/rejected": -12.75,
+      "logps/chosen": -708.0,
+      "logps/rejected": -736.0,
+      "loss": 0.6418,
+      "rewards/accuracies": 0.606249988079071,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 0.349609375,
+      "rewards/rejected": -3.5625,
       "step": 630
     },
     {
       "epoch": 0.6701570680628273,
-      "grad_norm": 2.2929882591980926,
+      "grad_norm": 2.1799778101410903,
       "learning_rate": 4.772235575174776e-05,
-      "logits/chosen": -9.5625,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -732.0,
-      "logps/rejected": -816.0,
-      "loss": 0.5859,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -3.09375,
-      "rewards/margins": 0.490234375,
-      "rewards/rejected": -3.59375,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -12.4375,
+      "logps/chosen": -716.0,
+      "logps/rejected": -792.0,
+      "loss": 0.583,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 0.427734375,
+      "rewards/rejected": -3.34375,
       "step": 640
     },
     {
       "epoch": 0.6701570680628273,
-      "eval_logits/chosen": -6.40625,
-      "eval_logits/rejected": -5.125,
-      "eval_logps/chosen": -684.0,
-      "eval_logps/rejected": -704.0,
-      "eval_loss": 0.7149375081062317,
-      "eval_rewards/accuracies": 0.619140625,
-      "eval_rewards/chosen": -3.671875,
+      "eval_logits/chosen": -10.0625,
+      "eval_logits/rejected": -9.0,
+      "eval_logps/chosen": -644.0,
+      "eval_logps/rejected": -660.0,
+      "eval_loss": 0.7081406116485596,
+      "eval_rewards/accuracies": 0.640625,
+      "eval_rewards/chosen": -3.25,
       "eval_rewards/margins": 0.46484375,
-      "eval_rewards/rejected": -4.125,
-      "eval_runtime": 46.5437,
-      "eval_samples_per_second": 42.97,
-      "eval_steps_per_second": 0.688,
+      "eval_rewards/rejected": -3.703125,
+      "eval_runtime": 58.2916,
+      "eval_samples_per_second": 34.31,
+      "eval_steps_per_second": 0.549,
       "step": 640
     },
     {
       "epoch": 0.680628272251309,
-      "grad_norm": 2.445288423492824,
+      "grad_norm": 2.4054215803524857,
       "learning_rate": 4.7593622872782754e-05,
-      "logits/chosen": -6.875,
-      "logits/rejected": -7.03125,
-      "logps/chosen": -740.0,
-      "logps/rejected": -840.0,
-      "loss": 0.5915,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 0.5625,
-      "rewards/rejected": -4.09375,
+      "logits/chosen": -9.375,
+      "logits/rejected": -9.375,
+      "logps/chosen": -732.0,
+      "logps/rejected": -836.0,
+      "loss": 0.5961,
+      "rewards/accuracies": 0.6187499761581421,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 0.60546875,
+      "rewards/rejected": -4.0625,
       "step": 650
     },
     {
       "epoch": 0.6910994764397905,
-      "grad_norm": 2.2614673448089895,
+      "grad_norm": 2.252928200142982,
       "learning_rate": 4.746153482466542e-05,
-      "logits/chosen": -8.0,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -732.0,
-      "logps/rejected": -828.0,
-      "loss": 0.6067,
-      "rewards/accuracies": 0.643750011920929,
-      "rewards/chosen": -3.328125,
-      "rewards/margins": 0.5234375,
-      "rewards/rejected": -3.84375,
+      "logits/chosen": -10.0625,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -780.0,
+      "logps/rejected": -872.0,
+      "loss": 0.6094,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -3.8125,
+      "rewards/margins": 0.498046875,
+      "rewards/rejected": -4.3125,
       "step": 660
     },
     {
       "epoch": 0.7015706806282722,
-      "grad_norm": 2.1888838015960026,
+      "grad_norm": 2.0997218503241344,
       "learning_rate": 4.732611122256537e-05,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -11.1875,
-      "logps/chosen": -744.0,
-      "logps/rejected": -816.0,
-      "loss": 0.5966,
-      "rewards/accuracies": 0.6187499761581421,
-      "rewards/chosen": -3.03125,
-      "rewards/margins": 0.373046875,
-      "rewards/rejected": -3.40625,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -13.3125,
+      "logps/chosen": -756.0,
+      "logps/rejected": -828.0,
+      "loss": 0.5997,
+      "rewards/accuracies": 0.59375,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 0.361328125,
+      "rewards/rejected": -3.53125,
       "step": 670
     },
     {
       "epoch": 0.7120418848167539,
-      "grad_norm": 2.3641764820607043,
+      "grad_norm": 2.8265544949434416,
       "learning_rate": 4.718737217698435e-05,
-      "logits/chosen": -10.1875,
-      "logits/rejected": -10.125,
+      "logits/chosen": -11.9375,
+      "logits/rejected": -11.625,
       "logps/chosen": -724.0,
-      "logps/rejected": -828.0,
-      "loss": 0.6137,
-      "rewards/accuracies": 0.668749988079071,
+      "logps/rejected": -836.0,
+      "loss": 0.6234,
+      "rewards/accuracies": 0.675000011920929,
       "rewards/chosen": -3.34375,
-      "rewards/margins": 0.56640625,
-      "rewards/rejected": -3.90625,
+      "rewards/margins": 0.62109375,
+      "rewards/rejected": -3.96875,
       "step": 680
     },
     {
       "epoch": 0.7225130890052356,
-      "grad_norm": 2.3793754160176253,
+      "grad_norm": 2.230220722979213,
       "learning_rate": 4.704533829076982e-05,
-      "logits/chosen": -9.25,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -684.0,
-      "logps/rejected": -788.0,
-      "loss": 0.5884,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -2.8125,
-      "rewards/margins": 0.44921875,
-      "rewards/rejected": -3.25,
+      "logits/chosen": -10.9375,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -704.0,
+      "logps/rejected": -812.0,
+      "loss": 0.587,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 0.486328125,
+      "rewards/rejected": -3.484375,
       "step": 690
     },
     {
       "epoch": 0.7329842931937173,
-      "grad_norm": 2.530673480396828,
+      "grad_norm": 2.604940301203485,
       "learning_rate": 4.6900030656055455e-05,
-      "logits/chosen": -9.3125,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -728.0,
-      "logps/rejected": -832.0,
-      "loss": 0.5728,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -3.09375,
-      "rewards/margins": 0.51953125,
-      "rewards/rejected": -3.625,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -744.0,
+      "logps/rejected": -856.0,
+      "loss": 0.5733,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 0.5859375,
+      "rewards/rejected": -3.875,
       "step": 700
     },
     {
       "epoch": 0.743455497382199,
-      "grad_norm": 2.412968373276436,
+      "grad_norm": 2.2293369353650996,
       "learning_rate": 4.6751470851128906e-05,
-      "logits/chosen": -8.25,
-      "logits/rejected": -7.96875,
-      "logps/chosen": -832.0,
-      "logps/rejected": -904.0,
-      "loss": 0.6029,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -3.921875,
-      "rewards/margins": 0.5546875,
-      "rewards/rejected": -4.46875,
+      "logits/chosen": -10.6875,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -804.0,
+      "logps/rejected": -876.0,
+      "loss": 0.6059,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": -3.65625,
+      "rewards/margins": 0.5390625,
+      "rewards/rejected": -4.1875,
       "step": 710
     },
     {
       "epoch": 0.7539267015706806,
-      "grad_norm": 2.08555056991513,
+      "grad_norm": 2.146040474575339,
       "learning_rate": 4.6599680937227454e-05,
-      "logits/chosen": -8.75,
-      "logits/rejected": -8.1875,
+      "logits/chosen": -10.0625,
+      "logits/rejected": -9.6875,
       "logps/chosen": -812.0,
-      "logps/rejected": -884.0,
-      "loss": 0.6127,
-      "rewards/accuracies": 0.6499999761581421,
+      "logps/rejected": -880.0,
+      "loss": 0.6183,
+      "rewards/accuracies": 0.643750011920929,
       "rewards/chosen": -3.828125,
-      "rewards/margins": 0.439453125,
-      "rewards/rejected": -4.28125,
+      "rewards/margins": 0.392578125,
+      "rewards/rejected": -4.21875,
       "step": 720
     },
     {
       "epoch": 0.7539267015706806,
-      "eval_logits/chosen": -8.0,
-      "eval_logits/rejected": -6.875,
-      "eval_logps/chosen": -688.0,
-      "eval_logps/rejected": -692.0,
-      "eval_loss": 0.7057031393051147,
-      "eval_rewards/accuracies": 0.6015625,
-      "eval_rewards/chosen": -3.703125,
-      "eval_rewards/margins": 0.33203125,
-      "eval_rewards/rejected": -4.03125,
-      "eval_runtime": 46.5391,
-      "eval_samples_per_second": 42.975,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -9.4375,
+      "eval_logits/rejected": -8.5625,
+      "eval_logps/chosen": -696.0,
+      "eval_logps/rejected": -700.0,
+      "eval_loss": 0.7397187352180481,
+      "eval_rewards/accuracies": 0.599609375,
+      "eval_rewards/chosen": -3.78125,
+      "eval_rewards/margins": 0.32421875,
+      "eval_rewards/rejected": -4.09375,
+      "eval_runtime": 58.3373,
+      "eval_samples_per_second": 34.283,
+      "eval_steps_per_second": 0.549,
       "step": 720
     },
     {
       "epoch": 0.7643979057591623,
-      "grad_norm": 1.92892075673009,
+      "grad_norm": 1.7733719464709947,
       "learning_rate": 4.644468345526183e-05,
-      "logits/chosen": -10.125,
-      "logits/rejected": -9.9375,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -11.0,
       "logps/chosen": -744.0,
-      "logps/rejected": -856.0,
-      "loss": 0.6141,
-      "rewards/accuracies": 0.65625,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 0.45703125,
-      "rewards/rejected": -3.671875,
+      "logps/rejected": -864.0,
+      "loss": 0.614,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -3.25,
+      "rewards/margins": 0.51171875,
+      "rewards/rejected": -3.75,
       "step": 730
     },
     {
       "epoch": 0.774869109947644,
-      "grad_norm": 1.9896793453255133,
+      "grad_norm": 1.8774216813881301,
       "learning_rate": 4.628650142246894e-05,
-      "logits/chosen": -9.375,
-      "logits/rejected": -8.875,
-      "logps/chosen": -728.0,
-      "logps/rejected": -844.0,
-      "loss": 0.5413,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -3.390625,
-      "rewards/margins": 0.62890625,
-      "rewards/rejected": -4.03125,
+      "logits/chosen": -10.25,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -668.0,
+      "logps/rejected": -776.0,
+      "loss": 0.5488,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -2.78125,
+      "rewards/margins": 0.55078125,
+      "rewards/rejected": -3.34375,
       "step": 740
     },
     {
       "epoch": 0.7853403141361257,
-      "grad_norm": 3.275997185894457,
+      "grad_norm": 2.580640844314193,
       "learning_rate": 4.6125158328993765e-05,
-      "logits/chosen": -7.34375,
-      "logits/rejected": -6.625,
-      "logps/chosen": -824.0,
-      "logps/rejected": -908.0,
-      "loss": 0.5912,
-      "rewards/accuracies": 0.643750011920929,
-      "rewards/chosen": -4.28125,
-      "rewards/margins": 0.625,
-      "rewards/rejected": -4.90625,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -8.5625,
+      "logps/chosen": -748.0,
+      "logps/rejected": -816.0,
+      "loss": 0.5962,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 0.453125,
+      "rewards/rejected": -3.984375,
       "step": 750
     },
     {
       "epoch": 0.7958115183246073,
-      "grad_norm": 1.8974160625820071,
+      "grad_norm": 2.129665237264523,
       "learning_rate": 4.5960678134401064e-05,
-      "logits/chosen": -9.6875,
+      "logits/chosen": -9.1875,
       "logits/rejected": -9.0,
-      "logps/chosen": -748.0,
-      "logps/rejected": -884.0,
-      "loss": 0.6161,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 0.6171875,
-      "rewards/rejected": -3.953125,
+      "logps/chosen": -732.0,
+      "logps/rejected": -876.0,
+      "loss": 0.5916,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 0.69921875,
+      "rewards/rejected": -3.859375,
       "step": 760
     },
     {
       "epoch": 0.806282722513089,
-      "grad_norm": 2.5471196954282576,
+      "grad_norm": 2.505944929814888,
       "learning_rate": 4.579308526411734e-05,
-      "logits/chosen": -10.875,
-      "logits/rejected": -10.375,
+      "logits/chosen": -10.25,
+      "logits/rejected": -9.9375,
       "logps/chosen": -684.0,
-      "logps/rejected": -776.0,
-      "loss": 0.5786,
-      "rewards/accuracies": 0.65625,
-      "rewards/chosen": -2.796875,
-      "rewards/margins": 0.52734375,
+      "logps/rejected": -772.0,
+      "loss": 0.5868,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -2.765625,
+      "rewards/margins": 0.54296875,
       "rewards/rejected": -3.3125,
       "step": 770
     },
     {
       "epoch": 0.8167539267015707,
-      "grad_norm": 2.8853511091166433,
+      "grad_norm": 2.6719352596011725,
       "learning_rate": 4.562240460580367e-05,
       "logits/chosen": -11.0,
-      "logits/rejected": -10.75,
-      "logps/chosen": -708.0,
-      "logps/rejected": -828.0,
-      "loss": 0.5956,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -3.03125,
-      "rewards/margins": 0.55859375,
-      "rewards/rejected": -3.59375,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -676.0,
+      "logps/rejected": -796.0,
+      "loss": 0.5903,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -2.703125,
+      "rewards/margins": 0.54296875,
+      "rewards/rejected": -3.25,
       "step": 780
     },
     {
       "epoch": 0.8272251308900523,
-      "grad_norm": 2.4361601497112177,
+      "grad_norm": 2.501868213095307,
       "learning_rate": 4.544866150565985e-05,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -11.1875,
-      "logps/chosen": -724.0,
-      "logps/rejected": -800.0,
-      "loss": 0.5984,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": -3.171875,
-      "rewards/margins": 0.44140625,
-      "rewards/rejected": -3.625,
+      "logits/chosen": -10.8125,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -728.0,
+      "logps/rejected": -812.0,
+      "loss": 0.5957,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 0.50390625,
+      "rewards/rejected": -3.734375,
       "step": 790
     },
     {
       "epoch": 0.837696335078534,
-      "grad_norm": 1.980712134960137,
+      "grad_norm": 2.386957909687042,
       "learning_rate": 4.527188176466051e-05,
-      "logits/chosen": -11.625,
-      "logits/rejected": -11.5,
-      "logps/chosen": -784.0,
-      "logps/rejected": -892.0,
-      "loss": 0.6012,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 0.69140625,
-      "rewards/rejected": -4.125,
+      "logits/chosen": -10.1875,
+      "logits/rejected": -9.875,
+      "logps/chosen": -868.0,
+      "logps/rejected": -992.0,
+      "loss": 0.5988,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -4.25,
+      "rewards/margins": 0.85546875,
+      "rewards/rejected": -5.125,
       "step": 800
     },
     {
       "epoch": 0.837696335078534,
-      "eval_logits/chosen": -11.125,
-      "eval_logits/rejected": -10.75,
-      "eval_logps/chosen": -684.0,
-      "eval_logps/rejected": -684.0,
-      "eval_loss": 0.751171886920929,
-      "eval_rewards/accuracies": 0.59375,
-      "eval_rewards/chosen": -3.640625,
-      "eval_rewards/margins": 0.298828125,
-      "eval_rewards/rejected": -3.9375,
-      "eval_runtime": 46.5274,
-      "eval_samples_per_second": 42.985,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -8.9375,
+      "eval_logits/rejected": -7.90625,
+      "eval_logps/chosen": -764.0,
+      "eval_logps/rejected": -784.0,
+      "eval_loss": 0.798578143119812,
+      "eval_rewards/accuracies": 0.58984375,
+      "eval_rewards/chosen": -4.46875,
+      "eval_rewards/margins": 0.4609375,
+      "eval_rewards/rejected": -4.9375,
+      "eval_runtime": 58.3286,
+      "eval_samples_per_second": 34.289,
+      "eval_steps_per_second": 0.549,
       "step": 800
     },
     {
       "epoch": 0.8481675392670157,
-      "grad_norm": 2.27220921615105,
+      "grad_norm": 2.9290901015785558,
       "learning_rate": 4.509209163472357e-05,
-      "logits/chosen": -11.5625,
-      "logits/rejected": -11.3125,
-      "logps/chosen": -792.0,
-      "logps/rejected": -888.0,
-      "loss": 0.5517,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -3.59375,
-      "rewards/margins": 0.515625,
-      "rewards/rejected": -4.125,
+      "logits/chosen": -10.375,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -860.0,
+      "logps/rejected": -960.0,
+      "loss": 0.5445,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -4.28125,
+      "rewards/margins": 0.5859375,
+      "rewards/rejected": -4.84375,
       "step": 810
     },
     {
       "epoch": 0.8586387434554974,
-      "grad_norm": 1.8781778595375338,
+      "grad_norm": 1.8154958275117952,
       "learning_rate": 4.4909317814811896e-05,
-      "logits/chosen": -11.5,
-      "logits/rejected": -11.1875,
-      "logps/chosen": -768.0,
+      "logits/chosen": -10.4375,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -772.0,
       "logps/rejected": -856.0,
-      "loss": 0.5997,
+      "loss": 0.6204,
       "rewards/accuracies": 0.65625,
-      "rewards/chosen": -3.578125,
-      "rewards/margins": 0.5625,
+      "rewards/chosen": -3.609375,
+      "rewards/margins": 0.5078125,
       "rewards/rejected": -4.125,
       "step": 820
     },
     {
       "epoch": 0.8691099476439791,
-      "grad_norm": 2.0503694088452327,
+      "grad_norm": 2.6409968043577163,
       "learning_rate": 4.472358744696845e-05,
-      "logits/chosen": -10.875,
-      "logits/rejected": -10.8125,
-      "logps/chosen": -760.0,
-      "logps/rejected": -880.0,
-      "loss": 0.5914,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 0.69140625,
-      "rewards/rejected": -4.21875,
+      "logits/chosen": -9.5,
+      "logits/rejected": -9.375,
+      "logps/chosen": -756.0,
+      "logps/rejected": -864.0,
+      "loss": 0.5984,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -3.46875,
+      "rewards/margins": 0.5859375,
+      "rewards/rejected": -4.0625,
       "step": 830
     },
     {
       "epoch": 0.8795811518324608,
-      "grad_norm": 2.5036439323716286,
+      "grad_norm": 2.8859153046540396,
       "learning_rate": 4.4534928112285645e-05,
-      "logits/chosen": -9.9375,
-      "logits/rejected": -9.625,
-      "logps/chosen": -844.0,
-      "logps/rejected": -924.0,
-      "loss": 0.5934,
-      "rewards/accuracies": 0.6312500238418579,
-      "rewards/chosen": -4.34375,
-      "rewards/margins": 0.451171875,
-      "rewards/rejected": -4.8125,
+      "logits/chosen": -7.1875,
+      "logits/rejected": -7.09375,
+      "logps/chosen": -904.0,
+      "logps/rejected": -984.0,
+      "loss": 0.6061,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 0.46484375,
+      "rewards/rejected": -5.40625,
       "step": 840
     },
     {
       "epoch": 0.8900523560209425,
-      "grad_norm": 2.2500597761643526,
+      "grad_norm": 2.183911563681,
       "learning_rate": 4.43433678268096e-05,
-      "logits/chosen": -9.375,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -780.0,
-      "logps/rejected": -840.0,
-      "loss": 0.5839,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -3.921875,
-      "rewards/margins": 0.36328125,
-      "rewards/rejected": -4.28125,
+      "logits/chosen": -7.5,
+      "logits/rejected": -7.375,
+      "logps/chosen": -820.0,
+      "logps/rejected": -876.0,
+      "loss": 0.6027,
+      "rewards/accuracies": 0.606249988079071,
+      "rewards/chosen": -4.28125,
+      "rewards/margins": 0.345703125,
+      "rewards/rejected": -4.625,
       "step": 850
     },
     {
       "epoch": 0.900523560209424,
-      "grad_norm": 2.065726853368458,
+      "grad_norm": 2.0913394364549047,
       "learning_rate": 4.414893503737968e-05,
-      "logits/chosen": -7.6875,
-      "logits/rejected": -7.1875,
-      "logps/chosen": -788.0,
-      "logps/rejected": -888.0,
-      "loss": 0.599,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -4.15625,
-      "rewards/margins": 0.546875,
-      "rewards/rejected": -4.6875,
+      "logits/chosen": -7.90625,
+      "logits/rejected": -7.8125,
+      "logps/chosen": -724.0,
+      "logps/rejected": -824.0,
+      "loss": 0.5899,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -3.5,
+      "rewards/margins": 0.51953125,
+      "rewards/rejected": -4.03125,
       "step": 860
     },
     {
       "epoch": 0.9109947643979057,
-      "grad_norm": 3.113134426519889,
+      "grad_norm": 2.4837155588325492,
       "learning_rate": 4.395165861740414e-05,
-      "logits/chosen": -6.46875,
-      "logits/rejected": -6.28125,
-      "logps/chosen": -788.0,
-      "logps/rejected": -920.0,
-      "loss": 0.594,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -4.09375,
-      "rewards/margins": 0.80078125,
-      "rewards/rejected": -4.90625,
+      "logits/chosen": -9.25,
+      "logits/rejected": -9.125,
+      "logps/chosen": -704.0,
+      "logps/rejected": -828.0,
+      "loss": 0.5831,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 0.7265625,
+      "rewards/rejected": -4.0,
       "step": 870
     },
     {
       "epoch": 0.9214659685863874,
-      "grad_norm": 2.085663920718786,
+      "grad_norm": 2.2922956694602155,
       "learning_rate": 4.375156786257241e-05,
-      "logits/chosen": -7.25,
-      "logits/rejected": -6.78125,
-      "logps/chosen": -852.0,
-      "logps/rejected": -952.0,
-      "loss": 0.5948,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -4.0,
-      "rewards/margins": 0.5546875,
-      "rewards/rejected": -4.5625,
+      "logits/chosen": -10.5,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -788.0,
+      "logps/rejected": -888.0,
+      "loss": 0.5882,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 0.53125,
+      "rewards/rejected": -3.921875,
       "step": 880
     },
     {
       "epoch": 0.9214659685863874,
-      "eval_logits/chosen": -6.875,
-      "eval_logits/rejected": -4.96875,
-      "eval_logps/chosen": -628.0,
-      "eval_logps/rejected": -632.0,
-      "eval_loss": 0.7495625019073486,
-      "eval_rewards/accuracies": 0.591796875,
-      "eval_rewards/chosen": -3.09375,
-      "eval_rewards/margins": 0.345703125,
-      "eval_rewards/rejected": -3.453125,
-      "eval_runtime": 46.5354,
-      "eval_samples_per_second": 42.978,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -9.1875,
+      "eval_logits/rejected": -8.3125,
+      "eval_logps/chosen": -644.0,
+      "eval_logps/rejected": -656.0,
+      "eval_loss": 0.7996875047683716,
+      "eval_rewards/accuracies": 0.587890625,
+      "eval_rewards/chosen": -3.265625,
+      "eval_rewards/margins": 0.390625,
+      "eval_rewards/rejected": -3.65625,
+      "eval_runtime": 58.3441,
+      "eval_samples_per_second": 34.279,
+      "eval_steps_per_second": 0.548,
       "step": 880
     },
     {
       "epoch": 0.9319371727748691,
-      "grad_norm": 1.8552387007374331,
+      "grad_norm": 1.8781466831056906,
       "learning_rate": 4.354869248650466e-05,
-      "logits/chosen": -8.625,
-      "logits/rejected": -7.96875,
-      "logps/chosen": -712.0,
-      "logps/rejected": -868.0,
-      "loss": 0.5634,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -2.921875,
-      "rewards/margins": 0.59765625,
-      "rewards/rejected": -3.53125,
+      "logits/chosen": -9.5625,
+      "logits/rejected": -9.125,
+      "logps/chosen": -752.0,
+      "logps/rejected": -928.0,
+      "loss": 0.5483,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -3.328125,
+      "rewards/margins": 0.78515625,
+      "rewards/rejected": -4.125,
       "step": 890
     },
     {
       "epoch": 0.9424083769633508,
-      "grad_norm": 3.0857784764570444,
+      "grad_norm": 2.844065787101728,
       "learning_rate": 4.334306261633931e-05,
-      "logits/chosen": -5.8125,
-      "logits/rejected": -5.5625,
-      "logps/chosen": -856.0,
-      "logps/rejected": -984.0,
-      "loss": 0.5874,
-      "rewards/accuracies": 0.637499988079071,
+      "logits/chosen": -8.375,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -860.0,
+      "logps/rejected": -992.0,
+      "loss": 0.5688,
+      "rewards/accuracies": 0.675000011920929,
       "rewards/chosen": -4.15625,
-      "rewards/margins": 0.625,
-      "rewards/rejected": -4.78125,
+      "rewards/margins": 0.68359375,
+      "rewards/rejected": -4.84375,
       "step": 900
     },
     {
       "epoch": 0.9528795811518325,
-      "grad_norm": 2.455266774958976,
+      "grad_norm": 2.420482039360459,
       "learning_rate": 4.313470878825911e-05,
-      "logits/chosen": -5.96875,
-      "logits/rejected": -5.46875,
-      "logps/chosen": -828.0,
-      "logps/rejected": -912.0,
-      "loss": 0.6125,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -4.1875,
-      "rewards/margins": 0.63671875,
-      "rewards/rejected": -4.84375,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -8.625,
+      "logps/chosen": -756.0,
+      "logps/rejected": -836.0,
+      "loss": 0.6142,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -3.484375,
+      "rewards/margins": 0.59765625,
+      "rewards/rejected": -4.0625,
       "step": 910
     },
     {
       "epoch": 0.9633507853403142,
-      "grad_norm": 1.8740293737898663,
+      "grad_norm": 1.9289631829505745,
       "learning_rate": 4.292366194295652e-05,
-      "logits/chosen": -5.6875,
-      "logits/rejected": -4.96875,
-      "logps/chosen": -840.0,
-      "logps/rejected": -904.0,
-      "loss": 0.5646,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -4.09375,
-      "rewards/margins": 0.60546875,
-      "rewards/rejected": -4.71875,
+      "logits/chosen": -8.75,
+      "logits/rejected": -8.375,
+      "logps/chosen": -760.0,
+      "logps/rejected": -824.0,
+      "loss": 0.579,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 0.5625,
+      "rewards/rejected": -3.90625,
       "step": 920
     },
     {
       "epoch": 0.9738219895287958,
-      "grad_norm": 2.494935804478786,
+      "grad_norm": 2.222476171506996,
       "learning_rate": 4.2709953421038964e-05,
-      "logits/chosen": -5.15625,
-      "logits/rejected": -4.90625,
-      "logps/chosen": -776.0,
-      "logps/rejected": -900.0,
-      "loss": 0.6009,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -3.828125,
-      "rewards/margins": 0.59375,
-      "rewards/rejected": -4.4375,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -704.0,
+      "logps/rejected": -824.0,
+      "loss": 0.592,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 0.58203125,
+      "rewards/rejected": -3.6875,
       "step": 930
     },
     {
       "epoch": 0.9842931937172775,
-      "grad_norm": 2.002069849447359,
+      "grad_norm": 2.091928898579189,
       "learning_rate": 4.249361495837478e-05,
-      "logits/chosen": -5.65625,
-      "logits/rejected": -5.28125,
-      "logps/chosen": -780.0,
-      "logps/rejected": -916.0,
-      "loss": 0.589,
-      "rewards/accuracies": 0.65625,
-      "rewards/chosen": -3.859375,
-      "rewards/margins": 0.58203125,
-      "rewards/rejected": -4.4375,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -712.0,
+      "logps/rejected": -852.0,
+      "loss": 0.5939,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 0.62109375,
+      "rewards/rejected": -3.828125,
       "step": 940
     },
     {
       "epoch": 0.9947643979057592,
-      "grad_norm": 2.0512402994781462,
+      "grad_norm": 1.9187007533600982,
       "learning_rate": 4.227467868138035e-05,
-      "logits/chosen": -6.625,
-      "logits/rejected": -6.0,
-      "logps/chosen": -800.0,
-      "logps/rejected": -956.0,
-      "loss": 0.5763,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": -3.8125,
-      "rewards/margins": 0.68359375,
-      "rewards/rejected": -4.5,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -8.5625,
+      "logps/chosen": -736.0,
+      "logps/rejected": -896.0,
+      "loss": 0.5637,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 0.7265625,
+      "rewards/rejected": -3.890625,
       "step": 950
     },
     {
       "epoch": 1.0052356020942408,
-      "grad_norm": 1.066837918927463,
+      "grad_norm": 1.2835431765487857,
       "learning_rate": 4.205317710224938e-05,
-      "logits/chosen": -8.0,
-      "logits/rejected": -6.9375,
-      "logps/chosen": -772.0,
-      "logps/rejected": -920.0,
-      "loss": 0.434,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 1.3671875,
-      "rewards/rejected": -4.8125,
+      "logits/chosen": -9.75,
+      "logits/rejected": -9.125,
+      "logps/chosen": -752.0,
+      "logps/rejected": -904.0,
+      "loss": 0.4256,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 1.3828125,
+      "rewards/rejected": -4.59375,
       "step": 960
     },
     {
       "epoch": 1.0052356020942408,
-      "eval_logits/chosen": -4.75,
-      "eval_logits/rejected": -2.890625,
-      "eval_logps/chosen": -760.0,
-      "eval_logps/rejected": -784.0,
-      "eval_loss": 0.7349374890327454,
-      "eval_rewards/accuracies": 0.62890625,
-      "eval_rewards/chosen": -4.40625,
-      "eval_rewards/margins": 0.5703125,
-      "eval_rewards/rejected": -4.96875,
-      "eval_runtime": 46.5293,
-      "eval_samples_per_second": 42.984,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -7.90625,
+      "eval_logits/rejected": -6.75,
+      "eval_logps/chosen": -772.0,
+      "eval_logps/rejected": -808.0,
+      "eval_loss": 0.7816093564033508,
+      "eval_rewards/accuracies": 0.6171875,
+      "eval_rewards/chosen": -4.53125,
+      "eval_rewards/margins": 0.63671875,
+      "eval_rewards/rejected": -5.1875,
+      "eval_runtime": 58.3925,
+      "eval_samples_per_second": 34.251,
+      "eval_steps_per_second": 0.548,
       "step": 960
     },
     {
       "epoch": 1.0157068062827226,
-      "grad_norm": 1.2213397070491476,
+      "grad_norm": 1.3372085103118,
       "learning_rate": 4.1829143114124733e-05,
-      "logits/chosen": -6.5625,
-      "logits/rejected": -5.875,
-      "logps/chosen": -836.0,
-      "logps/rejected": -1160.0,
-      "loss": 0.231,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -4.25,
-      "rewards/margins": 2.515625,
-      "rewards/rejected": -6.75,
+      "logits/chosen": -9.8125,
+      "logits/rejected": -9.25,
+      "logps/chosen": -832.0,
+      "logps/rejected": -1168.0,
+      "loss": 0.229,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -4.21875,
+      "rewards/margins": 2.625,
+      "rewards/rejected": -6.84375,
       "step": 970
     },
     {
       "epoch": 1.0261780104712042,
-      "grad_norm": 2.202008442076433,
+      "grad_norm": 2.0386632087058287,
       "learning_rate": 4.160260998621387e-05,
-      "logits/chosen": -4.84375,
-      "logits/rejected": -4.1875,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -1496.0,
-      "loss": 0.1876,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -6.5625,
-      "rewards/margins": 3.59375,
-      "rewards/rejected": -10.125,
+      "logits/chosen": -10.0,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1440.0,
+      "loss": 0.1941,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -5.96875,
+      "rewards/margins": 3.640625,
+      "rewards/rejected": -9.625,
       "step": 980
     },
     {
       "epoch": 1.036649214659686,
-      "grad_norm": 1.5717852056531012,
+      "grad_norm": 1.4638663475573397,
       "learning_rate": 4.137361135884824e-05,
-      "logits/chosen": -5.5,
-      "logits/rejected": -4.8125,
-      "logps/chosen": -940.0,
-      "logps/rejected": -1320.0,
-      "loss": 0.1877,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -5.15625,
-      "rewards/margins": 3.140625,
-      "rewards/rejected": -8.3125,
+      "logits/chosen": -10.6875,
+      "logits/rejected": -9.875,
+      "logps/chosen": -876.0,
+      "logps/rejected": -1256.0,
+      "loss": 0.1863,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -4.53125,
+      "rewards/margins": 3.171875,
+      "rewards/rejected": -7.6875,
       "step": 990
     },
     {
       "epoch": 1.0471204188481675,
-      "grad_norm": 2.317546921932559,
+      "grad_norm": 1.9684360770743778,
       "learning_rate": 4.114218123848777e-05,
-      "logits/chosen": -5.53125,
-      "logits/rejected": -4.875,
-      "logps/chosen": -856.0,
-      "logps/rejected": -1240.0,
-      "loss": 0.1871,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -4.53125,
-      "rewards/margins": 3.21875,
-      "rewards/rejected": -7.75,
+      "logits/chosen": -9.25,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -968.0,
+      "logps/rejected": -1360.0,
+      "loss": 0.1954,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -5.65625,
+      "rewards/margins": 3.265625,
+      "rewards/rejected": -8.9375,
       "step": 1000
     },
     {
       "epoch": 1.057591623036649,
-      "grad_norm": 2.371243839247179,
+      "grad_norm": 4.18066978689384,
       "learning_rate": 4.0908353992670824e-05,
-      "logits/chosen": -6.15625,
-      "logits/rejected": -5.1875,
-      "logps/chosen": -872.0,
-      "logps/rejected": -1248.0,
-      "loss": 0.2321,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -4.75,
-      "rewards/margins": 3.125,
-      "rewards/rejected": -7.875,
+      "logits/chosen": -10.8125,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -820.0,
+      "logps/rejected": -1176.0,
+      "loss": 0.233,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.28125,
+      "rewards/margins": 2.890625,
+      "rewards/rejected": -7.15625,
       "step": 1010
     },
     {
       "epoch": 1.068062827225131,
-      "grad_norm": 2.099749505595064,
+      "grad_norm": 2.069438843716654,
       "learning_rate": 4.067216434491059e-05,
-      "logits/chosen": -4.1875,
-      "logits/rejected": -3.765625,
-      "logps/chosen": -980.0,
-      "logps/rejected": -1368.0,
-      "loss": 0.1798,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -5.71875,
-      "rewards/margins": 3.15625,
-      "rewards/rejected": -8.875,
+      "logits/chosen": -9.8125,
+      "logits/rejected": -9.25,
+      "logps/chosen": -896.0,
+      "logps/rejected": -1288.0,
+      "loss": 0.1865,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 3.234375,
+      "rewards/rejected": -8.125,
       "step": 1020
     },
     {
       "epoch": 1.0785340314136125,
-      "grad_norm": 2.5253449214411603,
+      "grad_norm": 2.544150864766691,
       "learning_rate": 4.043364736953866e-05,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -2.8125,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1472.0,
-      "loss": 0.209,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.3125,
-      "rewards/margins": 3.578125,
-      "rewards/rejected": -9.875,
+      "logits/chosen": -8.75,
+      "logits/rejected": -7.84375,
+      "logps/chosen": -992.0,
+      "logps/rejected": -1368.0,
+      "loss": 0.2215,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.25,
+      "rewards/margins": 3.59375,
+      "rewards/rejected": -8.8125,
       "step": 1030
     },
     {
       "epoch": 1.0890052356020943,
-      "grad_norm": 2.85630609552847,
+      "grad_norm": 2.448939646513708,
       "learning_rate": 4.019283848649643e-05,
-      "logits/chosen": -5.59375,
-      "logits/rejected": -4.5625,
-      "logps/chosen": -1008.0,
-      "logps/rejected": -1424.0,
-      "loss": 0.1948,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -5.84375,
-      "rewards/margins": 3.640625,
-      "rewards/rejected": -9.5,
+      "logits/chosen": -9.75,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -860.0,
+      "logps/rejected": -1248.0,
+      "loss": 0.2006,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.34375,
+      "rewards/margins": 3.359375,
+      "rewards/rejected": -7.71875,
       "step": 1040
     },
     {
       "epoch": 1.0890052356020943,
-      "eval_logits/chosen": -0.84765625,
-      "eval_logits/rejected": 1.125,
-      "eval_logps/chosen": -1016.0,
-      "eval_logps/rejected": -1064.0,
-      "eval_loss": 1.037656307220459,
-      "eval_rewards/accuracies": 0.62109375,
-      "eval_rewards/chosen": -7.0,
-      "eval_rewards/margins": 0.75390625,
-      "eval_rewards/rejected": -7.75,
-      "eval_runtime": 46.553,
-      "eval_samples_per_second": 42.962,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -6.0625,
+      "eval_logits/rejected": -4.78125,
+      "eval_logps/chosen": -916.0,
+      "eval_logps/rejected": -960.0,
+      "eval_loss": 0.973437488079071,
+      "eval_rewards/accuracies": 0.609375,
+      "eval_rewards/chosen": -5.96875,
+      "eval_rewards/margins": 0.73828125,
+      "eval_rewards/rejected": -6.6875,
+      "eval_runtime": 58.4025,
+      "eval_samples_per_second": 34.245,
+      "eval_steps_per_second": 0.548,
       "step": 1040
     },
     {
       "epoch": 1.0994764397905759,
-      "grad_norm": 1.9142716366939394,
+      "grad_norm": 2.1497227730336754,
       "learning_rate": 3.9949773456075215e-05,
-      "logits/chosen": -3.796875,
-      "logits/rejected": -2.65625,
-      "logps/chosen": -856.0,
-      "logps/rejected": -1256.0,
-      "loss": 0.1838,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -4.75,
-      "rewards/margins": 3.34375,
-      "rewards/rejected": -8.0625,
+      "logits/chosen": -7.625,
+      "logits/rejected": -6.84375,
+      "logps/chosen": -800.0,
+      "logps/rejected": -1176.0,
+      "loss": 0.1983,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -4.1875,
+      "rewards/margins": 3.078125,
+      "rewards/rejected": -7.28125,
       "step": 1050
     },
     {
       "epoch": 1.1099476439790577,
-      "grad_norm": 1.615766539007503,
+      "grad_norm": 1.5023449068212837,
       "learning_rate": 3.970448837360585e-05,
-      "logits/chosen": -3.78125,
-      "logits/rejected": -2.9375,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -1512.0,
-      "loss": 0.1607,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -6.25,
-      "rewards/margins": 3.5,
-      "rewards/rejected": -9.75,
+      "logits/chosen": -7.6875,
+      "logits/rejected": -7.0,
+      "logps/chosen": -980.0,
+      "logps/rejected": -1384.0,
+      "loss": 0.1782,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -5.25,
+      "rewards/margins": 3.234375,
+      "rewards/rejected": -8.5,
       "step": 1060
     },
     {
       "epoch": 1.1204188481675392,
-      "grad_norm": 1.8704512495057821,
+      "grad_norm": 1.9771244287178242,
       "learning_rate": 3.945701966409846e-05,
-      "logits/chosen": -2.078125,
-      "logits/rejected": -2.0,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -1744.0,
-      "loss": 0.1973,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.125,
-      "rewards/margins": 4.1875,
-      "rewards/rejected": -12.3125,
+      "logits/chosen": -6.65625,
+      "logits/rejected": -6.34375,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -1584.0,
+      "loss": 0.2099,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.71875,
+      "rewards/margins": 4.03125,
+      "rewards/rejected": -10.75,
       "step": 1070
     },
     {
       "epoch": 1.130890052356021,
-      "grad_norm": 1.6907597612921286,
+      "grad_norm": 1.5727244264505325,
       "learning_rate": 3.9207404076833374e-05,
-      "logits/chosen": -4.0625,
-      "logits/rejected": -3.5,
-      "logps/chosen": -1020.0,
-      "logps/rejected": -1368.0,
-      "loss": 0.1769,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -5.6875,
-      "rewards/margins": 3.421875,
-      "rewards/rejected": -9.125,
+      "logits/chosen": -8.0,
+      "logits/rejected": -7.4375,
+      "logps/chosen": -920.0,
+      "logps/rejected": -1256.0,
+      "loss": 0.1869,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 3.359375,
+      "rewards/rejected": -8.0625,
       "step": 1080
     },
     {
       "epoch": 1.1413612565445026,
-      "grad_norm": 1.665355486984576,
+      "grad_norm": 2.204039552908947,
       "learning_rate": 3.895567867990379e-05,
-      "logits/chosen": -5.0625,
-      "logits/rejected": -3.640625,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -1440.0,
-      "loss": 0.1457,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -5.8125,
-      "rewards/margins": 3.578125,
-      "rewards/rejected": -9.375,
+      "logits/chosen": -9.3125,
+      "logits/rejected": -8.125,
+      "logps/chosen": -932.0,
+      "logps/rejected": -1296.0,
+      "loss": 0.1554,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -4.53125,
+      "rewards/margins": 3.484375,
+      "rewards/rejected": -8.0,
       "step": 1090
     },
     {
       "epoch": 1.1518324607329844,
-      "grad_norm": 2.240797057948719,
+      "grad_norm": 1.91059798524668,
       "learning_rate": 3.870188085471114e-05,
-      "logits/chosen": -2.84375,
-      "logits/rejected": -2.0625,
-      "logps/chosen": -1144.0,
-      "logps/rejected": -1504.0,
-      "loss": 0.1654,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -7.28125,
-      "rewards/margins": 3.671875,
-      "rewards/rejected": -10.9375,
+      "logits/chosen": -7.96875,
+      "logits/rejected": -7.21875,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1384.0,
+      "loss": 0.1682,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -6.21875,
+      "rewards/margins": 3.5,
+      "rewards/rejected": -9.75,
       "step": 1100
     },
     {
       "epoch": 1.162303664921466,
-      "grad_norm": 2.5272516941083296,
+      "grad_norm": 2.4306822581106156,
       "learning_rate": 3.844604829041395e-05,
-      "logits/chosen": -2.0,
-      "logits/rejected": -1.6484375,
-      "logps/chosen": -980.0,
-      "logps/rejected": -1368.0,
-      "loss": 0.1849,
+      "logits/chosen": -8.5,
+      "logits/rejected": -7.71875,
+      "logps/chosen": -872.0,
+      "logps/rejected": -1256.0,
+      "loss": 0.1879,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -6.0625,
-      "rewards/margins": 3.578125,
-      "rewards/rejected": -9.625,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 3.53125,
+      "rewards/rejected": -8.5,
       "step": 1110
     },
     {
       "epoch": 1.1727748691099475,
-      "grad_norm": 2.692359212013677,
+      "grad_norm": 2.7649015881461234,
       "learning_rate": 3.818821897833095e-05,
-      "logits/chosen": -0.88671875,
-      "logits/rejected": -0.87109375,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1504.0,
-      "loss": 0.2139,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -6.65625,
-      "rewards/margins": 3.59375,
-      "rewards/rejected": -10.25,
+      "logits/chosen": -7.75,
+      "logits/rejected": -7.4375,
+      "logps/chosen": -924.0,
+      "logps/rejected": -1320.0,
+      "loss": 0.1977,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.90625,
+      "rewards/margins": 3.46875,
+      "rewards/rejected": -8.375,
       "step": 1120
     },
     {
       "epoch": 1.1727748691099475,
-      "eval_logits/chosen": 1.546875,
-      "eval_logits/rejected": 3.296875,
-      "eval_logps/chosen": -1088.0,
-      "eval_logps/rejected": -1128.0,
-      "eval_loss": 0.9297031164169312,
-      "eval_rewards/accuracies": 0.61328125,
-      "eval_rewards/chosen": -7.71875,
-      "eval_rewards/margins": 0.65625,
-      "eval_rewards/rejected": -8.375,
-      "eval_runtime": 46.5308,
-      "eval_samples_per_second": 42.982,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -6.25,
+      "eval_logits/rejected": -5.0,
+      "eval_logps/chosen": -944.0,
+      "eval_logps/rejected": -988.0,
+      "eval_loss": 0.9420312643051147,
+      "eval_rewards/accuracies": 0.609375,
+      "eval_rewards/chosen": -6.25,
+      "eval_rewards/margins": 0.7578125,
+      "eval_rewards/rejected": -7.0,
+      "eval_runtime": 58.3871,
+      "eval_samples_per_second": 34.254,
+      "eval_steps_per_second": 0.548,
       "step": 1120
     },
     {
       "epoch": 1.1832460732984293,
-      "grad_norm": 2.276901711157189,
+      "grad_norm": 2.4001157771202584,
       "learning_rate": 3.792843120629935e-05,
-      "logits/chosen": -1.3046875,
-      "logits/rejected": -0.53125,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -1456.0,
-      "loss": 0.1958,
+      "logits/chosen": -7.5625,
+      "logits/rejected": -6.71875,
+      "logps/chosen": -944.0,
+      "logps/rejected": -1368.0,
+      "loss": 0.1961,
       "rewards/accuracies": 0.9375,
-      "rewards/chosen": -6.3125,
-      "rewards/margins": 3.34375,
-      "rewards/rejected": -9.625,
+      "rewards/chosen": -5.40625,
+      "rewards/margins": 3.390625,
+      "rewards/rejected": -8.8125,
       "step": 1130
     },
     {
       "epoch": 1.193717277486911,
-      "grad_norm": 2.4669133987809158,
+      "grad_norm": 2.3819166366576687,
       "learning_rate": 3.766672355298907e-05,
-      "logits/chosen": -1.4453125,
-      "logits/rejected": -1.15625,
-      "logps/chosen": -1012.0,
-      "logps/rejected": -1392.0,
-      "loss": 0.185,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -6.34375,
-      "rewards/margins": 3.359375,
-      "rewards/rejected": -9.6875,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -7.8125,
+      "logps/chosen": -908.0,
+      "logps/rejected": -1272.0,
+      "loss": 0.1914,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.28125,
+      "rewards/margins": 3.234375,
+      "rewards/rejected": -8.5,
       "step": 1140
     },
     {
       "epoch": 1.2041884816753927,
-      "grad_norm": 1.7276286196060537,
+      "grad_norm": 2.1368239751680687,
       "learning_rate": 3.740313488217373e-05,
-      "logits/chosen": -1.78125,
-      "logits/rejected": -1.7109375,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -1528.0,
-      "loss": 0.1866,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -6.625,
-      "rewards/margins": 3.75,
-      "rewards/rejected": -10.375,
+      "logits/chosen": -7.9375,
+      "logits/rejected": -7.5,
+      "logps/chosen": -992.0,
+      "logps/rejected": -1464.0,
+      "loss": 0.1904,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -6.0625,
+      "rewards/margins": 3.640625,
+      "rewards/rejected": -9.6875,
       "step": 1150
     },
     {
       "epoch": 1.2146596858638743,
-      "grad_norm": 2.646351680097224,
+      "grad_norm": 2.2543317571072503,
       "learning_rate": 3.7137704336959464e-05,
-      "logits/chosen": -0.9296875,
-      "logits/rejected": -0.84375,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -1416.0,
-      "loss": 0.198,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -6.8125,
-      "rewards/margins": 3.1875,
-      "rewards/rejected": -10.0,
+      "logits/chosen": -6.59375,
+      "logits/rejected": -6.3125,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -1408.0,
+      "loss": 0.2087,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -6.71875,
+      "rewards/margins": 3.15625,
+      "rewards/rejected": -9.875,
       "step": 1160
     },
     {
       "epoch": 1.225130890052356,
-      "grad_norm": 2.34402813671587,
+      "grad_norm": 1.9764340447099664,
       "learning_rate": 3.687047133397201e-05,
-      "logits/chosen": -3.9375,
-      "logits/rejected": -2.6875,
-      "logps/chosen": -936.0,
-      "logps/rejected": -1312.0,
-      "loss": 0.1866,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -5.40625,
-      "rewards/margins": 3.40625,
-      "rewards/rejected": -8.8125,
+      "logits/chosen": -8.375,
+      "logits/rejected": -7.34375,
+      "logps/chosen": -904.0,
+      "logps/rejected": -1280.0,
+      "loss": 0.1852,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 3.390625,
+      "rewards/rejected": -8.5,
       "step": 1170
     },
     {
       "epoch": 1.2356020942408377,
-      "grad_norm": 1.9751499004445527,
+      "grad_norm": 2.003203347467155,
       "learning_rate": 3.6601475557503415e-05,
-      "logits/chosen": -3.578125,
-      "logits/rejected": -2.5,
-      "logps/chosen": -964.0,
-      "logps/rejected": -1304.0,
-      "loss": 0.1918,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -5.5,
-      "rewards/margins": 2.921875,
-      "rewards/rejected": -8.375,
+      "logits/chosen": -8.1875,
+      "logits/rejected": -7.15625,
+      "logps/chosen": -948.0,
+      "logps/rejected": -1280.0,
+      "loss": 0.1982,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.3125,
+      "rewards/margins": 2.84375,
+      "rewards/rejected": -8.1875,
       "step": 1180
     },
     {
       "epoch": 1.2460732984293195,
-      "grad_norm": 2.3255112663690993,
+      "grad_norm": 1.8077521135856842,
       "learning_rate": 3.633075695361882e-05,
-      "logits/chosen": -1.8828125,
-      "logits/rejected": -1.3125,
-      "logps/chosen": -964.0,
-      "logps/rejected": -1392.0,
-      "loss": 0.1748,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -5.78125,
-      "rewards/margins": 3.71875,
-      "rewards/rejected": -9.5,
+      "logits/chosen": -7.25,
+      "logits/rejected": -6.4375,
+      "logps/chosen": -936.0,
+      "logps/rejected": -1360.0,
+      "loss": 0.1874,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -5.53125,
+      "rewards/margins": 3.671875,
+      "rewards/rejected": -9.1875,
       "step": 1190
     },
     {
       "epoch": 1.256544502617801,
-      "grad_norm": 2.1361076374285806,
+      "grad_norm": 1.73549578290112,
       "learning_rate": 3.6058355724224477e-05,
-      "logits/chosen": -2.109375,
-      "logits/rejected": -1.5390625,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -1528.0,
-      "loss": 0.1666,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -6.78125,
-      "rewards/margins": 3.65625,
-      "rewards/rejected": -10.4375,
+      "logits/chosen": -7.8125,
+      "logits/rejected": -7.09375,
+      "logps/chosen": -1020.0,
+      "logps/rejected": -1464.0,
+      "loss": 0.1717,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.15625,
+      "rewards/margins": 3.640625,
+      "rewards/rejected": -9.8125,
       "step": 1200
     },
     {
       "epoch": 1.256544502617801,
-      "eval_logits/chosen": 1.0859375,
-      "eval_logits/rejected": 2.921875,
-      "eval_logps/chosen": -1152.0,
-      "eval_logps/rejected": -1200.0,
-      "eval_loss": 1.0331875085830688,
-      "eval_rewards/accuracies": 0.583984375,
-      "eval_rewards/chosen": -8.375,
-      "eval_rewards/margins": 0.7109375,
-      "eval_rewards/rejected": -9.0625,
-      "eval_runtime": 46.5366,
-      "eval_samples_per_second": 42.977,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -5.8125,
+      "eval_logits/rejected": -4.5,
+      "eval_logps/chosen": -1064.0,
+      "eval_logps/rejected": -1112.0,
+      "eval_loss": 1.054781198501587,
+      "eval_rewards/accuracies": 0.591796875,
+      "eval_rewards/chosen": -7.46875,
+      "eval_rewards/margins": 0.78515625,
+      "eval_rewards/rejected": -8.25,
+      "eval_runtime": 58.3831,
+      "eval_samples_per_second": 34.256,
+      "eval_steps_per_second": 0.548,
       "step": 1200
     },
     {
       "epoch": 1.2670157068062826,
-      "grad_norm": 3.0725363592181147,
+      "grad_norm": 2.6059069333752314,
       "learning_rate": 3.5784312321097744e-05,
-      "logits/chosen": -1.625,
-      "logits/rejected": -1.015625,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -1568.0,
-      "loss": 0.1829,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.15625,
-      "rewards/margins": 3.921875,
-      "rewards/rejected": -11.0625,
+      "logits/chosen": -7.25,
+      "logits/rejected": -6.53125,
+      "logps/chosen": -1024.0,
+      "logps/rejected": -1456.0,
+      "loss": 0.1733,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.03125,
+      "rewards/margins": 3.890625,
+      "rewards/rejected": -9.9375,
       "step": 1210
     },
     {
       "epoch": 1.2774869109947644,
-      "grad_norm": 2.121287380029104,
+      "grad_norm": 2.458974665082335,
       "learning_rate": 3.550866743987994e-05,
-      "logits/chosen": -3.34375,
-      "logits/rejected": -2.609375,
-      "logps/chosen": -1120.0,
-      "logps/rejected": -1536.0,
-      "loss": 0.1949,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.71875,
-      "rewards/margins": 3.65625,
-      "rewards/rejected": -10.375,
+      "logits/chosen": -7.125,
+      "logits/rejected": -6.4375,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -1528.0,
+      "loss": 0.182,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -6.40625,
+      "rewards/margins": 3.84375,
+      "rewards/rejected": -10.25,
       "step": 1220
     },
     {
       "epoch": 1.2879581151832462,
-      "grad_norm": 1.491496770075195,
+      "grad_norm": 1.9092757234052242,
       "learning_rate": 3.5231462014033024e-05,
-      "logits/chosen": -4.5625,
-      "logits/rejected": -3.859375,
-      "logps/chosen": -1008.0,
-      "logps/rejected": -1392.0,
-      "loss": 0.1605,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -6.09375,
-      "rewards/margins": 3.296875,
-      "rewards/rejected": -9.375,
+      "logits/chosen": -7.53125,
+      "logits/rejected": -6.9375,
+      "logps/chosen": -964.0,
+      "logps/rejected": -1344.0,
+      "loss": 0.1734,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.6875,
+      "rewards/margins": 3.28125,
+      "rewards/rejected": -8.9375,
       "step": 1230
     },
     {
       "epoch": 1.2984293193717278,
-      "grad_norm": 3.019117996273314,
+      "grad_norm": 2.339231191517696,
       "learning_rate": 3.495273720876095e-05,
-      "logits/chosen": -2.140625,
-      "logits/rejected": -1.71875,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -1600.0,
-      "loss": 0.1599,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.5625,
-      "rewards/margins": 3.90625,
-      "rewards/rejected": -11.4375,
+      "logits/chosen": -5.78125,
+      "logits/rejected": -5.5,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1440.0,
+      "loss": 0.1671,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -6.25,
+      "rewards/margins": 3.609375,
+      "rewards/rejected": -9.875,
       "step": 1240
     },
     {
       "epoch": 1.3089005235602094,
-      "grad_norm": 2.5004299185202004,
+      "grad_norm": 2.2625415077647304,
       "learning_rate": 3.4672534414896616e-05,
-      "logits/chosen": -3.0,
-      "logits/rejected": -1.96875,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -1744.0,
-      "loss": 0.1897,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -8.625,
-      "rewards/margins": 4.125,
-      "rewards/rejected": -12.75,
+      "logits/chosen": -6.6875,
+      "logits/rejected": -6.125,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -1544.0,
+      "loss": 0.1814,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -7.0,
+      "rewards/margins": 3.6875,
+      "rewards/rejected": -10.6875,
       "step": 1250
     },
     {
       "epoch": 1.3193717277486912,
-      "grad_norm": 1.587001097569015,
+      "grad_norm": 1.4560634390897056,
       "learning_rate": 3.439089524275526e-05,
-      "logits/chosen": -2.578125,
-      "logits/rejected": -1.1015625,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -1416.0,
-      "loss": 0.168,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -6.5625,
-      "rewards/margins": 3.203125,
-      "rewards/rejected": -9.75,
-      "step": 1260
-    },
+      "logits/chosen": -6.21875,
+      "logits/rejected": -5.375,
+      "logps/chosen": -1020.0,
+      "logps/rejected": -1376.0,
+      "loss": 0.1635,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -6.21875,
+      "rewards/margins": 3.21875,
+      "rewards/rejected": -9.4375,
+      "step": 1260
+    },
     {
       "epoch": 1.3298429319371727,
-      "grad_norm": 1.977571290493062,
+      "grad_norm": 2.4980264471385825,
       "learning_rate": 3.410786151595533e-05,
-      "logits/chosen": -2.859375,
-      "logits/rejected": -2.265625,
-      "logps/chosen": -1000.0,
-      "logps/rejected": -1408.0,
-      "loss": 0.1702,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -6.09375,
-      "rewards/margins": 3.390625,
+      "logits/chosen": -6.15625,
+      "logits/rejected": -5.75,
+      "logps/chosen": -988.0,
+      "logps/rejected": -1400.0,
+      "loss": 0.1804,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -5.96875,
+      "rewards/margins": 3.453125,
       "rewards/rejected": -9.4375,
       "step": 1270
     },
     {
       "epoch": 1.3403141361256545,
-      "grad_norm": 1.9617940818632646,
+      "grad_norm": 2.008585083483667,
       "learning_rate": 3.382347526520765e-05,
-      "logits/chosen": -3.34375,
-      "logits/rejected": -1.953125,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -1392.0,
-      "loss": 0.1994,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -6.3125,
-      "rewards/margins": 3.40625,
-      "rewards/rejected": -9.6875,
+      "logits/chosen": -6.3125,
+      "logits/rejected": -5.59375,
+      "logps/chosen": -968.0,
+      "logps/rejected": -1312.0,
+      "loss": 0.1881,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -5.6875,
+      "rewards/margins": 3.15625,
+      "rewards/rejected": -8.875,
       "step": 1280
     },
     {
       "epoch": 1.3403141361256545,
-      "eval_logits/chosen": -1.6796875,
-      "eval_logits/rejected": 0.486328125,
-      "eval_logps/chosen": -1064.0,
-      "eval_logps/rejected": -1112.0,
-      "eval_loss": 0.9937343597412109,
-      "eval_rewards/accuracies": 0.591796875,
-      "eval_rewards/chosen": -7.46875,
-      "eval_rewards/margins": 0.796875,
-      "eval_rewards/rejected": -8.25,
-      "eval_runtime": 46.5375,
-      "eval_samples_per_second": 42.976,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -4.46875,
+      "eval_logits/rejected": -3.21875,
+      "eval_logps/chosen": -1012.0,
+      "eval_logps/rejected": -1072.0,
+      "eval_loss": 0.956656277179718,
+      "eval_rewards/accuracies": 0.603515625,
+      "eval_rewards/chosen": -6.96875,
+      "eval_rewards/margins": 0.8671875,
+      "eval_rewards/rejected": -7.8125,
+      "eval_runtime": 58.3583,
+      "eval_samples_per_second": 34.271,
+      "eval_steps_per_second": 0.548,
       "step": 1280
     },
     {
       "epoch": 1.350785340314136,
-      "grad_norm": 3.072252987469703,
+      "grad_norm": 2.3661176844397604,
       "learning_rate": 3.353777872207381e-05,
-      "logits/chosen": -3.25,
-      "logits/rejected": -2.484375,
-      "logps/chosen": -1008.0,
-      "logps/rejected": -1480.0,
-      "loss": 0.1668,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -5.9375,
-      "rewards/margins": 3.796875,
-      "rewards/rejected": -9.75,
+      "logits/chosen": -6.09375,
+      "logits/rejected": -5.8125,
+      "logps/chosen": -940.0,
+      "logps/rejected": -1392.0,
+      "loss": 0.1837,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.28125,
+      "rewards/margins": 3.625,
+      "rewards/rejected": -8.875,
       "step": 1290
     },
     {
       "epoch": 1.3612565445026177,
-      "grad_norm": 2.189078126101991,
+      "grad_norm": 2.106031536408807,
       "learning_rate": 3.325081431269475e-05,
-      "logits/chosen": -1.3125,
-      "logits/rejected": -0.45703125,
-      "logps/chosen": -936.0,
-      "logps/rejected": -1296.0,
-      "loss": 0.1741,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -5.65625,
-      "rewards/margins": 3.40625,
-      "rewards/rejected": -9.0625,
+      "logits/chosen": -5.34375,
+      "logits/rejected": -4.9375,
+      "logps/chosen": -868.0,
+      "logps/rejected": -1200.0,
+      "loss": 0.1795,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 3.09375,
+      "rewards/rejected": -8.0625,
       "step": 1300
     },
     {
       "epoch": 1.3717277486910995,
-      "grad_norm": 3.2965708942244003,
+      "grad_norm": 3.1932735518252398,
       "learning_rate": 3.296262465149046e-05,
-      "logits/chosen": -3.21875,
-      "logits/rejected": -1.4609375,
-      "logps/chosen": -976.0,
-      "logps/rejected": -1392.0,
-      "loss": 0.1674,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -5.5,
-      "rewards/margins": 3.515625,
-      "rewards/rejected": -9.0,
+      "logits/chosen": -7.1875,
+      "logits/rejected": -6.15625,
+      "logps/chosen": -900.0,
+      "logps/rejected": -1296.0,
+      "loss": 0.1741,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -4.75,
+      "rewards/margins": 3.328125,
+      "rewards/rejected": -8.0625,
       "step": 1310
     },
     {
       "epoch": 1.3821989528795813,
-      "grad_norm": 2.004056972467981,
+      "grad_norm": 2.0299107380831525,
       "learning_rate": 3.2673252534831687e-05,
-      "logits/chosen": -4.375,
-      "logits/rejected": -2.921875,
-      "logps/chosen": -944.0,
-      "logps/rejected": -1352.0,
-      "loss": 0.1756,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -5.34375,
-      "rewards/margins": 3.46875,
-      "rewards/rejected": -8.8125,
+      "logits/chosen": -8.375,
+      "logits/rejected": -7.40625,
+      "logps/chosen": -884.0,
+      "logps/rejected": -1272.0,
+      "loss": 0.1878,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -4.78125,
+      "rewards/margins": 3.25,
+      "rewards/rejected": -8.0,
       "step": 1320
     },
     {
       "epoch": 1.3926701570680629,
-      "grad_norm": 2.8981255120153295,
+      "grad_norm": 2.294618041082223,
       "learning_rate": 3.2382740934684696e-05,
-      "logits/chosen": -3.390625,
-      "logits/rejected": -2.234375,
-      "logps/chosen": -1024.0,
-      "logps/rejected": -1464.0,
-      "loss": 0.2091,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -6.25,
-      "rewards/margins": 3.578125,
-      "rewards/rejected": -9.8125,
+      "logits/chosen": -7.46875,
+      "logits/rejected": -6.53125,
+      "logps/chosen": -984.0,
+      "logps/rejected": -1424.0,
+      "loss": 0.2006,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -5.84375,
+      "rewards/margins": 3.625,
+      "rewards/rejected": -9.4375,
       "step": 1330
     },
     {
       "epoch": 1.4031413612565444,
-      "grad_norm": 2.1596527740451754,
+      "grad_norm": 3.1441128836415615,
       "learning_rate": 3.209113299222982e-05,
-      "logits/chosen": -1.0546875,
-      "logits/rejected": -0.9765625,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -1432.0,
-      "loss": 0.1772,
+      "logits/chosen": -6.125,
+      "logits/rejected": -5.875,
+      "logps/chosen": -964.0,
+      "logps/rejected": -1368.0,
+      "loss": 0.1694,
       "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -6.875,
-      "rewards/margins": 3.28125,
-      "rewards/rejected": -10.125,
+      "rewards/chosen": -6.15625,
+      "rewards/margins": 3.40625,
+      "rewards/rejected": -9.5625,
       "step": 1340
     },
     {
       "epoch": 1.4136125654450262,
-      "grad_norm": 2.5553340028780855,
+      "grad_norm": 2.4194545992019654,
       "learning_rate": 3.179847201145505e-05,
-      "logits/chosen": -1.96875,
-      "logits/rejected": -1.1484375,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -1480.0,
-      "loss": 0.176,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -6.53125,
-      "rewards/margins": 3.5,
-      "rewards/rejected": -10.0,
+      "logits/chosen": -7.21875,
+      "logits/rejected": -6.5,
+      "logps/chosen": -980.0,
+      "logps/rejected": -1384.0,
+      "loss": 0.1586,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -5.5,
+      "rewards/margins": 3.5625,
+      "rewards/rejected": -9.0625,
       "step": 1350
     },
     {
       "epoch": 1.4240837696335078,
-      "grad_norm": 2.2213766606851757,
+      "grad_norm": 2.0568237139401213,
       "learning_rate": 3.150480145272528e-05,
-      "logits/chosen": -2.78125,
-      "logits/rejected": -2.125,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -1448.0,
-      "loss": 0.1836,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -6.53125,
+      "logits/chosen": -7.4375,
+      "logits/rejected": -6.75,
+      "logps/chosen": -1000.0,
+      "logps/rejected": -1400.0,
+      "loss": 0.1897,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.0,
       "rewards/margins": 3.390625,
-      "rewards/rejected": -9.9375,
+      "rewards/rejected": -9.375,
       "step": 1360
     },
     {
       "epoch": 1.4240837696335078,
-      "eval_logits/chosen": -0.73828125,
-      "eval_logits/rejected": 1.3984375,
-      "eval_logps/chosen": -1064.0,
-      "eval_logps/rejected": -1120.0,
-      "eval_loss": 0.9651718735694885,
-      "eval_rewards/accuracies": 0.607421875,
-      "eval_rewards/chosen": -7.46875,
-      "eval_rewards/margins": 0.8359375,
-      "eval_rewards/rejected": -8.3125,
-      "eval_runtime": 46.5399,
-      "eval_samples_per_second": 42.974,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -5.6875,
+      "eval_logits/rejected": -4.28125,
+      "eval_logps/chosen": -1016.0,
+      "eval_logps/rejected": -1072.0,
+      "eval_loss": 0.9563124775886536,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -6.96875,
+      "eval_rewards/margins": 0.88671875,
+      "eval_rewards/rejected": -7.84375,
+      "eval_runtime": 58.3802,
+      "eval_samples_per_second": 34.258,
+      "eval_steps_per_second": 0.548,
       "step": 1360
     },
     {
       "epoch": 1.4345549738219896,
-      "grad_norm": 1.6470552755303483,
+      "grad_norm": 1.338259187688985,
       "learning_rate": 3.121016492632848e-05,
-      "logits/chosen": -4.15625,
-      "logits/rejected": -3.1875,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -1416.0,
-      "loss": 0.1949,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -6.34375,
-      "rewards/margins": 3.265625,
-      "rewards/rejected": -9.625,
+      "logits/chosen": -7.53125,
+      "logits/rejected": -6.75,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1424.0,
+      "loss": 0.1705,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.1875,
+      "rewards/margins": 3.46875,
+      "rewards/rejected": -9.6875,
       "step": 1370
     },
     {
       "epoch": 1.4450261780104712,
-      "grad_norm": 2.2338652713930505,
+      "grad_norm": 2.1517111469908725,
       "learning_rate": 3.091460618599951e-05,
-      "logits/chosen": -2.484375,
-      "logits/rejected": -1.4453125,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -1488.0,
-      "loss": 0.1744,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -6.3125,
-      "rewards/margins": 3.8125,
-      "rewards/rejected": -10.125,
+      "logits/chosen": -6.1875,
+      "logits/rejected": -5.3125,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -1512.0,
+      "loss": 0.1723,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -6.5,
+      "rewards/margins": 3.859375,
+      "rewards/rejected": -10.375,
       "step": 1380
     },
     {
       "epoch": 1.4554973821989527,
-      "grad_norm": 2.5245572636328357,
+      "grad_norm": 2.589356351516992,
       "learning_rate": 3.061816912242265e-05,
-      "logits/chosen": -2.8125,
-      "logits/rejected": -2.28125,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -1560.0,
-      "loss": 0.1848,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -7.03125,
-      "rewards/margins": 3.859375,
-      "rewards/rejected": -10.875,
+      "logits/chosen": -6.0625,
+      "logits/rejected": -5.6875,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -1552.0,
+      "loss": 0.1796,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.9375,
+      "rewards/margins": 3.890625,
+      "rewards/rejected": -10.8125,
       "step": 1390
     },
     {
       "epoch": 1.4659685863874345,
-      "grad_norm": 2.1094242222376574,
+      "grad_norm": 2.7055916329758047,
       "learning_rate": 3.0320897756713785e-05,
-      "logits/chosen": -2.140625,
-      "logits/rejected": -1.5234375,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -1544.0,
-      "loss": 0.1661,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -6.8125,
-      "rewards/margins": 3.53125,
-      "rewards/rejected": -10.3125,
+      "logits/chosen": -6.15625,
+      "logits/rejected": -5.65625,
+      "logps/chosen": -1032.0,
+      "logps/rejected": -1472.0,
+      "loss": 0.1552,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -6.0625,
+      "rewards/margins": 3.609375,
+      "rewards/rejected": -9.6875,
       "step": 1400
     },
     {
       "epoch": 1.4764397905759163,
-      "grad_norm": 1.5932733379830823,
+      "grad_norm": 1.8228637366032696,
       "learning_rate": 3.002283623388332e-05,
-      "logits/chosen": -1.53125,
-      "logits/rejected": -0.59375,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -1416.0,
-      "loss": 0.1884,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -6.65625,
-      "rewards/margins": 3.328125,
-      "rewards/rejected": -10.0,
+      "logits/chosen": -6.625,
+      "logits/rejected": -5.90625,
+      "logps/chosen": -936.0,
+      "logps/rejected": -1328.0,
+      "loss": 0.1877,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -5.625,
+      "rewards/margins": 3.4375,
+      "rewards/rejected": -9.0625,
       "step": 1410
     },
     {
       "epoch": 1.486910994764398,
-      "grad_norm": 3.5049985218975004,
+      "grad_norm": 7.458094011627475,
       "learning_rate": 2.9724028816280507e-05,
-      "logits/chosen": -1.4921875,
-      "logits/rejected": -1.2578125,
-      "logps/chosen": -1024.0,
-      "logps/rejected": -1424.0,
-      "loss": 0.2048,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -6.34375,
-      "rewards/margins": 3.375,
-      "rewards/rejected": -9.75,
+      "logits/chosen": -6.5625,
+      "logits/rejected": -6.28125,
+      "logps/chosen": -948.0,
+      "logps/rejected": -1344.0,
+      "loss": 0.1951,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -5.5625,
+      "rewards/margins": 3.359375,
+      "rewards/rejected": -8.9375,
       "step": 1420
     },
     {
       "epoch": 1.4973821989528795,
-      "grad_norm": 2.657944801696889,
+      "grad_norm": 3.0840732265902013,
       "learning_rate": 2.9424519877020524e-05,
-      "logits/chosen": -3.015625,
-      "logits/rejected": -2.21875,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -1544.0,
-      "loss": 0.166,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.84375,
-      "rewards/margins": 3.625,
-      "rewards/rejected": -10.4375,
+      "logits/chosen": -7.34375,
+      "logits/rejected": -6.625,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1488.0,
+      "loss": 0.1644,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -6.0625,
+      "rewards/margins": 3.75,
+      "rewards/rejected": -9.8125,
       "step": 1430
     },
     {
       "epoch": 1.5078534031413613,
-      "grad_norm": 2.743824080856472,
+      "grad_norm": 2.0154314863348888,
       "learning_rate": 2.912435389339504e-05,
-      "logits/chosen": -1.4453125,
-      "logits/rejected": 0.34375,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -1560.0,
-      "loss": 0.153,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.34375,
-      "rewards/margins": 3.953125,
-      "rewards/rejected": -11.3125,
+      "logits/chosen": -5.875,
+      "logits/rejected": -4.625,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -1480.0,
+      "loss": 0.1383,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -6.46875,
+      "rewards/margins": 4.0625,
+      "rewards/rejected": -10.5625,
       "step": 1440
     },
     {
       "epoch": 1.5078534031413613,
-      "eval_logits/chosen": 1.921875,
-      "eval_logits/rejected": 4.0,
-      "eval_logps/chosen": -1288.0,
-      "eval_logps/rejected": -1352.0,
-      "eval_loss": 1.1796562671661377,
-      "eval_rewards/accuracies": 0.59375,
-      "eval_rewards/chosen": -9.6875,
-      "eval_rewards/margins": 0.91015625,
-      "eval_rewards/rejected": -10.625,
-      "eval_runtime": 46.5384,
-      "eval_samples_per_second": 42.975,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -3.90625,
+      "eval_logits/rejected": -2.59375,
+      "eval_logps/chosen": -1176.0,
+      "eval_logps/rejected": -1240.0,
+      "eval_loss": 1.1195625066757202,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -8.5625,
+      "eval_rewards/margins": 0.9921875,
+      "eval_rewards/rejected": -9.5,
+      "eval_runtime": 58.3935,
+      "eval_samples_per_second": 34.25,
+      "eval_steps_per_second": 0.548,
       "step": 1440
     },
     {
       "epoch": 1.518324607329843,
-      "grad_norm": 3.278790504977696,
+      "grad_norm": 2.7703357473930295,
       "learning_rate": 2.8823575440267226e-05,
-      "logits/chosen": -0.51953125,
-      "logits/rejected": 0.359375,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1624.0,
-      "loss": 0.1776,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -7.9375,
-      "rewards/margins": 3.671875,
-      "rewards/rejected": -11.625,
+      "logits/chosen": -5.34375,
+      "logits/rejected": -4.71875,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -1536.0,
+      "loss": 0.1859,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -7.25,
+      "rewards/margins": 3.515625,
+      "rewards/rejected": -10.75,
       "step": 1450
     },
     {
       "epoch": 1.5287958115183247,
-      "grad_norm": 1.9412671583001393,
+      "grad_norm": 2.3144297007102073,
       "learning_rate": 2.8522229183452465e-05,
-      "logits/chosen": -0.2392578125,
-      "logits/rejected": 0.87890625,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1536.0,
-      "loss": 0.1986,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -7.6875,
-      "rewards/margins": 3.4375,
-      "rewards/rejected": -11.125,
+      "logits/chosen": -5.4375,
+      "logits/rejected": -4.5625,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -1448.0,
+      "loss": 0.2097,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.90625,
+      "rewards/margins": 3.40625,
+      "rewards/rejected": -10.3125,
       "step": 1460
     },
     {
       "epoch": 1.5392670157068062,
-      "grad_norm": 2.933864206225955,
+      "grad_norm": 2.4767618519892967,
       "learning_rate": 2.8220359873085373e-05,
-      "logits/chosen": -1.4375,
-      "logits/rejected": -1.1875,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -1416.0,
-      "loss": 0.1789,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -6.1875,
-      "rewards/margins": 3.671875,
-      "rewards/rejected": -9.875,
+      "logits/chosen": -6.03125,
+      "logits/rejected": -5.78125,
+      "logps/chosen": -976.0,
+      "logps/rejected": -1360.0,
+      "loss": 0.1669,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -5.65625,
+      "rewards/margins": 3.6875,
+      "rewards/rejected": -9.3125,
       "step": 1470
     },
     {
       "epoch": 1.5497382198952878,
-      "grad_norm": 2.581174000120746,
+      "grad_norm": 2.208371028450818,
       "learning_rate": 2.7918012336974388e-05,
-      "logits/chosen": -1.390625,
-      "logits/rejected": -1.359375,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -1456.0,
-      "loss": 0.1653,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -6.46875,
-      "rewards/margins": 3.734375,
-      "rewards/rejected": -10.1875,
+      "logits/chosen": -5.75,
+      "logits/rejected": -5.5,
+      "logps/chosen": -1024.0,
+      "logps/rejected": -1448.0,
+      "loss": 0.2199,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.21875,
+      "rewards/margins": 3.9375,
+      "rewards/rejected": -10.125,
       "step": 1480
     },
     {
       "epoch": 1.5602094240837696,
-      "grad_norm": 2.8994060608112786,
+      "grad_norm": 3.0235966451359317,
       "learning_rate": 2.7615231473944815e-05,
-      "logits/chosen": -0.08935546875,
-      "logits/rejected": 0.5625,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -1488.0,
-      "loss": 0.1668,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -6.96875,
-      "rewards/margins": 3.65625,
-      "rewards/rejected": -10.625,
+      "logits/chosen": -4.71875,
+      "logits/rejected": -4.21875,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -1472.0,
+      "loss": 0.1721,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -6.84375,
+      "rewards/margins": 3.546875,
+      "rewards/rejected": -10.375,
       "step": 1490
     },
     {
       "epoch": 1.5706806282722514,
-      "grad_norm": 2.332911962136739,
+      "grad_norm": 1.9341998638460187,
       "learning_rate": 2.7312062247171327e-05,
-      "logits/chosen": -0.48046875,
-      "logits/rejected": -0.14453125,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -1512.0,
-      "loss": 0.1892,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -7.09375,
-      "rewards/margins": 3.484375,
-      "rewards/rejected": -10.5625,
+      "logits/chosen": -5.75,
+      "logits/rejected": -5.28125,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1456.0,
+      "loss": 0.1798,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.5625,
+      "rewards/margins": 3.46875,
+      "rewards/rejected": -10.0625,
       "step": 1500
     },
     {
       "epoch": 1.581151832460733,
-      "grad_norm": 2.3163869046136787,
+      "grad_norm": 2.0144052805549464,
       "learning_rate": 2.7008549677500876e-05,
-      "logits/chosen": -1.5859375,
-      "logits/rejected": -0.70703125,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -1528.0,
-      "loss": 0.1715,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -6.8125,
-      "rewards/margins": 3.59375,
-      "rewards/rejected": -10.375,
+      "logits/chosen": -6.1875,
+      "logits/rejected": -5.5,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1488.0,
+      "loss": 0.1786,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.5625,
+      "rewards/margins": 3.421875,
+      "rewards/rejected": -10.0,
       "step": 1510
     },
     {
       "epoch": 1.5916230366492146,
-      "grad_norm": 2.5633932930876253,
+      "grad_norm": 2.4639219813948703,
       "learning_rate": 2.670473883676709e-05,
-      "logits/chosen": -0.37109375,
-      "logits/rejected": 0.44140625,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -1496.0,
-      "loss": 0.1398,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -6.71875,
+      "logits/chosen": -5.0625,
+      "logits/rejected": -4.40625,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1512.0,
+      "loss": 0.146,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.8125,
       "rewards/margins": 3.671875,
-      "rewards/rejected": -10.375,
+      "rewards/rejected": -10.5,
       "step": 1520
     },
     {
       "epoch": 1.5916230366492146,
-      "eval_logits/chosen": 2.15625,
-      "eval_logits/rejected": 4.1875,
-      "eval_logps/chosen": -1216.0,
-      "eval_logps/rejected": -1280.0,
-      "eval_loss": 1.0458593368530273,
-      "eval_rewards/accuracies": 0.599609375,
-      "eval_rewards/chosen": -9.0,
-      "eval_rewards/margins": 0.9609375,
-      "eval_rewards/rejected": -9.9375,
-      "eval_runtime": 46.5597,
-      "eval_samples_per_second": 42.956,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -3.03125,
+      "eval_logits/rejected": -1.6796875,
+      "eval_logps/chosen": -1264.0,
+      "eval_logps/rejected": -1336.0,
+      "eval_loss": 1.0766562223434448,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -9.5,
+      "eval_rewards/margins": 1.0078125,
+      "eval_rewards/rejected": -10.5,
+      "eval_runtime": 58.3775,
+      "eval_samples_per_second": 34.26,
+      "eval_steps_per_second": 0.548,
       "step": 1520
     },
     {
       "epoch": 1.6020942408376964,
-      "grad_norm": 2.6089032432465964,
+      "grad_norm": 2.6809441371028067,
       "learning_rate": 2.6400674841097074e-05,
-      "logits/chosen": -1.3359375,
-      "logits/rejected": -0.10595703125,
-      "logps/chosen": -1208.0,
+      "logits/chosen": -5.21875,
+      "logits/rejected": -4.34375,
+      "logps/chosen": -1216.0,
       "logps/rejected": -1624.0,
-      "loss": 0.1563,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -7.84375,
-      "rewards/margins": 3.78125,
+      "loss": 0.1651,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -7.90625,
+      "rewards/margins": 3.734375,
       "rewards/rejected": -11.625,
       "step": 1530
     },
     {
       "epoch": 1.6125654450261782,
-      "grad_norm": 2.0057571784102723,
+      "grad_norm": 2.560176807721602,
       "learning_rate": 2.609640284421158e-05,
-      "logits/chosen": -1.515625,
-      "logits/rejected": -1.0234375,
-      "logps/chosen": -1152.0,
-      "logps/rejected": -1608.0,
-      "loss": 0.1659,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.4375,
-      "rewards/margins": 3.796875,
-      "rewards/rejected": -11.25,
+      "logits/chosen": -5.3125,
+      "logits/rejected": -4.78125,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -1592.0,
+      "loss": 0.1713,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -7.28125,
+      "rewards/margins": 3.890625,
+      "rewards/rejected": -11.125,
       "step": 1540
     },
     {
       "epoch": 1.6230366492146597,
-      "grad_norm": 3.4498242396076084,
+      "grad_norm": 3.424368217114417,
       "learning_rate": 2.5791968030719726e-05,
-      "logits/chosen": -3.15625,
-      "logits/rejected": -1.65625,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -1488.0,
-      "loss": 0.1882,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -6.28125,
-      "rewards/margins": 3.65625,
-      "rewards/rejected": -9.9375,
+      "logits/chosen": -5.46875,
+      "logits/rejected": -4.625,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -1504.0,
+      "loss": 0.1828,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -6.46875,
+      "rewards/margins": 3.625,
+      "rewards/rejected": -10.0625,
       "step": 1550
     },
     {
       "epoch": 1.6335078534031413,
-      "grad_norm": 2.025237361481366,
+      "grad_norm": 1.8069783236744463,
       "learning_rate": 2.548741560940902e-05,
-      "logits/chosen": -3.0625,
-      "logits/rejected": -2.15625,
-      "logps/chosen": -984.0,
-      "logps/rejected": -1360.0,
-      "loss": 0.1862,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -5.875,
-      "rewards/margins": 3.328125,
-      "rewards/rejected": -9.1875,
+      "logits/chosen": -5.75,
+      "logits/rejected": -5.0625,
+      "logps/chosen": -976.0,
+      "logps/rejected": -1376.0,
+      "loss": 0.1847,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.78125,
+      "rewards/margins": 3.5625,
+      "rewards/rejected": -9.3125,
       "step": 1560
     },
     {
       "epoch": 1.6439790575916229,
-      "grad_norm": 2.5619218621897963,
+      "grad_norm": 3.01788054883275,
       "learning_rate": 2.5182790806531787e-05,
-      "logits/chosen": -1.3828125,
-      "logits/rejected": -0.80859375,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -1448.0,
-      "loss": 0.1865,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -6.5625,
-      "rewards/margins": 3.359375,
-      "rewards/rejected": -9.9375,
+      "logits/chosen": -5.34375,
+      "logits/rejected": -4.875,
+      "logps/chosen": -996.0,
+      "logps/rejected": -1384.0,
+      "loss": 0.1872,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -5.8125,
+      "rewards/margins": 3.5,
+      "rewards/rejected": -9.3125,
       "step": 1570
     },
     {
       "epoch": 1.6544502617801047,
-      "grad_norm": 2.5739097608736405,
+      "grad_norm": 4.440471116458592,
       "learning_rate": 2.4878138859089074e-05,
-      "logits/chosen": -1.875,
-      "logits/rejected": -0.78515625,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -1456.0,
-      "loss": 0.1429,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.6875,
-      "rewards/margins": 3.59375,
-      "rewards/rejected": -10.25,
+      "logits/chosen": -5.8125,
+      "logits/rejected": -4.875,
+      "logps/chosen": -988.0,
+      "logps/rejected": -1376.0,
+      "loss": 0.1514,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -5.84375,
+      "rewards/margins": 3.640625,
+      "rewards/rejected": -9.5,
       "step": 1580
     },
     {
       "epoch": 1.6649214659685865,
-      "grad_norm": 1.886698354769222,
+      "grad_norm": 2.353567274517742,
       "learning_rate": 2.4573505008112924e-05,
-      "logits/chosen": -3.1875,
-      "logits/rejected": -1.9921875,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -1584.0,
-      "loss": 0.1867,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -7.25,
-      "rewards/margins": 3.84375,
-      "rewards/rejected": -11.0625,
+      "logits/chosen": -6.15625,
+      "logits/rejected": -5.21875,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -1536.0,
+      "loss": 0.193,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 3.859375,
+      "rewards/rejected": -10.6875,
       "step": 1590
     },
     {
       "epoch": 1.675392670157068,
-      "grad_norm": 2.300067683322435,
+      "grad_norm": 2.150090769099789,
       "learning_rate": 2.4268934491948032e-05,
-      "logits/chosen": -2.90625,
-      "logits/rejected": -1.8125,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -1616.0,
-      "loss": 0.1634,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -7.5625,
-      "rewards/margins": 3.6875,
-      "rewards/rejected": -11.25,
+      "logits/chosen": -5.59375,
+      "logits/rejected": -5.03125,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -1536.0,
+      "loss": 0.1831,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -7.0,
+      "rewards/margins": 3.421875,
+      "rewards/rejected": -10.4375,
       "step": 1600
     },
     {
       "epoch": 1.675392670157068,
-      "eval_logits/chosen": 0.1005859375,
-      "eval_logits/rejected": 2.40625,
-      "eval_logps/chosen": -1224.0,
-      "eval_logps/rejected": -1288.0,
-      "eval_loss": 1.0130780935287476,
-      "eval_rewards/accuracies": 0.59765625,
-      "eval_rewards/chosen": -9.0625,
-      "eval_rewards/margins": 0.9375,
-      "eval_rewards/rejected": -10.0,
-      "eval_runtime": 46.5425,
-      "eval_samples_per_second": 42.971,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -3.59375,
+      "eval_logits/rejected": -2.234375,
+      "eval_logps/chosen": -1128.0,
+      "eval_logps/rejected": -1184.0,
+      "eval_loss": 0.9775624871253967,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -8.0625,
+      "eval_rewards/margins": 0.8515625,
+      "eval_rewards/rejected": -8.9375,
+      "eval_runtime": 58.3335,
+      "eval_samples_per_second": 34.286,
+      "eval_steps_per_second": 0.549,
       "step": 1600
     },
     {
       "epoch": 1.6858638743455496,
-      "grad_norm": 1.914030930549274,
+      "grad_norm": 1.6767452703337336,
       "learning_rate": 2.396447253953385e-05,
-      "logits/chosen": -4.21875,
-      "logits/rejected": -3.25,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1600.0,
-      "loss": 0.1611,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.53125,
-      "rewards/margins": 3.625,
-      "rewards/rejected": -11.1875,
+      "logits/chosen": -6.28125,
+      "logits/rejected": -5.75,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -1504.0,
+      "loss": 0.1527,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.5625,
+      "rewards/margins": 3.65625,
+      "rewards/rejected": -10.25,
       "step": 1610
     },
     {
       "epoch": 1.6963350785340314,
-      "grad_norm": 2.558816044799574,
+      "grad_norm": 2.8728288660119934,
       "learning_rate": 2.3660164363687998e-05,
-      "logits/chosen": -4.15625,
-      "logits/rejected": -3.296875,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1616.0,
-      "loss": 0.1638,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.34375,
-      "rewards/margins": 3.703125,
-      "rewards/rejected": -11.0625,
+      "logits/chosen": -5.96875,
+      "logits/rejected": -5.5625,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -1544.0,
+      "loss": 0.1708,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.71875,
+      "rewards/margins": 3.65625,
+      "rewards/rejected": -10.375,
       "step": 1620
     },
     {
       "epoch": 1.7068062827225132,
-      "grad_norm": 3.3794766792060593,
+      "grad_norm": 2.9340571429381237,
       "learning_rate": 2.33560551543922e-05,
-      "logits/chosen": -3.3125,
-      "logits/rejected": -2.421875,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -1560.0,
-      "loss": 0.1789,
+      "logits/chosen": -6.4375,
+      "logits/rejected": -5.84375,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -1432.0,
+      "loss": 0.1817,
       "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -7.3125,
-      "rewards/margins": 3.609375,
-      "rewards/rejected": -10.9375,
+      "rewards/chosen": -6.15625,
+      "rewards/margins": 3.46875,
+      "rewards/rejected": -9.625,
       "step": 1630
     },
     {
       "epoch": 1.7172774869109948,
-      "grad_norm": 2.57951933161477,
+      "grad_norm": 2.5142070466008954,
       "learning_rate": 2.3052190072081493e-05,
-      "logits/chosen": -4.0625,
-      "logits/rejected": -2.59375,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -1496.0,
-      "loss": 0.1831,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -6.90625,
-      "rewards/margins": 3.546875,
-      "rewards/rejected": -10.4375,
+      "logits/chosen": -6.3125,
+      "logits/rejected": -5.3125,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1440.0,
+      "loss": 0.1886,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.25,
+      "rewards/margins": 3.6875,
+      "rewards/rejected": -9.9375,
       "step": 1640
     },
     {
       "epoch": 1.7277486910994764,
-      "grad_norm": 2.301766751101889,
+      "grad_norm": 2.6510280338696033,
       "learning_rate": 2.2748614240937864e-05,
-      "logits/chosen": -3.53125,
-      "logits/rejected": -2.921875,
-      "logps/chosen": -1020.0,
-      "logps/rejected": -1416.0,
-      "loss": 0.1645,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -6.3125,
-      "rewards/margins": 3.421875,
-      "rewards/rejected": -9.75,
+      "logits/chosen": -5.53125,
+      "logits/rejected": -5.15625,
+      "logps/chosen": -1012.0,
+      "logps/rejected": -1400.0,
+      "loss": 0.1688,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -6.28125,
+      "rewards/margins": 3.296875,
+      "rewards/rejected": -9.5625,
       "step": 1650
     },
     {
       "epoch": 1.738219895287958,
-      "grad_norm": 2.144668727248553,
+      "grad_norm": 1.6658800737883042,
       "learning_rate": 2.2445372742189334e-05,
-      "logits/chosen": -2.890625,
-      "logits/rejected": -2.0625,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -1632.0,
-      "loss": 0.1458,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.375,
-      "rewards/margins": 3.84375,
-      "rewards/rejected": -11.1875,
+      "logits/chosen": -5.1875,
+      "logits/rejected": -4.53125,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -1608.0,
+      "loss": 0.146,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -7.0625,
+      "rewards/margins": 3.890625,
+      "rewards/rejected": -10.9375,
       "step": 1660
     },
     {
       "epoch": 1.7486910994764397,
-      "grad_norm": 2.9442423780596894,
+      "grad_norm": 2.9766335974687412,
       "learning_rate": 2.214251060741528e-05,
-      "logits/chosen": -1.640625,
-      "logits/rejected": -0.82421875,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -1808.0,
-      "loss": 0.1622,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.125,
-      "rewards/margins": 4.28125,
-      "rewards/rejected": -12.4375,
+      "logits/chosen": -5.4375,
+      "logits/rejected": -4.65625,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -1688.0,
+      "loss": 0.1825,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.9375,
+      "rewards/margins": 4.21875,
+      "rewards/rejected": -11.1875,
       "step": 1670
     },
     {
       "epoch": 1.7591623036649215,
-      "grad_norm": 4.185881575644846,
+      "grad_norm": 3.3717185070255984,
       "learning_rate": 2.1840072811859302e-05,
-      "logits/chosen": -1.734375,
-      "logits/rejected": -0.7734375,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -1576.0,
-      "loss": 0.1777,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.5,
-      "rewards/margins": 3.765625,
-      "rewards/rejected": -11.25,
+      "logits/chosen": -5.53125,
+      "logits/rejected": -4.8125,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1424.0,
+      "loss": 0.1667,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -6.03125,
+      "rewards/margins": 3.703125,
+      "rewards/rejected": -9.75,
       "step": 1680
     },
     {
       "epoch": 1.7591623036649215,
-      "eval_logits/chosen": 1.1875,
-      "eval_logits/rejected": 3.53125,
-      "eval_logps/chosen": -1192.0,
-      "eval_logps/rejected": -1248.0,
-      "eval_loss": 1.0902031660079956,
-      "eval_rewards/accuracies": 0.625,
-      "eval_rewards/chosen": -8.6875,
-      "eval_rewards/margins": 0.921875,
-      "eval_rewards/rejected": -9.625,
-      "eval_runtime": 46.5456,
-      "eval_samples_per_second": 42.969,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -3.234375,
+      "eval_logits/rejected": -1.734375,
+      "eval_logps/chosen": -1088.0,
+      "eval_logps/rejected": -1152.0,
+      "eval_loss": 1.0210468769073486,
+      "eval_rewards/accuracies": 0.595703125,
+      "eval_rewards/chosen": -7.75,
+      "eval_rewards/margins": 0.90234375,
+      "eval_rewards/rejected": -8.625,
+      "eval_runtime": 58.3114,
+      "eval_samples_per_second": 34.299,
+      "eval_steps_per_second": 0.549,
       "step": 1680
     },
     {
       "epoch": 1.7696335078534031,
-      "grad_norm": 2.1964272411669428,
+      "grad_norm": 1.9499355522968773,
       "learning_rate": 2.1538104267750287e-05,
-      "logits/chosen": -0.69921875,
-      "logits/rejected": -0.16015625,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -1536.0,
-      "loss": 0.163,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.0625,
-      "rewards/margins": 3.703125,
-      "rewards/rejected": -10.75,
+      "logits/chosen": -4.28125,
+      "logits/rejected": -3.890625,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1464.0,
+      "loss": 0.154,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.4375,
+      "rewards/margins": 3.578125,
+      "rewards/rejected": -10.0,
       "step": 1690
     },
     {
       "epoch": 1.7801047120418847,
-      "grad_norm": 2.4817498044729036,
+      "grad_norm": 1.934436251327707,
       "learning_rate": 2.1236649817632952e-05,
-      "logits/chosen": -1.96875,
-      "logits/rejected": -1.7421875,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -1472.0,
-      "loss": 0.1436,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -6.6875,
-      "rewards/margins": 3.515625,
-      "rewards/rejected": -10.1875,
+      "logits/chosen": -4.6875,
+      "logits/rejected": -4.5,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1480.0,
+      "loss": 0.1533,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -6.71875,
+      "rewards/margins": 3.546875,
+      "rewards/rejected": -10.25,
       "step": 1700
     },
     {
       "epoch": 1.7905759162303665,
-      "grad_norm": 1.710320464091893,
+      "grad_norm": 1.873560519381576,
       "learning_rate": 2.093575422770872e-05,
-      "logits/chosen": -1.5234375,
-      "logits/rejected": -0.37890625,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -1424.0,
-      "loss": 0.1472,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -6.65625,
-      "rewards/margins": 3.65625,
-      "rewards/rejected": -10.3125,
+      "logits/chosen": -4.53125,
+      "logits/rejected": -3.78125,
+      "logps/chosen": -1000.0,
+      "logps/rejected": -1400.0,
+      "loss": 0.1439,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.25,
+      "rewards/margins": 3.84375,
+      "rewards/rejected": -10.125,
       "step": 1710
     },
     {
       "epoch": 1.8010471204188483,
-      "grad_norm": 3.6075834713291406,
+      "grad_norm": 3.84391842729347,
       "learning_rate": 2.0635462181187827e-05,
-      "logits/chosen": -2.734375,
-      "logits/rejected": -1.78125,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -1640.0,
-      "loss": 0.1683,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -7.53125,
-      "rewards/margins": 4.03125,
-      "rewards/rejected": -11.5625,
+      "logits/chosen": -5.375,
+      "logits/rejected": -4.65625,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -1584.0,
+      "loss": 0.1617,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -7.09375,
+      "rewards/margins": 3.953125,
+      "rewards/rejected": -11.0625,
       "step": 1720
     },
     {
       "epoch": 1.8115183246073299,
-      "grad_norm": 2.2216462813633084,
+      "grad_norm": 2.0619169655199223,
       "learning_rate": 2.0335818271653917e-05,
-      "logits/chosen": -2.546875,
-      "logits/rejected": -1.96875,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -1472.0,
-      "loss": 0.1518,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -6.59375,
+      "logits/chosen": -5.21875,
+      "logits/rejected": -4.65625,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1440.0,
+      "loss": 0.155,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -6.3125,
       "rewards/margins": 3.53125,
-      "rewards/rejected": -10.125,
+      "rewards/rejected": -9.875,
       "step": 1730
     },
     {
       "epoch": 1.8219895287958114,
-      "grad_norm": 3.00625657088484,
+      "grad_norm": 2.368898261919784,
       "learning_rate": 2.0036866996441815e-05,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -2.015625,
-      "logps/chosen": -1064.0,
+      "logits/chosen": -5.59375,
+      "logits/rejected": -4.5625,
+      "logps/chosen": -1056.0,
       "logps/rejected": -1448.0,
-      "loss": 0.1593,
+      "loss": 0.1438,
       "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.5625,
-      "rewards/margins": 3.8125,
-      "rewards/rejected": -10.375,
+      "rewards/chosen": -6.53125,
+      "rewards/margins": 3.921875,
+      "rewards/rejected": -10.4375,
       "step": 1740
     },
     {
       "epoch": 1.8324607329842932,
-      "grad_norm": 2.430313566922212,
+      "grad_norm": 1.725992484978568,
       "learning_rate": 1.9738652750029677e-05,
-      "logits/chosen": -2.3125,
-      "logits/rejected": -1.09375,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -1520.0,
-      "loss": 0.1583,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -6.875,
-      "rewards/margins": 3.8125,
+      "logits/chosen": -4.875,
+      "logits/rejected": -3.984375,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1512.0,
+      "loss": 0.1564,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 3.859375,
       "rewards/rejected": -10.6875,
       "step": 1750
     },
     {
       "epoch": 1.8429319371727748,
-      "grad_norm": 2.9259915595918375,
+      "grad_norm": 3.397622908776909,
       "learning_rate": 1.944121981744631e-05,
-      "logits/chosen": -2.859375,
-      "logits/rejected": -1.546875,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -1648.0,
-      "loss": 0.1536,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.40625,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -4.5,
+      "logps/chosen": -1160.0,
+      "logps/rejected": -1608.0,
+      "loss": 0.1514,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -7.0,
       "rewards/margins": 4.03125,
-      "rewards/rejected": -11.4375,
+      "rewards/rejected": -11.0625,
       "step": 1760
     },
     {
       "epoch": 1.8429319371727748,
-      "eval_logits/chosen": 1.6796875,
-      "eval_logits/rejected": 3.828125,
-      "eval_logps/chosen": -1232.0,
-      "eval_logps/rejected": -1296.0,
-      "eval_loss": 1.0046093463897705,
-      "eval_rewards/accuracies": 0.607421875,
-      "eval_rewards/chosen": -9.125,
-      "eval_rewards/margins": 0.91796875,
-      "eval_rewards/rejected": -10.0625,
-      "eval_runtime": 46.5361,
-      "eval_samples_per_second": 42.977,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -2.53125,
+      "eval_logits/rejected": -1.171875,
+      "eval_logps/chosen": -1184.0,
+      "eval_logps/rejected": -1256.0,
+      "eval_loss": 1.0214062929153442,
+      "eval_rewards/accuracies": 0.61328125,
+      "eval_rewards/chosen": -8.6875,
+      "eval_rewards/margins": 0.98046875,
+      "eval_rewards/rejected": -9.6875,
+      "eval_runtime": 58.3645,
+      "eval_samples_per_second": 34.267,
+      "eval_steps_per_second": 0.548,
       "step": 1760
     },
     {
       "epoch": 1.8534031413612566,
-      "grad_norm": 1.9069888597534437,
+      "grad_norm": 2.0852633231840647,
       "learning_rate": 1.9144612367694887e-05,
-      "logits/chosen": -1.78125,
-      "logits/rejected": -0.8125,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1552.0,
-      "loss": 0.1556,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.0,
-      "rewards/margins": 3.875,
-      "rewards/rejected": -10.875,
+      "logits/chosen": -4.71875,
+      "logits/rejected": -4.0625,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -1520.0,
+      "loss": 0.1617,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.75,
+      "rewards/margins": 3.78125,
+      "rewards/rejected": -10.5,
       "step": 1770
     },
     {
       "epoch": 1.8638743455497382,
-      "grad_norm": 3.027359130683597,
+      "grad_norm": 2.8843354889585244,
       "learning_rate": 1.8848874447193803e-05,
-      "logits/chosen": -0.6953125,
-      "logits/rejected": -0.0380859375,
-      "logps/chosen": -1120.0,
-      "logps/rejected": -1536.0,
-      "loss": 0.1705,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -7.25,
-      "rewards/margins": 3.546875,
-      "rewards/rejected": -10.8125,
+      "logits/chosen": -4.375,
+      "logits/rejected": -3.921875,
+      "logps/chosen": -1064.0,
+      "logps/rejected": -1472.0,
+      "loss": 0.165,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -6.6875,
+      "rewards/margins": 3.515625,
+      "rewards/rejected": -10.1875,
       "step": 1780
     },
     {
       "epoch": 1.8743455497382198,
-      "grad_norm": 2.399025253575211,
+      "grad_norm": 1.9641769529809405,
       "learning_rate": 1.855404997323571e-05,
-      "logits/chosen": -2.3125,
-      "logits/rejected": -1.2734375,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -1592.0,
-      "loss": 0.1839,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -7.46875,
-      "rewards/margins": 3.578125,
-      "rewards/rejected": -11.0625,
+      "logits/chosen": -5.53125,
+      "logits/rejected": -4.90625,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1520.0,
+      "loss": 0.1758,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -6.90625,
+      "rewards/margins": 3.421875,
+      "rewards/rejected": -10.3125,
       "step": 1790
     },
     {
       "epoch": 1.8848167539267016,
-      "grad_norm": 2.018568472273811,
+      "grad_norm": 1.3319542802482134,
       "learning_rate": 1.82601827274658e-05,
-      "logits/chosen": -2.5,
-      "logits/rejected": -1.578125,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -1488.0,
-      "loss": 0.1506,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -6.8125,
-      "rewards/margins": 3.5625,
-      "rewards/rejected": -10.375,
+      "logits/chosen": -5.75,
+      "logits/rejected": -5.09375,
+      "logps/chosen": -1008.0,
+      "logps/rejected": -1448.0,
+      "loss": 0.1495,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.1875,
+      "rewards/margins": 3.75,
+      "rewards/rejected": -9.9375,
       "step": 1800
     },
     {
       "epoch": 1.8952879581151834,
-      "grad_norm": 3.379581309581174,
+      "grad_norm": 3.0211489854534728,
       "learning_rate": 1.7967316349380213e-05,
-      "logits/chosen": -1.9765625,
-      "logits/rejected": -1.0625,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -1472.0,
-      "loss": 0.1575,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -6.90625,
-      "rewards/margins": 3.625,
-      "rewards/rejected": -10.5,
+      "logits/chosen": -5.0625,
+      "logits/rejected": -4.4375,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1448.0,
+      "loss": 0.1556,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.625,
+      "rewards/margins": 3.65625,
+      "rewards/rejected": -10.3125,
       "step": 1810
     },
     {
       "epoch": 1.905759162303665,
-      "grad_norm": 2.6962837037532195,
+      "grad_norm": 2.1653308803353766,
       "learning_rate": 1.7675494329845516e-05,
-      "logits/chosen": -0.859375,
-      "logits/rejected": 0.173828125,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1496.0,
-      "loss": 0.152,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -7.0,
-      "rewards/margins": 3.640625,
-      "rewards/rejected": -10.625,
+      "logits/chosen": -3.953125,
+      "logits/rejected": -3.296875,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -1552.0,
+      "loss": 0.1563,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -7.59375,
+      "rewards/margins": 3.625,
+      "rewards/rejected": -11.25,
       "step": 1820
     },
     {
       "epoch": 1.9162303664921465,
-      "grad_norm": 2.3013242821310103,
+      "grad_norm": 2.196424496809726,
       "learning_rate": 1.738476000464026e-05,
-      "logits/chosen": -0.486328125,
-      "logits/rejected": 0.6640625,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -1552.0,
-      "loss": 0.1458,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -6.84375,
+      "logits/chosen": -3.921875,
+      "logits/rejected": -3.375,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -1592.0,
+      "loss": 0.1598,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.21875,
       "rewards/margins": 4.25,
-      "rewards/rejected": -11.125,
+      "rewards/rejected": -11.5,
       "step": 1830
     },
     {
       "epoch": 1.9267015706806283,
-      "grad_norm": 3.3533598923784735,
+      "grad_norm": 2.86175715506911,
       "learning_rate": 1.7095156548019646e-05,
-      "logits/chosen": -0.15625,
-      "logits/rejected": 0.345703125,
-      "logps/chosen": -1120.0,
-      "logps/rejected": -1608.0,
-      "loss": 0.1713,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.3125,
-      "rewards/margins": 4.15625,
-      "rewards/rejected": -11.5,
+      "logits/chosen": -4.5625,
+      "logits/rejected": -4.125,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -1592.0,
+      "loss": 0.1594,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -7.25,
+      "rewards/margins": 4.0625,
+      "rewards/rejected": -11.3125,
       "step": 1840
     },
     {
       "epoch": 1.9267015706806283,
-      "eval_logits/chosen": 3.078125,
-      "eval_logits/rejected": 5.28125,
-      "eval_logps/chosen": -1248.0,
-      "eval_logps/rejected": -1328.0,
-      "eval_loss": 1.0627187490463257,
-      "eval_rewards/accuracies": 0.623046875,
-      "eval_rewards/chosen": -9.3125,
-      "eval_rewards/margins": 1.078125,
-      "eval_rewards/rejected": -10.375,
-      "eval_runtime": 46.5457,
-      "eval_samples_per_second": 42.969,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -2.625,
+      "eval_logits/rejected": -1.234375,
+      "eval_logps/chosen": -1200.0,
+      "eval_logps/rejected": -1264.0,
+      "eval_loss": 1.0633437633514404,
+      "eval_rewards/accuracies": 0.59765625,
+      "eval_rewards/chosen": -8.8125,
+      "eval_rewards/margins": 0.97265625,
+      "eval_rewards/rejected": -9.75,
+      "eval_runtime": 58.3667,
+      "eval_samples_per_second": 34.266,
+      "eval_steps_per_second": 0.548,
       "step": 1840
     },
     {
       "epoch": 1.93717277486911,
-      "grad_norm": 2.7188459999623036,
+      "grad_norm": 2.470536022896026,
       "learning_rate": 1.6806726966304064e-05,
-      "logits/chosen": -0.84765625,
-      "logits/rejected": 0.310546875,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1640.0,
-      "loss": 0.1938,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -7.5,
-      "rewards/margins": 4.09375,
-      "rewards/rejected": -11.625,
+      "logits/chosen": -4.90625,
+      "logits/rejected": -4.21875,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -1576.0,
+      "loss": 0.1934,
+      "rewards/accuracies": 0.8687499761581421,
+      "rewards/chosen": -7.1875,
+      "rewards/margins": 3.796875,
+      "rewards/rejected": -11.0,
       "step": 1850
     },
     {
       "epoch": 1.9476439790575917,
-      "grad_norm": 1.5551903506216613,
+      "grad_norm": 1.9737307134594215,
       "learning_rate": 1.6519514091492625e-05,
-      "logits/chosen": -1.4765625,
-      "logits/rejected": -0.400390625,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -1456.0,
-      "loss": 0.193,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -6.78125,
-      "rewards/margins": 3.546875,
-      "rewards/rejected": -10.375,
+      "logits/chosen": -5.34375,
+      "logits/rejected": -4.5625,
+      "logps/chosen": -988.0,
+      "logps/rejected": -1368.0,
+      "loss": 0.1874,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -6.09375,
+      "rewards/margins": 3.34375,
+      "rewards/rejected": -9.4375,
       "step": 1860
     },
     {
       "epoch": 1.9581151832460733,
-      "grad_norm": 1.8179149995134052,
+      "grad_norm": 1.6325115090387947,
       "learning_rate": 1.6233560574902627e-05,
-      "logits/chosen": -2.1875,
-      "logits/rejected": -1.3671875,
-      "logps/chosen": -1000.0,
-      "logps/rejected": -1368.0,
-      "loss": 0.1724,
+      "logits/chosen": -5.5625,
+      "logits/rejected": -4.96875,
+      "logps/chosen": -928.0,
+      "logps/rejected": -1304.0,
+      "loss": 0.1795,
       "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.375,
-      "rewards/margins": 3.453125,
-      "rewards/rejected": -9.8125,
+      "rewards/chosen": -5.625,
+      "rewards/margins": 3.5,
+      "rewards/rejected": -9.125,
       "step": 1870
     },
     {
       "epoch": 1.9685863874345548,
-      "grad_norm": 2.91158247694518,
+      "grad_norm": 2.443253992106041,
       "learning_rate": 1.5948908880835753e-05,
-      "logits/chosen": -2.375,
-      "logits/rejected": -1.625,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -1472.0,
-      "loss": 0.1746,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.53125,
-      "rewards/margins": 3.75,
-      "rewards/rejected": -10.25,
+      "logits/chosen": -5.25,
+      "logits/rejected": -4.78125,
+      "logps/chosen": -1016.0,
+      "logps/rejected": -1432.0,
+      "loss": 0.1665,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.125,
+      "rewards/margins": 3.765625,
+      "rewards/rejected": -9.875,
       "step": 1880
     },
     {
       "epoch": 1.9790575916230366,
-      "grad_norm": 2.8480567404966797,
+      "grad_norm": 2.1290088740383073,
       "learning_rate": 1.5665601280272124e-05,
-      "logits/chosen": -1.9765625,
-      "logits/rejected": -0.2734375,
+      "logits/chosen": -4.90625,
+      "logits/rejected": -3.859375,
       "logps/chosen": -1064.0,
-      "logps/rejected": -1472.0,
-      "loss": 0.1738,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -6.8125,
-      "rewards/margins": 3.75,
-      "rewards/rejected": -10.5625,
+      "logps/rejected": -1480.0,
+      "loss": 0.1632,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.75,
+      "rewards/margins": 3.875,
+      "rewards/rejected": -10.625,
       "step": 1890
     },
     {
       "epoch": 1.9895287958115184,
-      "grad_norm": 1.3661340738977774,
+      "grad_norm": 1.8410620375092406,
       "learning_rate": 1.538367984459298e-05,
-      "logits/chosen": -0.63671875,
-      "logits/rejected": 0.2578125,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -1488.0,
-      "loss": 0.1291,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -6.5,
-      "rewards/margins": 3.875,
-      "rewards/rejected": -10.375,
+      "logits/chosen": -4.09375,
+      "logits/rejected": -3.4375,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -1504.0,
+      "loss": 0.1429,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.75,
+      "rewards/margins": 3.8125,
+      "rewards/rejected": -10.5625,
       "step": 1900
     },
     {
       "epoch": 2.0,
-      "grad_norm": 2.5130221055959563,
+      "grad_norm": 2.0869587456660628,
       "learning_rate": 1.5103186439333133e-05,
-      "logits/chosen": -0.73046875,
-      "logits/rejected": -0.173828125,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -1512.0,
-      "loss": 0.1425,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.625,
-      "rewards/margins": 3.953125,
-      "rewards/rejected": -10.5625,
+      "logits/chosen": -5.0625,
+      "logits/rejected": -4.5625,
+      "logps/chosen": -988.0,
+      "logps/rejected": -1424.0,
+      "loss": 0.166,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -6.03125,
+      "rewards/margins": 3.703125,
+      "rewards/rejected": -9.75,
       "step": 1910
     },
     {
       "epoch": 2.0104712041884816,
-      "grad_norm": 0.505945540708177,
+      "grad_norm": 0.5304486601237802,
       "learning_rate": 1.4824162717963829e-05,
-      "logits/chosen": -1.703125,
-      "logits/rejected": 1.3203125,
-      "logps/chosen": -1000.0,
-      "logps/rejected": -1624.0,
-      "loss": 0.0282,
+      "logits/chosen": -5.78125,
+      "logits/rejected": -3.5625,
+      "logps/chosen": -944.0,
+      "logps/rejected": -1552.0,
+      "loss": 0.0307,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -6.0625,
-      "rewards/margins": 5.6875,
-      "rewards/rejected": -11.75,
+      "rewards/chosen": -5.53125,
+      "rewards/margins": 5.5,
+      "rewards/rejected": -11.0,
       "step": 1920
     },
     {
       "epoch": 2.0104712041884816,
-      "eval_logits/chosen": 1.828125,
-      "eval_logits/rejected": 4.125,
-      "eval_logps/chosen": -1224.0,
-      "eval_logps/rejected": -1304.0,
-      "eval_loss": 1.0637500286102295,
-      "eval_rewards/accuracies": 0.61328125,
-      "eval_rewards/chosen": -9.0625,
-      "eval_rewards/margins": 1.0625,
-      "eval_rewards/rejected": -10.125,
-      "eval_runtime": 46.551,
-      "eval_samples_per_second": 42.964,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -2.984375,
+      "eval_logits/rejected": -1.453125,
+      "eval_logps/chosen": -1192.0,
+      "eval_logps/rejected": -1264.0,
+      "eval_loss": 1.0948125123977661,
+      "eval_rewards/accuracies": 0.6171875,
+      "eval_rewards/chosen": -8.75,
+      "eval_rewards/margins": 1.03125,
+      "eval_rewards/rejected": -9.75,
+      "eval_runtime": 58.3906,
+      "eval_samples_per_second": 34.252,
+      "eval_steps_per_second": 0.548,
       "step": 1920
     },
     {
       "epoch": 2.020942408376963,
-      "grad_norm": 0.6827442694941761,
+      "grad_norm": 0.9795540969367579,
       "learning_rate": 1.4546650115707192e-05,
-      "logits/chosen": -1.34375,
-      "logits/rejected": 0.44921875,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -1744.0,
-      "loss": 0.0245,
+      "logits/chosen": -4.59375,
+      "logits/rejected": -3.25,
+      "logps/chosen": -1080.0,
+      "logps/rejected": -1736.0,
+      "loss": 0.024,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.125,
-      "rewards/margins": 6.25,
-      "rewards/rejected": -13.375,
+      "rewards/chosen": -7.09375,
+      "rewards/margins": 6.21875,
+      "rewards/rejected": -13.3125,
       "step": 1930
     },
     {
       "epoch": 2.031413612565445,
-      "grad_norm": 0.28323703950912593,
+      "grad_norm": 0.19293286252964736,
       "learning_rate": 1.4270689843383111e-05,
-      "logits/chosen": 1.03125,
-      "logits/rejected": 2.1875,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -1928.0,
-      "loss": 0.0294,
+      "logits/chosen": -3.046875,
+      "logits/rejected": -2.03125,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -1960.0,
+      "loss": 0.0286,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 6.6875,
-      "rewards/rejected": -14.9375,
+      "rewards/chosen": -8.4375,
+      "rewards/margins": 6.875,
+      "rewards/rejected": -15.25,
       "step": 1940
     },
     {
       "epoch": 2.0418848167539267,
-      "grad_norm": 0.533120184877145,
+      "grad_norm": 0.4495019098497561,
       "learning_rate": 1.3996322881289347e-05,
-      "logits/chosen": -0.326171875,
-      "logits/rejected": 1.7265625,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -1944.0,
-      "loss": 0.0299,
+      "logits/chosen": -4.125,
+      "logits/rejected": -2.515625,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -1968.0,
+      "loss": 0.0289,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -8.625,
-      "rewards/margins": 6.625,
-      "rewards/rejected": -15.25,
+      "rewards/chosen": -8.75,
+      "rewards/margins": 6.71875,
+      "rewards/rejected": -15.5,
       "step": 1950
     },
     {
       "epoch": 2.0523560209424083,
-      "grad_norm": 0.6931813313556514,
+      "grad_norm": 0.5334073353991419,
       "learning_rate": 1.3723589973115963e-05,
-      "logits/chosen": 0.248046875,
-      "logits/rejected": 1.5,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -1984.0,
-      "loss": 0.0302,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -8.1875,
-      "rewards/margins": 7.0,
-      "rewards/rejected": -15.25,
+      "logits/chosen": -3.703125,
+      "logits/rejected": -2.78125,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -1976.0,
+      "loss": 0.0322,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -8.375,
+      "rewards/margins": 6.8125,
+      "rewards/rejected": -15.1875,
       "step": 1960
     },
     {
       "epoch": 2.06282722513089,
-      "grad_norm": 0.2749098422011923,
+      "grad_norm": 0.2708320400396081,
       "learning_rate": 1.3452531619894837e-05,
-      "logits/chosen": 0.83984375,
-      "logits/rejected": 2.28125,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -2048.0,
-      "loss": 0.0233,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 6.9375,
-      "rewards/rejected": -15.75,
+      "logits/chosen": -3.140625,
+      "logits/rejected": -1.9765625,
+      "logps/chosen": -1320.0,
+      "logps/rejected": -2112.0,
+      "loss": 0.0234,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -9.125,
+      "rewards/margins": 7.125,
+      "rewards/rejected": -16.25,
       "step": 1970
     },
     {
       "epoch": 2.073298429319372,
-      "grad_norm": 0.2938087162993104,
+      "grad_norm": 0.7094881971284149,
       "learning_rate": 1.3183188073985245e-05,
-      "logits/chosen": 0.41796875,
-      "logits/rejected": 2.046875,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -2048.0,
-      "loss": 0.0238,
+      "logits/chosen": -3.671875,
+      "logits/rejected": -2.28125,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -2128.0,
+      "loss": 0.0291,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 7.0,
-      "rewards/rejected": -15.8125,
+      "rewards/chosen": -9.375,
+      "rewards/margins": 7.1875,
+      "rewards/rejected": -16.625,
       "step": 1980
     },
     {
       "epoch": 2.0837696335078535,
-      "grad_norm": 0.48008190123542704,
+      "grad_norm": 0.44420344034550036,
       "learning_rate": 1.2915599333096351e-05,
-      "logits/chosen": 0.58984375,
-      "logits/rejected": 1.7265625,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -2080.0,
-      "loss": 0.0234,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.5,
-      "rewards/margins": 6.65625,
-      "rewards/rejected": -16.25,
+      "logits/chosen": -3.609375,
+      "logits/rejected": -2.609375,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -2144.0,
+      "loss": 0.0196,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -9.9375,
+      "rewards/margins": 6.90625,
+      "rewards/rejected": -16.875,
       "step": 1990
     },
     {
       "epoch": 2.094240837696335,
-      "grad_norm": 0.6894165128753994,
+      "grad_norm": 0.23683641427994503,
       "learning_rate": 1.264980513434752e-05,
-      "logits/chosen": 1.046875,
-      "logits/rejected": 2.765625,
-      "logps/chosen": -1328.0,
-      "logps/rejected": -2096.0,
-      "loss": 0.0234,
+      "logits/chosen": -3.375,
+      "logits/rejected": -2.109375,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -2144.0,
+      "loss": 0.0214,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 6.8125,
-      "rewards/rejected": -15.8125,
+      "rewards/chosen": -9.625,
+      "rewards/margins": 6.78125,
+      "rewards/rejected": -16.375,
       "step": 2000
     },
     {
       "epoch": 2.094240837696335,
-      "eval_logits/chosen": 4.5,
-      "eval_logits/rejected": 6.59375,
-      "eval_logps/chosen": -1480.0,
-      "eval_logps/rejected": -1568.0,
-      "eval_loss": 1.3751250505447388,
-      "eval_rewards/accuracies": 0.595703125,
-      "eval_rewards/chosen": -11.625,
-      "eval_rewards/margins": 1.2265625,
-      "eval_rewards/rejected": -12.8125,
-      "eval_runtime": 46.5434,
-      "eval_samples_per_second": 42.971,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -1.203125,
+      "eval_logits/rejected": 0.197265625,
+      "eval_logps/chosen": -1544.0,
+      "eval_logps/rejected": -1624.0,
+      "eval_loss": 1.5353906154632568,
+      "eval_rewards/accuracies": 0.609375,
+      "eval_rewards/chosen": -12.25,
+      "eval_rewards/margins": 1.1015625,
+      "eval_rewards/rejected": -13.3125,
+      "eval_runtime": 58.3851,
+      "eval_samples_per_second": 34.255,
+      "eval_steps_per_second": 0.548,
       "step": 2000
     },
     {
       "epoch": 2.1047120418848166,
-      "grad_norm": 0.31795412384725125,
+      "grad_norm": 0.26196429588688636,
       "learning_rate": 1.2385844948367324e-05,
-      "logits/chosen": 0.84375,
-      "logits/rejected": 2.0,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -2000.0,
-      "loss": 0.0366,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 6.5625,
-      "rewards/rejected": -15.4375,
+      "logits/chosen": -3.703125,
+      "logits/rejected": -2.65625,
+      "logps/chosen": -1304.0,
+      "logps/rejected": -2080.0,
+      "loss": 0.037,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -9.3125,
+      "rewards/margins": 6.90625,
+      "rewards/rejected": -16.25,
       "step": 2010
     },
     {
       "epoch": 2.115183246073298,
-      "grad_norm": 0.15587259667727116,
+      "grad_norm": 0.20124111190306695,
       "learning_rate": 1.2123757973432114e-05,
-      "logits/chosen": 0.470703125,
-      "logits/rejected": 1.90625,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -2112.0,
-      "loss": 0.0299,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -9.5625,
-      "rewards/margins": 7.0,
+      "logits/chosen": -4.09375,
+      "logits/rejected": -2.859375,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -2128.0,
+      "loss": 0.0237,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -9.375,
+      "rewards/margins": 7.15625,
       "rewards/rejected": -16.5,
       "step": 2020
     },
     {
       "epoch": 2.1256544502617802,
-      "grad_norm": 0.3474987892053459,
+      "grad_norm": 0.39381225826078137,
       "learning_rate": 1.1863583129645106e-05,
-      "logits/chosen": 1.421875,
-      "logits/rejected": 2.703125,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -2016.0,
-      "loss": 0.0316,
+      "logits/chosen": -3.46875,
+      "logits/rejected": -2.46875,
+      "logps/chosen": -1304.0,
+      "logps/rejected": -2064.0,
+      "loss": 0.032,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 6.875,
-      "rewards/rejected": -15.75,
+      "rewards/chosen": -9.125,
+      "rewards/margins": 7.09375,
+      "rewards/rejected": -16.25,
       "step": 2030
     },
     {
       "epoch": 2.136125654450262,
-      "grad_norm": 0.9942111051273294,
+      "grad_norm": 0.3769545888355418,
       "learning_rate": 1.1605359053156607e-05,
-      "logits/chosen": 1.0234375,
-      "logits/rejected": 2.984375,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -2128.0,
-      "loss": 0.0229,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.1875,
-      "rewards/margins": 7.3125,
-      "rewards/rejected": -16.5,
+      "logits/chosen": -3.59375,
+      "logits/rejected": -2.296875,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -2208.0,
+      "loss": 0.0232,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -9.625,
+      "rewards/margins": 7.625,
+      "rewards/rejected": -17.25,
       "step": 2040
     },
     {
       "epoch": 2.1465968586387434,
-      "grad_norm": 0.9481551402152563,
+      "grad_norm": 0.3477578926611757,
       "learning_rate": 1.1349124090426644e-05,
-      "logits/chosen": 2.015625,
-      "logits/rejected": 3.34375,
-      "logps/chosen": -1384.0,
-      "logps/rejected": -2224.0,
-      "loss": 0.0258,
+      "logits/chosen": -2.921875,
+      "logits/rejected": -1.84375,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -2256.0,
+      "loss": 0.0262,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -9.8125,
-      "rewards/margins": 7.5625,
-      "rewards/rejected": -17.375,
+      "rewards/chosen": -10.125,
+      "rewards/margins": 7.46875,
+      "rewards/rejected": -17.625,
       "step": 2050
     },
     {
       "epoch": 2.157068062827225,
-      "grad_norm": 0.6482072843527241,
+      "grad_norm": 0.3358957461152107,
       "learning_rate": 1.1094916292530406e-05,
-      "logits/chosen": 3.234375,
-      "logits/rejected": 3.796875,
-      "logps/chosen": -1400.0,
-      "logps/rejected": -2160.0,
-      "loss": 0.0425,
+      "logits/chosen": -2.234375,
+      "logits/rejected": -1.71875,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -2176.0,
+      "loss": 0.0223,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.8125,
-      "rewards/margins": 7.125,
-      "rewards/rejected": -17.0,
+      "rewards/chosen": -10.125,
+      "rewards/margins": 7.03125,
+      "rewards/rejected": -17.125,
       "step": 2060
     },
     {
       "epoch": 2.167539267015707,
-      "grad_norm": 0.2825658903152756,
+      "grad_norm": 0.4263981236077306,
       "learning_rate": 1.0842773409507623e-05,
-      "logits/chosen": 2.25,
-      "logits/rejected": 3.640625,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -2128.0,
-      "loss": 0.0248,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 7.625,
-      "rewards/rejected": -16.5,
+      "logits/chosen": -2.6875,
+      "logits/rejected": -1.609375,
+      "logps/chosen": -1376.0,
+      "logps/rejected": -2224.0,
+      "loss": 0.0233,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -9.9375,
+      "rewards/margins": 7.5625,
+      "rewards/rejected": -17.5,
       "step": 2070
     },
     {
       "epoch": 2.1780104712041886,
-      "grad_norm": 0.40355584735292016,
+      "grad_norm": 0.4102416159370556,
       "learning_rate": 1.0592732884756754e-05,
-      "logits/chosen": 1.4296875,
-      "logits/rejected": 2.65625,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -2288.0,
-      "loss": 0.0167,
+      "logits/chosen": -3.25,
+      "logits/rejected": -2.25,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.0186,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.875,
-      "rewards/margins": 7.875,
-      "rewards/rejected": -17.75,
+      "rewards/chosen": -10.875,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -19.125,
       "step": 2080
     },
     {
       "epoch": 2.1780104712041886,
-      "eval_logits/chosen": 5.4375,
-      "eval_logits/rejected": 7.5,
-      "eval_logps/chosen": -1584.0,
-      "eval_logps/rejected": -1688.0,
-      "eval_loss": 1.485531210899353,
-      "eval_rewards/accuracies": 0.59765625,
-      "eval_rewards/chosen": -12.625,
-      "eval_rewards/margins": 1.375,
-      "eval_rewards/rejected": -14.0,
-      "eval_runtime": 46.5572,
-      "eval_samples_per_second": 42.958,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -0.91015625,
+      "eval_logits/rejected": 0.490234375,
+      "eval_logps/chosen": -1680.0,
+      "eval_logps/rejected": -1784.0,
+      "eval_loss": 1.5789999961853027,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -13.5625,
+      "eval_rewards/margins": 1.390625,
+      "eval_rewards/rejected": -14.9375,
+      "eval_runtime": 58.382,
+      "eval_samples_per_second": 34.257,
+      "eval_steps_per_second": 0.548,
       "step": 2080
     },
     {
       "epoch": 2.18848167539267,
-      "grad_norm": 1.7264983132076623,
+      "grad_norm": 0.8842746504219667,
       "learning_rate": 1.0344831849474507e-05,
-      "logits/chosen": 2.1875,
-      "logits/rejected": 3.71875,
-      "logps/chosen": -1432.0,
-      "logps/rejected": -2224.0,
-      "loss": 0.0306,
+      "logits/chosen": -3.140625,
+      "logits/rejected": -1.859375,
+      "logps/chosen": -1504.0,
+      "logps/rejected": -2320.0,
+      "loss": 0.0228,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.1875,
-      "rewards/margins": 7.375,
-      "rewards/rejected": -17.5,
+      "rewards/chosen": -10.9375,
+      "rewards/margins": 7.53125,
+      "rewards/rejected": -18.375,
       "step": 2090
     },
     {
       "epoch": 2.1989528795811517,
-      "grad_norm": 0.3042699666231012,
+      "grad_norm": 0.7925437133179385,
       "learning_rate": 1.0099107117141879e-05,
-      "logits/chosen": 2.4375,
-      "logits/rejected": 4.15625,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -2192.0,
-      "loss": 0.0391,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -9.1875,
+      "logits/chosen": -3.109375,
+      "logits/rejected": -1.734375,
+      "logps/chosen": -1376.0,
+      "logps/rejected": -2272.0,
+      "loss": 0.0266,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -9.9375,
       "rewards/margins": 8.0625,
-      "rewards/rejected": -17.25,
+      "rewards/rejected": -18.0,
       "step": 2100
     },
     {
       "epoch": 2.2094240837696333,
-      "grad_norm": 0.2629249152738321,
+      "grad_norm": 1.8721532767967497,
       "learning_rate": 9.855595178057334e-06,
-      "logits/chosen": 0.71484375,
-      "logits/rejected": 2.390625,
-      "logps/chosen": -1360.0,
-      "logps/rejected": -2176.0,
-      "loss": 0.022,
+      "logits/chosen": -3.609375,
+      "logits/rejected": -2.5,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -2288.0,
+      "loss": 0.0212,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.375,
-      "rewards/margins": 7.71875,
-      "rewards/rejected": -17.125,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 7.90625,
+      "rewards/rejected": -18.25,
       "step": 2110
     },
     {
       "epoch": 2.2198952879581153,
-      "grad_norm": 0.2807081185276977,
+      "grad_norm": 0.6894065456195657,
       "learning_rate": 9.614332193917949e-06,
-      "logits/chosen": 2.21875,
-      "logits/rejected": 3.5,
-      "logps/chosen": -1416.0,
-      "logps/rejected": -2272.0,
-      "loss": 0.0279,
+      "logits/chosen": -2.53125,
+      "logits/rejected": -1.703125,
+      "logps/chosen": -1512.0,
+      "logps/rejected": -2352.0,
+      "loss": 0.027,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.875,
-      "rewards/margins": 7.9375,
-      "rewards/rejected": -17.875,
+      "rewards/chosen": -10.875,
+      "rewards/margins": 7.78125,
+      "rewards/rejected": -18.625,
       "step": 2120
     },
     {
       "epoch": 2.230366492146597,
-      "grad_norm": 0.36875019300866685,
-      "learning_rate": 9.375353992449385e-06,
-      "logits/chosen": 1.421875,
-      "logits/rejected": 2.6875,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -2272.0,
+      "grad_norm": 0.17834479810287274,
+      "learning_rate": 9.375353992449385e-06,
+      "logits/chosen": -3.171875,
+      "logits/rejected": -2.359375,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2336.0,
       "loss": 0.017,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.0,
+      "rewards/chosen": -10.6875,
       "rewards/margins": 7.71875,
-      "rewards/rejected": -17.75,
+      "rewards/rejected": -18.375,
       "step": 2130
     },
     {
       "epoch": 2.2408376963350785,
-      "grad_norm": 1.1407207424945793,
+      "grad_norm": 0.40291891657603873,
       "learning_rate": 9.138696062085442e-06,
-      "logits/chosen": 1.1875,
-      "logits/rejected": 3.046875,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -2160.0,
+      "logits/chosen": -3.140625,
+      "logits/rejected": -1.9453125,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -2224.0,
       "loss": 0.0238,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.5,
-      "rewards/margins": 7.3125,
-      "rewards/rejected": -16.75,
+      "rewards/chosen": -10.125,
+      "rewards/margins": 7.34375,
+      "rewards/rejected": -17.5,
       "step": 2140
     },
     {
       "epoch": 2.25130890052356,
-      "grad_norm": 0.36192661839771645,
+      "grad_norm": 0.34487096485258734,
       "learning_rate": 8.904393546698007e-06,
-      "logits/chosen": 1.359375,
-      "logits/rejected": 2.84375,
-      "logps/chosen": -1352.0,
-      "logps/rejected": -2208.0,
-      "loss": 0.0203,
+      "logits/chosen": -3.125,
+      "logits/rejected": -2.234375,
+      "logps/chosen": -1376.0,
+      "logps/rejected": -2224.0,
+      "loss": 0.0262,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.375,
-      "rewards/margins": 8.25,
-      "rewards/rejected": -17.625,
+      "rewards/chosen": -9.625,
+      "rewards/margins": 8.125,
+      "rewards/rejected": -17.75,
       "step": 2150
     },
     {
       "epoch": 2.261780104712042,
-      "grad_norm": 0.39592939486939527,
+      "grad_norm": 0.39517389471491227,
       "learning_rate": 8.672481240378142e-06,
-      "logits/chosen": 1.796875,
-      "logits/rejected": 3.609375,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -2080.0,
-      "loss": 0.0366,
+      "logits/chosen": -3.25,
+      "logits/rejected": -1.8984375,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -2040.0,
+      "loss": 0.0395,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.375,
-      "rewards/margins": 7.09375,
-      "rewards/rejected": -16.5,
+      "rewards/chosen": -8.9375,
+      "rewards/margins": 7.0625,
+      "rewards/rejected": -16.0,
       "step": 2160
     },
     {
       "epoch": 2.261780104712042,
-      "eval_logits/chosen": 5.25,
-      "eval_logits/rejected": 7.3125,
-      "eval_logps/chosen": -1536.0,
-      "eval_logps/rejected": -1648.0,
-      "eval_loss": 1.4665000438690186,
-      "eval_rewards/accuracies": 0.595703125,
-      "eval_rewards/chosen": -12.1875,
-      "eval_rewards/margins": 1.359375,
-      "eval_rewards/rejected": -13.5625,
-      "eval_runtime": 46.5493,
-      "eval_samples_per_second": 42.965,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -0.765625,
+      "eval_logits/rejected": 0.5390625,
+      "eval_logps/chosen": -1520.0,
+      "eval_logps/rejected": -1608.0,
+      "eval_loss": 1.5234375,
+      "eval_rewards/accuracies": 0.603515625,
+      "eval_rewards/chosen": -12.0625,
+      "eval_rewards/margins": 1.140625,
+      "eval_rewards/rejected": -13.1875,
+      "eval_runtime": 58.3709,
+      "eval_samples_per_second": 34.264,
+      "eval_steps_per_second": 0.548,
       "step": 2160
     },
     {
       "epoch": 2.2722513089005236,
-      "grad_norm": 0.8579897661137699,
+      "grad_norm": 0.46571112928466485,
       "learning_rate": 8.44299358226919e-06,
-      "logits/chosen": 2.0625,
-      "logits/rejected": 3.40625,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -2080.0,
-      "loss": 0.0329,
+      "logits/chosen": -2.75,
+      "logits/rejected": -1.7890625,
+      "logps/chosen": -1320.0,
+      "logps/rejected": -2112.0,
+      "loss": 0.032,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 7.3125,
-      "rewards/rejected": -16.375,
+      "rewards/chosen": -9.125,
+      "rewards/margins": 7.46875,
+      "rewards/rejected": -16.625,
       "step": 2170
     },
     {
       "epoch": 2.282722513089005,
-      "grad_norm": 0.35084016033212284,
+      "grad_norm": 0.707446903747919,
       "learning_rate": 8.215964651452457e-06,
-      "logits/chosen": 2.28125,
-      "logits/rejected": 3.828125,
-      "logps/chosen": -1336.0,
-      "logps/rejected": -2208.0,
-      "loss": 0.0186,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -9.375,
-      "rewards/margins": 7.78125,
-      "rewards/rejected": -17.125,
+      "logits/chosen": -2.390625,
+      "logits/rejected": -1.3984375,
+      "logps/chosen": -1384.0,
+      "logps/rejected": -2256.0,
+      "loss": 0.02,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -9.8125,
+      "rewards/margins": 7.71875,
+      "rewards/rejected": -17.5,
       "step": 2180
     },
     {
       "epoch": 2.2931937172774868,
-      "grad_norm": 0.056951293426571126,
+      "grad_norm": 0.13804405945869258,
       "learning_rate": 7.991428161886502e-06,
-      "logits/chosen": 1.9140625,
-      "logits/rejected": 3.015625,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -2128.0,
-      "loss": 0.0186,
+      "logits/chosen": -2.59375,
+      "logits/rejected": -1.8359375,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -2224.0,
+      "loss": 0.0182,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.3125,
-      "rewards/margins": 7.28125,
-      "rewards/rejected": -16.625,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 7.46875,
+      "rewards/rejected": -17.5,
       "step": 2190
     },
     {
       "epoch": 2.303664921465969,
-      "grad_norm": 0.2082322336678535,
+      "grad_norm": 0.23850356165212758,
       "learning_rate": 7.76941745740061e-06,
-      "logits/chosen": 2.171875,
-      "logits/rejected": 3.28125,
-      "logps/chosen": -1456.0,
-      "logps/rejected": -2288.0,
-      "loss": 0.0284,
+      "logits/chosen": -2.5625,
+      "logits/rejected": -1.6953125,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2384.0,
+      "loss": 0.0257,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -10.25,
-      "rewards/margins": 7.8125,
-      "rewards/rejected": -18.0,
+      "rewards/chosen": -11.125,
+      "rewards/margins": 7.90625,
+      "rewards/rejected": -19.125,
       "step": 2200
     },
     {
       "epoch": 2.3141361256544504,
-      "grad_norm": 0.30053173087859586,
+      "grad_norm": 0.6193436019404015,
       "learning_rate": 7.549965506743123e-06,
-      "logits/chosen": 2.75,
-      "logits/rejected": 4.125,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -2112.0,
-      "loss": 0.0218,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 7.34375,
-      "rewards/rejected": -17.0,
+      "logits/chosen": -2.5625,
+      "logits/rejected": -1.4765625,
+      "logps/chosen": -1384.0,
+      "logps/rejected": -2160.0,
+      "loss": 0.0242,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 7.40625,
+      "rewards/rejected": -17.375,
       "step": 2210
     },
     {
       "epoch": 2.324607329842932,
-      "grad_norm": 0.35272020324828024,
+      "grad_norm": 0.33592373037548906,
       "learning_rate": 7.333104898685631e-06,
-      "logits/chosen": 1.6484375,
-      "logits/rejected": 2.734375,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -2096.0,
-      "loss": 0.0302,
+      "logits/chosen": -2.9375,
+      "logits/rejected": -2.03125,
+      "logps/chosen": -1384.0,
+      "logps/rejected": -2144.0,
+      "loss": 0.0283,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.375,
+      "rewards/chosen": -9.875,
       "rewards/margins": 7.25,
-      "rewards/rejected": -16.625,
+      "rewards/rejected": -17.125,
       "step": 2220
     },
     {
       "epoch": 2.3350785340314135,
-      "grad_norm": 0.2952213372174239,
+      "grad_norm": 0.18341977992612982,
       "learning_rate": 7.118867837183518e-06,
-      "logits/chosen": 1.0078125,
-      "logits/rejected": 2.46875,
-      "logps/chosen": -1440.0,
-      "logps/rejected": -2240.0,
-      "loss": 0.0186,
+      "logits/chosen": -3.234375,
+      "logits/rejected": -2.203125,
+      "logps/chosen": -1480.0,
+      "logps/rejected": -2288.0,
+      "loss": 0.0206,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.6875,
-      "rewards/margins": 7.75,
-      "rewards/rejected": -17.375,
+      "rewards/chosen": -10.0625,
+      "rewards/margins": 7.8125,
+      "rewards/rejected": -17.875,
       "step": 2230
     },
     {
       "epoch": 2.345549738219895,
-      "grad_norm": 2.078969158986083,
+      "grad_norm": 0.16833754978760213,
       "learning_rate": 6.9072861365936045e-06,
-      "logits/chosen": 1.375,
-      "logits/rejected": 2.3125,
-      "logps/chosen": -1536.0,
-      "logps/rejected": -2320.0,
-      "loss": 0.0268,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.6875,
-      "rewards/margins": 7.59375,
-      "rewards/rejected": -18.25,
+      "logits/chosen": -3.15625,
+      "logits/rejected": -2.421875,
+      "logps/chosen": -1560.0,
+      "logps/rejected": -2368.0,
+      "loss": 0.0217,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -10.9375,
+      "rewards/margins": 7.71875,
+      "rewards/rejected": -18.625,
       "step": 2240
     },
     {
       "epoch": 2.345549738219895,
-      "eval_logits/chosen": 5.71875,
-      "eval_logits/rejected": 7.6875,
-      "eval_logps/chosen": -1600.0,
-      "eval_logps/rejected": -1712.0,
-      "eval_loss": 1.5464531183242798,
-      "eval_rewards/accuracies": 0.599609375,
-      "eval_rewards/chosen": -12.8125,
-      "eval_rewards/margins": 1.359375,
-      "eval_rewards/rejected": -14.1875,
-      "eval_runtime": 46.5537,
-      "eval_samples_per_second": 42.961,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -0.4140625,
+      "eval_logits/rejected": 0.89453125,
+      "eval_logps/chosen": -1632.0,
+      "eval_logps/rejected": -1744.0,
+      "eval_loss": 1.5866718292236328,
+      "eval_rewards/accuracies": 0.603515625,
+      "eval_rewards/chosen": -13.1875,
+      "eval_rewards/margins": 1.375,
+      "eval_rewards/rejected": -14.5625,
+      "eval_runtime": 58.3931,
+      "eval_samples_per_second": 34.251,
+      "eval_steps_per_second": 0.548,
       "step": 2240
     },
     {
       "epoch": 2.356020942408377,
-      "grad_norm": 0.8997006314893151,
+      "grad_norm": 0.5242611747482183,
       "learning_rate": 6.6983912169497005e-06,
-      "logits/chosen": 1.515625,
-      "logits/rejected": 2.890625,
-      "logps/chosen": -1424.0,
-      "logps/rejected": -2224.0,
-      "loss": 0.0268,
+      "logits/chosen": -2.984375,
+      "logits/rejected": -2.0,
+      "logps/chosen": -1464.0,
+      "logps/rejected": -2240.0,
+      "loss": 0.0298,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.8125,
-      "rewards/margins": 7.65625,
-      "rewards/rejected": -17.5,
+      "rewards/chosen": -10.1875,
+      "rewards/margins": 7.5,
+      "rewards/rejected": -17.75,
       "step": 2250
     },
     {
       "epoch": 2.3664921465968587,
-      "grad_norm": 0.30283973873202213,
+      "grad_norm": 0.15027913885981314,
       "learning_rate": 6.49221409929677e-06,
-      "logits/chosen": 3.140625,
-      "logits/rejected": 4.21875,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -2160.0,
-      "loss": 0.0165,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.5625,
-      "rewards/margins": 7.71875,
-      "rewards/rejected": -17.25,
+      "logits/chosen": -2.09375,
+      "logits/rejected": -1.1796875,
+      "logps/chosen": -1384.0,
+      "logps/rejected": -2192.0,
+      "loss": 0.0232,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 7.5625,
+      "rewards/rejected": -17.625,
       "step": 2260
     },
     {
       "epoch": 2.3769633507853403,
-      "grad_norm": 0.44282921904173,
+      "grad_norm": 0.373813967354651,
       "learning_rate": 6.2887854010842075e-06,
-      "logits/chosen": 3.015625,
-      "logits/rejected": 3.390625,
-      "logps/chosen": -1424.0,
-      "logps/rejected": -2288.0,
-      "loss": 0.0206,
+      "logits/chosen": -2.28125,
+      "logits/rejected": -1.7578125,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -2304.0,
+      "loss": 0.0214,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -10.4375,
-      "rewards/margins": 7.96875,
-      "rewards/rejected": -18.375,
+      "rewards/chosen": -10.5,
+      "rewards/margins": 8.0625,
+      "rewards/rejected": -18.625,
       "step": 2270
     },
     {
       "epoch": 2.387434554973822,
-      "grad_norm": 0.32382837530749703,
+      "grad_norm": 0.1649978540371382,
       "learning_rate": 6.0881353316191385e-06,
-      "logits/chosen": 1.6953125,
-      "logits/rejected": 3.078125,
-      "logps/chosen": -1368.0,
-      "logps/rejected": -2160.0,
-      "loss": 0.0184,
+      "logits/chosen": -2.78125,
+      "logits/rejected": -1.875,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -2208.0,
+      "loss": 0.0178,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.375,
-      "rewards/margins": 7.84375,
-      "rewards/rejected": -17.25,
+      "rewards/chosen": -9.625,
+      "rewards/margins": 8.0,
+      "rewards/rejected": -17.625,
       "step": 2280
     },
     {
       "epoch": 2.3979057591623034,
-      "grad_norm": 0.43433211076684164,
+      "grad_norm": 0.5073069876053594,
       "learning_rate": 5.890293687580381e-06,
-      "logits/chosen": 2.171875,
-      "logits/rejected": 3.8125,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -2272.0,
-      "loss": 0.0255,
+      "logits/chosen": -2.390625,
+      "logits/rejected": -1.1796875,
+      "logps/chosen": -1456.0,
+      "logps/rejected": -2336.0,
+      "loss": 0.0259,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.75,
-      "rewards/margins": 8.625,
-      "rewards/rejected": -18.375,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 8.6875,
+      "rewards/rejected": -19.0,
       "step": 2290
     },
     {
       "epoch": 2.4083769633507854,
-      "grad_norm": 0.3477608029134286,
+      "grad_norm": 0.4661606507629274,
       "learning_rate": 5.695289848593532e-06,
-      "logits/chosen": 1.46875,
-      "logits/rejected": 2.484375,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -2128.0,
-      "loss": 0.0235,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -9.5,
-      "rewards/margins": 7.4375,
-      "rewards/rejected": -16.875,
+      "logits/chosen": -2.765625,
+      "logits/rejected": -2.078125,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -2208.0,
+      "loss": 0.0252,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -10.375,
+      "rewards/margins": 7.40625,
+      "rewards/rejected": -17.75,
       "step": 2300
     },
     {
       "epoch": 2.418848167539267,
-      "grad_norm": 0.306592854618104,
+      "grad_norm": 0.2454860873179466,
       "learning_rate": 5.50315277286812e-06,
-      "logits/chosen": 1.8515625,
-      "logits/rejected": 2.953125,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -2192.0,
-      "loss": 0.0217,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.6875,
-      "rewards/margins": 7.75,
-      "rewards/rejected": -17.375,
+      "logits/chosen": -2.734375,
+      "logits/rejected": -1.9609375,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -2272.0,
+      "loss": 0.0225,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.4375,
+      "rewards/margins": 7.78125,
+      "rewards/rejected": -18.25,
       "step": 2310
     },
     {
       "epoch": 2.4293193717277486,
-      "grad_norm": 0.464354643862231,
+      "grad_norm": 0.8867505110385082,
       "learning_rate": 5.313910992897281e-06,
-      "logits/chosen": 2.453125,
-      "logits/rejected": 3.921875,
-      "logps/chosen": -1320.0,
-      "logps/rejected": -2176.0,
-      "loss": 0.0241,
+      "logits/chosen": -2.5,
+      "logits/rejected": -1.3671875,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -2224.0,
+      "loss": 0.0268,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.125,
-      "rewards/margins": 7.65625,
-      "rewards/rejected": -16.75,
+      "rewards/chosen": -9.5,
+      "rewards/margins": 7.8125,
+      "rewards/rejected": -17.375,
       "step": 2320
     },
     {
       "epoch": 2.4293193717277486,
-      "eval_logits/chosen": 5.1875,
-      "eval_logits/rejected": 7.25,
-      "eval_logps/chosen": -1544.0,
-      "eval_logps/rejected": -1656.0,
-      "eval_loss": 1.5025312900543213,
-      "eval_rewards/accuracies": 0.599609375,
-      "eval_rewards/chosen": -12.25,
+      "eval_logits/chosen": -0.67578125,
+      "eval_logits/rejected": 0.6796875,
+      "eval_logps/chosen": -1616.0,
+      "eval_logps/rejected": -1728.0,
+      "eval_loss": 1.5887500047683716,
+      "eval_rewards/accuracies": 0.603515625,
+      "eval_rewards/chosen": -13.0,
       "eval_rewards/margins": 1.421875,
-      "eval_rewards/rejected": -13.6875,
-      "eval_runtime": 46.5635,
-      "eval_samples_per_second": 42.952,
-      "eval_steps_per_second": 0.687,
+      "eval_rewards/rejected": -14.375,
+      "eval_runtime": 58.4006,
+      "eval_samples_per_second": 34.246,
+      "eval_steps_per_second": 0.548,
       "step": 2320
     },
     {
       "epoch": 2.4397905759162306,
-      "grad_norm": 0.42125888365504927,
+      "grad_norm": 0.30684774939895193,
       "learning_rate": 5.127592611220658e-06,
-      "logits/chosen": 1.046875,
-      "logits/rejected": 3.140625,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -2256.0,
-      "loss": 0.0302,
+      "logits/chosen": -3.4375,
+      "logits/rejected": -1.90625,
+      "logps/chosen": -1456.0,
+      "logps/rejected": -2304.0,
+      "loss": 0.0304,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.625,
+      "rewards/chosen": -10.1875,
       "rewards/margins": 8.25,
-      "rewards/rejected": -17.875,
+      "rewards/rejected": -18.5,
       "step": 2330
     },
     {
       "epoch": 2.450261780104712,
-      "grad_norm": 0.5222731895651916,
+      "grad_norm": 0.45073771624086617,
       "learning_rate": 4.944225296251159e-06,
-      "logits/chosen": 1.765625,
-      "logits/rejected": 3.109375,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -2160.0,
-      "loss": 0.0199,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -9.75,
-      "rewards/margins": 7.78125,
-      "rewards/rejected": -17.5,
+      "logits/chosen": -2.875,
+      "logits/rejected": -1.984375,
+      "logps/chosen": -1368.0,
+      "logps/rejected": -2176.0,
+      "loss": 0.02,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 7.65625,
+      "rewards/rejected": -17.625,
       "step": 2340
     },
     {
       "epoch": 2.4607329842931938,
-      "grad_norm": 0.586270317731517,
+      "grad_norm": 0.49167621153069013,
       "learning_rate": 4.763836278166172e-06,
-      "logits/chosen": 2.03125,
-      "logits/rejected": 3.96875,
-      "logps/chosen": -1352.0,
+      "logits/chosen": -2.921875,
+      "logits/rejected": -1.46875,
+      "logps/chosen": -1360.0,
       "logps/rejected": -2112.0,
-      "loss": 0.018,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -9.4375,
-      "rewards/margins": 7.5625,
+      "loss": 0.0166,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -9.5625,
+      "rewards/margins": 7.46875,
       "rewards/rejected": -17.0,
       "step": 2350
     },
     {
       "epoch": 2.4712041884816753,
-      "grad_norm": 1.4193626770389078,
+      "grad_norm": 0.5352599579735774,
       "learning_rate": 4.586452344863876e-06,
-      "logits/chosen": 1.140625,
-      "logits/rejected": 3.0,
-      "logps/chosen": -1384.0,
+      "logits/chosen": -3.375,
+      "logits/rejected": -2.0,
+      "logps/chosen": -1408.0,
       "logps/rejected": -2256.0,
-      "loss": 0.0241,
+      "loss": 0.0233,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 8.3125,
+      "rewards/chosen": -9.875,
+      "rewards/margins": 7.96875,
       "rewards/rejected": -17.875,
       "step": 2360
     },
     {
       "epoch": 2.481675392670157,
-      "grad_norm": 0.4125707439635376,
+      "grad_norm": 2.3092798257089684,
       "learning_rate": 4.412099837985192e-06,
-      "logits/chosen": 2.109375,
-      "logits/rejected": 3.75,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -2352.0,
-      "loss": 0.0263,
+      "logits/chosen": -2.859375,
+      "logits/rejected": -1.625,
+      "logps/chosen": -1440.0,
+      "logps/rejected": -2368.0,
+      "loss": 0.0281,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.875,
-      "rewards/margins": 8.3125,
-      "rewards/rejected": -18.25,
+      "rewards/chosen": -10.1875,
+      "rewards/margins": 8.25,
+      "rewards/rejected": -18.5,
       "step": 2370
     },
     {
       "epoch": 2.492146596858639,
-      "grad_norm": 0.11258623695863354,
+      "grad_norm": 0.05714738962128367,
       "learning_rate": 4.240804649002089e-06,
-      "logits/chosen": 1.546875,
-      "logits/rejected": 3.390625,
-      "logps/chosen": -1304.0,
+      "logits/chosen": -3.140625,
+      "logits/rejected": -1.7578125,
+      "logps/chosen": -1328.0,
       "logps/rejected": -2112.0,
-      "loss": 0.029,
+      "loss": 0.0258,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -9.3125,
-      "rewards/margins": 7.6875,
+      "rewards/chosen": -9.5625,
+      "rewards/margins": 7.46875,
       "rewards/rejected": -17.0,
       "step": 2380
     },
     {
       "epoch": 2.5026178010471205,
-      "grad_norm": 0.683513567240261,
+      "grad_norm": 0.6554378710301283,
       "learning_rate": 4.072592215372598e-06,
-      "logits/chosen": 0.8515625,
-      "logits/rejected": 2.25,
-      "logps/chosen": -1456.0,
-      "logps/rejected": -2336.0,
-      "loss": 0.0249,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -10.125,
-      "rewards/margins": 8.125,
-      "rewards/rejected": -18.25,
+      "logits/chosen": -3.359375,
+      "logits/rejected": -2.328125,
+      "logps/chosen": -1496.0,
+      "logps/rejected": -2400.0,
+      "loss": 0.0212,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.5625,
+      "rewards/margins": 8.25,
+      "rewards/rejected": -18.875,
       "step": 2390
     },
     {
       "epoch": 2.513089005235602,
-      "grad_norm": 0.6928452346894448,
+      "grad_norm": 0.3765649597477542,
       "learning_rate": 3.907487516763389e-06,
-      "logits/chosen": 2.125,
-      "logits/rejected": 3.609375,
-      "logps/chosen": -1360.0,
-      "logps/rejected": -2208.0,
-      "loss": 0.0261,
+      "logits/chosen": -2.5625,
+      "logits/rejected": -1.5546875,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -2272.0,
+      "loss": 0.0238,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.6875,
-      "rewards/margins": 7.84375,
-      "rewards/rejected": -17.5,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 8.0,
+      "rewards/rejected": -18.25,
       "step": 2400
     },
     {
       "epoch": 2.513089005235602,
-      "eval_logits/chosen": 5.5,
-      "eval_logits/rejected": 7.53125,
-      "eval_logps/chosen": -1592.0,
-      "eval_logps/rejected": -1712.0,
-      "eval_loss": 1.5371249914169312,
-      "eval_rewards/accuracies": 0.59375,
-      "eval_rewards/chosen": -12.75,
-      "eval_rewards/margins": 1.4375,
-      "eval_rewards/rejected": -14.1875,
-      "eval_runtime": 46.5454,
-      "eval_samples_per_second": 42.969,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -0.36328125,
+      "eval_logits/rejected": 0.96484375,
+      "eval_logps/chosen": -1680.0,
+      "eval_logps/rejected": -1792.0,
+      "eval_loss": 1.6646562814712524,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -13.625,
+      "eval_rewards/margins": 1.4453125,
+      "eval_rewards/rejected": -15.0625,
+      "eval_runtime": 58.3862,
+      "eval_samples_per_second": 34.255,
+      "eval_steps_per_second": 0.548,
       "step": 2400
     },
     {
       "epoch": 2.5235602094240837,
-      "grad_norm": 0.25545877506113723,
+      "grad_norm": 0.21690087912444653,
       "learning_rate": 3.745515071340286e-06,
-      "logits/chosen": 1.765625,
-      "logits/rejected": 3.1875,
-      "logps/chosen": -1400.0,
-      "logps/rejected": -2256.0,
-      "loss": 0.023,
+      "logits/chosen": -2.515625,
+      "logits/rejected": -1.5078125,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2368.0,
+      "loss": 0.0274,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -9.8125,
-      "rewards/margins": 7.84375,
-      "rewards/rejected": -17.625,
+      "rewards/chosen": -10.6875,
+      "rewards/margins": 8.0625,
+      "rewards/rejected": -18.75,
       "step": 2410
     },
     {
       "epoch": 2.5340314136125652,
-      "grad_norm": 0.5294373658951493,
+      "grad_norm": 0.6824251542367844,
       "learning_rate": 3.586698932127236e-06,
-      "logits/chosen": 0.76953125,
-      "logits/rejected": 2.8125,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -2176.0,
-      "loss": 0.015,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 8.3125,
-      "rewards/rejected": -17.25,
+      "logits/chosen": -3.203125,
+      "logits/rejected": -1.953125,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -2240.0,
+      "loss": 0.0151,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -10.0625,
+      "rewards/margins": 7.90625,
+      "rewards/rejected": -18.0,
       "step": 2420
     },
     {
       "epoch": 2.5445026178010473,
-      "grad_norm": 0.41812473025548513,
+      "grad_norm": 0.48685071962233567,
       "learning_rate": 3.4310626834344744e-06,
-      "logits/chosen": 0.7890625,
-      "logits/rejected": 1.578125,
-      "logps/chosen": -1480.0,
-      "logps/rejected": -2320.0,
-      "loss": 0.022,
+      "logits/chosen": -3.328125,
+      "logits/rejected": -2.75,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2400.0,
+      "loss": 0.024,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.4375,
-      "rewards/margins": 7.96875,
-      "rewards/rejected": -18.375,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -19.25,
       "step": 2430
     },
     {
       "epoch": 2.554973821989529,
-      "grad_norm": 1.3919450416461203,
+      "grad_norm": 0.19224102409835456,
       "learning_rate": 3.2786294373562343e-06,
-      "logits/chosen": 1.9375,
-      "logits/rejected": 3.484375,
-      "logps/chosen": -1424.0,
-      "logps/rejected": -2224.0,
-      "loss": 0.0183,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.0,
-      "rewards/margins": 7.6875,
-      "rewards/rejected": -17.75,
+      "logits/chosen": -2.765625,
+      "logits/rejected": -1.578125,
+      "logps/chosen": -1440.0,
+      "logps/rejected": -2272.0,
+      "loss": 0.0184,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -10.1875,
+      "rewards/margins": 8.0,
+      "rewards/rejected": -18.125,
       "step": 2440
     },
     {
       "epoch": 2.5654450261780104,
-      "grad_norm": 0.4938981100074194,
+      "grad_norm": 0.7896008277375696,
       "learning_rate": 3.1294218303385526e-06,
-      "logits/chosen": 1.3125,
-      "logits/rejected": 3.265625,
-      "logps/chosen": -1384.0,
-      "logps/rejected": -2176.0,
-      "loss": 0.0215,
+      "logits/chosen": -2.953125,
+      "logits/rejected": -1.6328125,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -2224.0,
+      "loss": 0.0213,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.4375,
-      "rewards/margins": 8.0,
-      "rewards/rejected": -17.5,
+      "rewards/chosen": -10.0625,
+      "rewards/margins": 7.8125,
+      "rewards/rejected": -17.875,
       "step": 2450
     },
     {
       "epoch": 2.5759162303664924,
-      "grad_norm": 0.609185065558062,
+      "grad_norm": 0.3218181782901593,
       "learning_rate": 2.9834620198177483e-06,
-      "logits/chosen": 1.9921875,
-      "logits/rejected": 3.40625,
-      "logps/chosen": -1424.0,
-      "logps/rejected": -2256.0,
-      "loss": 0.0294,
+      "logits/chosen": -2.71875,
+      "logits/rejected": -1.5859375,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2320.0,
+      "loss": 0.0298,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -10.25,
-      "rewards/margins": 7.65625,
-      "rewards/rejected": -17.875,
+      "rewards/chosen": -10.6875,
+      "rewards/margins": 7.9375,
+      "rewards/rejected": -18.625,
       "step": 2460
     },
     {
       "epoch": 2.5863874345549736,
-      "grad_norm": 0.619689929797116,
+      "grad_norm": 0.6042892005024296,
       "learning_rate": 2.840771680930068e-06,
-      "logits/chosen": 1.296875,
-      "logits/rejected": 2.265625,
-      "logps/chosen": -1456.0,
-      "logps/rejected": -2384.0,
-      "loss": 0.0131,
+      "logits/chosen": -3.015625,
+      "logits/rejected": -2.25,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2480.0,
+      "loss": 0.0176,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.25,
-      "rewards/margins": 8.25,
-      "rewards/rejected": -18.5,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 8.375,
+      "rewards/rejected": -19.5,
       "step": 2470
     },
     {
       "epoch": 2.5968586387434556,
-      "grad_norm": 0.2422197645568945,
+      "grad_norm": 0.4597042291720016,
       "learning_rate": 2.701372003292868e-06,
-      "logits/chosen": 2.4375,
-      "logits/rejected": 3.71875,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -2192.0,
-      "loss": 0.0245,
+      "logits/chosen": -2.25,
+      "logits/rejected": -1.359375,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -2256.0,
+      "loss": 0.0227,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.25,
-      "rewards/margins": 7.96875,
-      "rewards/rejected": -17.25,
+      "rewards/chosen": -9.75,
+      "rewards/margins": 8.0625,
+      "rewards/rejected": -17.875,
       "step": 2480
     },
     {
       "epoch": 2.5968586387434556,
-      "eval_logits/chosen": 5.5625,
-      "eval_logits/rejected": 7.625,
-      "eval_logps/chosen": -1584.0,
-      "eval_logps/rejected": -1696.0,
-      "eval_loss": 1.5284062623977661,
-      "eval_rewards/accuracies": 0.59765625,
-      "eval_rewards/chosen": -12.625,
-      "eval_rewards/margins": 1.4453125,
-      "eval_rewards/rejected": -14.0625,
-      "eval_runtime": 46.5535,
-      "eval_samples_per_second": 42.961,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -0.419921875,
+      "eval_logits/rejected": 0.92578125,
+      "eval_logps/chosen": -1632.0,
+      "eval_logps/rejected": -1744.0,
+      "eval_loss": 1.5872968435287476,
+      "eval_rewards/accuracies": 0.609375,
+      "eval_rewards/chosen": -13.125,
+      "eval_rewards/margins": 1.4375,
+      "eval_rewards/rejected": -14.5625,
+      "eval_runtime": 58.38,
+      "eval_samples_per_second": 34.258,
+      "eval_steps_per_second": 0.548,
       "step": 2480
     },
     {
       "epoch": 2.607329842931937,
-      "grad_norm": 0.4352445123603095,
+      "grad_norm": 0.6496482073621739,
       "learning_rate": 2.56528368785795e-06,
-      "logits/chosen": 1.6796875,
-      "logits/rejected": 3.671875,
-      "logps/chosen": -1360.0,
-      "logps/rejected": -2208.0,
-      "loss": 0.0227,
+      "logits/chosen": -3.0,
+      "logits/rejected": -1.5703125,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -2256.0,
+      "loss": 0.0213,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.375,
-      "rewards/margins": 8.1875,
-      "rewards/rejected": -17.5,
+      "rewards/chosen": -9.75,
+      "rewards/margins": 8.25,
+      "rewards/rejected": -18.0,
       "step": 2490
     },
     {
       "epoch": 2.6178010471204187,
-      "grad_norm": 1.4100622198841122,
+      "grad_norm": 0.8022946780517504,
       "learning_rate": 2.432526943837471e-06,
-      "logits/chosen": 2.1875,
-      "logits/rejected": 4.3125,
-      "logps/chosen": -1320.0,
-      "logps/rejected": -2160.0,
-      "loss": 0.0245,
+      "logits/chosen": -2.8125,
+      "logits/rejected": -1.046875,
+      "logps/chosen": -1328.0,
+      "logps/rejected": -2192.0,
+      "loss": 0.0217,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.375,
-      "rewards/margins": 7.875,
-      "rewards/rejected": -17.25,
+      "rewards/chosen": -9.4375,
+      "rewards/margins": 8.25,
+      "rewards/rejected": -17.625,
       "step": 2500
     },
     {
       "epoch": 2.6282722513089007,
-      "grad_norm": 0.34034919639058425,
+      "grad_norm": 0.24100837235041658,
       "learning_rate": 2.3031214857028446e-06,
-      "logits/chosen": 2.84375,
-      "logits/rejected": 3.9375,
-      "logps/chosen": -1360.0,
-      "logps/rejected": -2208.0,
-      "loss": 0.0181,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.375,
-      "rewards/margins": 8.125,
-      "rewards/rejected": -17.5,
+      "logits/chosen": -2.28125,
+      "logits/rejected": -1.3515625,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -2272.0,
+      "loss": 0.0202,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -9.6875,
+      "rewards/margins": 8.3125,
+      "rewards/rejected": -18.0,
       "step": 2510
     },
     {
       "epoch": 2.6387434554973823,
-      "grad_norm": 0.4830377701941637,
+      "grad_norm": 0.5098208768156812,
       "learning_rate": 2.177086530257139e-06,
-      "logits/chosen": 1.5234375,
-      "logits/rejected": 3.1875,
-      "logps/chosen": -1320.0,
-      "logps/rejected": -2144.0,
-      "loss": 0.0264,
+      "logits/chosen": -3.015625,
+      "logits/rejected": -1.90625,
+      "logps/chosen": -1368.0,
+      "logps/rejected": -2176.0,
+      "loss": 0.0275,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -9.25,
-      "rewards/margins": 7.90625,
-      "rewards/rejected": -17.125,
+      "rewards/chosen": -9.75,
+      "rewards/margins": 7.71875,
+      "rewards/rejected": -17.375,
       "step": 2520
     },
     {
       "epoch": 2.649214659685864,
-      "grad_norm": 0.3001706451772336,
+      "grad_norm": 0.24060810021367757,
       "learning_rate": 2.054440793781354e-06,
-      "logits/chosen": 1.671875,
-      "logits/rejected": 3.265625,
-      "logps/chosen": -1360.0,
-      "logps/rejected": -2176.0,
-      "loss": 0.0253,
+      "logits/chosen": -2.9375,
+      "logits/rejected": -1.828125,
+      "logps/chosen": -1376.0,
+      "logps/rejected": -2208.0,
+      "loss": 0.0219,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -9.5,
-      "rewards/margins": 7.625,
-      "rewards/rejected": -17.125,
+      "rewards/chosen": -9.6875,
+      "rewards/margins": 7.65625,
+      "rewards/rejected": -17.375,
       "step": 2530
     },
     {
       "epoch": 2.6596858638743455,
-      "grad_norm": 0.3281100282536891,
+      "grad_norm": 0.23265967627043627,
       "learning_rate": 1.9352024892550587e-06,
-      "logits/chosen": 2.015625,
-      "logits/rejected": 3.109375,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -2256.0,
-      "loss": 0.018,
+      "logits/chosen": -2.859375,
+      "logits/rejected": -1.8828125,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -2304.0,
+      "loss": 0.0172,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.6875,
-      "rewards/margins": 8.1875,
-      "rewards/rejected": -17.875,
+      "rewards/chosen": -10.0625,
+      "rewards/margins": 8.125,
+      "rewards/rejected": -18.25,
       "step": 2540
     },
     {
       "epoch": 2.670157068062827,
-      "grad_norm": 0.2115698470981628,
+      "grad_norm": 0.8279062684054203,
       "learning_rate": 1.8193893236517379e-06,
-      "logits/chosen": 1.6640625,
-      "logits/rejected": 3.328125,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -2240.0,
-      "loss": 0.0245,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.6875,
-      "rewards/margins": 8.375,
-      "rewards/rejected": -18.125,
+      "logits/chosen": -2.9375,
+      "logits/rejected": -1.796875,
+      "logps/chosen": -1440.0,
+      "logps/rejected": -2288.0,
+      "loss": 0.0208,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -18.375,
       "step": 2550
     },
     {
       "epoch": 2.680628272251309,
-      "grad_norm": 0.21860141060119193,
+      "grad_norm": 0.6500782804133279,
       "learning_rate": 1.7070184953092956e-06,
-      "logits/chosen": 1.9453125,
-      "logits/rejected": 3.453125,
-      "logps/chosen": -1384.0,
+      "logits/chosen": -2.921875,
+      "logits/rejected": -1.7578125,
+      "logps/chosen": -1408.0,
       "logps/rejected": -2208.0,
-      "loss": 0.0249,
+      "loss": 0.0233,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.6875,
-      "rewards/margins": 7.90625,
+      "rewards/chosen": -9.9375,
+      "rewards/margins": 7.625,
       "rewards/rejected": -17.625,
       "step": 2560
     },
     {
       "epoch": 2.680628272251309,
-      "eval_logits/chosen": 5.71875,
-      "eval_logits/rejected": 7.78125,
-      "eval_logps/chosen": -1600.0,
-      "eval_logps/rejected": -1712.0,
-      "eval_loss": 1.539156198501587,
-      "eval_rewards/accuracies": 0.59375,
-      "eval_rewards/chosen": -12.8125,
-      "eval_rewards/margins": 1.4765625,
-      "eval_rewards/rejected": -14.25,
-      "eval_runtime": 46.545,
-      "eval_samples_per_second": 42.969,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -0.41796875,
+      "eval_logits/rejected": 0.9296875,
+      "eval_logps/chosen": -1640.0,
+      "eval_logps/rejected": -1752.0,
+      "eval_loss": 1.5836249589920044,
+      "eval_rewards/accuracies": 0.603515625,
+      "eval_rewards/chosen": -13.1875,
+      "eval_rewards/margins": 1.4296875,
+      "eval_rewards/rejected": -14.625,
+      "eval_runtime": 58.3865,
+      "eval_samples_per_second": 34.254,
+      "eval_steps_per_second": 0.548,
       "step": 2560
     },
     {
       "epoch": 2.6910994764397906,
-      "grad_norm": 0.7739164436510806,
+      "grad_norm": 0.7171717882095547,
       "learning_rate": 1.5981066913760918e-06,
-      "logits/chosen": 3.296875,
-      "logits/rejected": 4.9375,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -2128.0,
-      "loss": 0.0182,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.1875,
-      "rewards/margins": 7.75,
-      "rewards/rejected": -17.0,
+      "logits/chosen": -1.9921875,
+      "logits/rejected": -0.76171875,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -2160.0,
+      "loss": 0.0157,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -9.5625,
+      "rewards/margins": 7.6875,
+      "rewards/rejected": -17.25,
       "step": 2570
     },
     {
       "epoch": 2.701570680628272,
-      "grad_norm": 0.5935585023275837,
+      "grad_norm": 0.46853039078379227,
       "learning_rate": 1.492670085332884e-06,
-      "logits/chosen": 2.21875,
-      "logits/rejected": 3.59375,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -2288.0,
-      "loss": 0.0219,
+      "logits/chosen": -2.734375,
+      "logits/rejected": -1.6875,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -2352.0,
+      "loss": 0.0256,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.875,
-      "rewards/margins": 8.0,
-      "rewards/rejected": -17.875,
+      "rewards/chosen": -10.1875,
+      "rewards/margins": 8.375,
+      "rewards/rejected": -18.5,
       "step": 2580
     },
     {
       "epoch": 2.712041884816754,
-      "grad_norm": 1.0223669766862518,
+      "grad_norm": 0.6760551204244232,
       "learning_rate": 1.3907243345910787e-06,
-      "logits/chosen": 2.671875,
-      "logits/rejected": 4.0,
-      "logps/chosen": -1400.0,
-      "logps/rejected": -2240.0,
-      "loss": 0.021,
+      "logits/chosen": -2.359375,
+      "logits/rejected": -1.4140625,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -2288.0,
+      "loss": 0.0205,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.5,
+      "rewards/chosen": -10.0,
       "rewards/margins": 8.0625,
-      "rewards/rejected": -17.625,
+      "rewards/rejected": -18.125,
       "step": 2590
     },
     {
       "epoch": 2.7225130890052354,
-      "grad_norm": 0.1909751195246789,
+      "grad_norm": 1.6118442746217418,
       "learning_rate": 1.292284578167538e-06,
-      "logits/chosen": 2.1875,
-      "logits/rejected": 3.40625,
-      "logps/chosen": -1352.0,
-      "logps/rejected": -2256.0,
-      "loss": 0.0131,
+      "logits/chosen": -2.625,
+      "logits/rejected": -1.71875,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -2288.0,
+      "loss": 0.0192,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 8.125,
-      "rewards/rejected": -17.75,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 8.0625,
+      "rewards/rejected": -18.0,
       "step": 2600
     },
     {
       "epoch": 2.7329842931937174,
-      "grad_norm": 0.6010233527725186,
+      "grad_norm": 0.8062253611370775,
       "learning_rate": 1.1973654344364925e-06,
-      "logits/chosen": 2.0,
-      "logits/rejected": 3.078125,
-      "logps/chosen": -1432.0,
-      "logps/rejected": -2304.0,
-      "loss": 0.0176,
+      "logits/chosen": -3.03125,
+      "logits/rejected": -2.140625,
+      "logps/chosen": -1456.0,
+      "logps/rejected": -2320.0,
+      "loss": 0.021,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.0,
-      "rewards/margins": 8.25,
-      "rewards/rejected": -18.25,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 8.0625,
+      "rewards/rejected": -18.375,
       "step": 2610
     },
     {
       "epoch": 2.743455497382199,
-      "grad_norm": 0.7466670604642195,
+      "grad_norm": 0.6802693976981098,
       "learning_rate": 1.1059809989586583e-06,
-      "logits/chosen": 3.1875,
-      "logits/rejected": 3.875,
-      "logps/chosen": -1360.0,
-      "logps/rejected": -2240.0,
-      "loss": 0.022,
+      "logits/chosen": -2.1875,
+      "logits/rejected": -1.515625,
+      "logps/chosen": -1376.0,
+      "logps/rejected": -2256.0,
+      "loss": 0.0203,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.5625,
-      "rewards/margins": 8.125,
-      "rewards/rejected": -17.625,
+      "rewards/chosen": -9.625,
+      "rewards/margins": 8.0625,
+      "rewards/rejected": -17.75,
       "step": 2620
     },
     {
       "epoch": 2.7539267015706805,
-      "grad_norm": 0.46799603748389595,
+      "grad_norm": 0.48627568469172616,
       "learning_rate": 1.018144842388044e-06,
-      "logits/chosen": 3.09375,
-      "logits/rejected": 4.75,
-      "logps/chosen": -1400.0,
+      "logits/chosen": -2.125,
+      "logits/rejected": -0.95703125,
+      "logps/chosen": -1440.0,
       "logps/rejected": -2336.0,
-      "loss": 0.0252,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 8.5625,
+      "loss": 0.0215,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -10.0625,
+      "rewards/margins": 8.25,
       "rewards/rejected": -18.25,
       "step": 2630
     },
     {
       "epoch": 2.7643979057591626,
-      "grad_norm": 0.19630711018583963,
+      "grad_norm": 0.8219088239831188,
       "learning_rate": 9.338700084567109e-07,
-      "logits/chosen": 2.140625,
-      "logits/rejected": 3.546875,
-      "logps/chosen": -1368.0,
+      "logits/chosen": -2.78125,
+      "logits/rejected": -1.765625,
+      "logps/chosen": -1400.0,
       "logps/rejected": -2272.0,
-      "loss": 0.02,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 8.5625,
-      "rewards/rejected": -18.125,
+      "loss": 0.021,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -9.875,
+      "rewards/margins": 8.375,
+      "rewards/rejected": -18.25,
       "step": 2640
     },
     {
       "epoch": 2.7643979057591626,
-      "eval_logits/chosen": 5.9375,
-      "eval_logits/rejected": 7.9375,
-      "eval_logps/chosen": -1624.0,
-      "eval_logps/rejected": -1744.0,
-      "eval_loss": 1.5570937395095825,
-      "eval_rewards/accuracies": 0.58984375,
-      "eval_rewards/chosen": -13.0625,
-      "eval_rewards/margins": 1.4921875,
-      "eval_rewards/rejected": -14.5625,
-      "eval_runtime": 46.5462,
-      "eval_samples_per_second": 42.968,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -0.322265625,
+      "eval_logits/rejected": 1.0078125,
+      "eval_logps/chosen": -1664.0,
+      "eval_logps/rejected": -1776.0,
+      "eval_loss": 1.591703176498413,
+      "eval_rewards/accuracies": 0.609375,
+      "eval_rewards/chosen": -13.4375,
+      "eval_rewards/margins": 1.4609375,
+      "eval_rewards/rejected": -14.9375,
+      "eval_runtime": 58.4088,
+      "eval_samples_per_second": 34.241,
+      "eval_steps_per_second": 0.548,
       "step": 2640
     },
     {
       "epoch": 2.774869109947644,
-      "grad_norm": 0.7982434696988542,
+      "grad_norm": 0.8770217124347667,
       "learning_rate": 8.531690120377606e-07,
-      "logits/chosen": 2.125,
-      "logits/rejected": 3.71875,
-      "logps/chosen": -1536.0,
-      "logps/rejected": -2416.0,
+      "logits/chosen": -2.75,
+      "logits/rejected": -1.5234375,
+      "logps/chosen": -1576.0,
+      "logps/rejected": -2448.0,
       "loss": 0.0215,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.6875,
-      "rewards/margins": 8.625,
-      "rewards/rejected": -19.375,
+      "rewards/chosen": -11.125,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -19.625,
       "step": 2650
     },
     {
       "epoch": 2.7853403141361257,
-      "grad_norm": 0.4079637883313195,
+      "grad_norm": 0.3416868160268321,
       "learning_rate": 7.760538372868636e-07,
-      "logits/chosen": 2.171875,
-      "logits/rejected": 3.4375,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -2272.0,
-      "loss": 0.0217,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -10.0,
-      "rewards/margins": 8.125,
-      "rewards/rejected": -18.125,
+      "logits/chosen": -2.71875,
+      "logits/rejected": -1.7734375,
+      "logps/chosen": -1456.0,
+      "logps/rejected": -2304.0,
+      "loss": 0.0242,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -10.5,
+      "rewards/margins": 8.0,
+      "rewards/rejected": -18.5,
       "step": 2660
     },
     {
       "epoch": 2.7958115183246073,
-      "grad_norm": 0.5970693352578537,
+      "grad_norm": 0.4222954091992161,
       "learning_rate": 7.025359358626166e-07,
-      "logits/chosen": 2.09375,
-      "logits/rejected": 4.125,
-      "logps/chosen": -1488.0,
-      "logps/rejected": -2368.0,
-      "loss": 0.018,
+      "logits/chosen": -2.78125,
+      "logits/rejected": -1.4453125,
+      "logps/chosen": -1504.0,
+      "logps/rejected": -2384.0,
+      "loss": 0.0164,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.4375,
+      "rewards/chosen": -10.625,
       "rewards/margins": 8.4375,
-      "rewards/rejected": -18.875,
+      "rewards/rejected": -19.125,
       "step": 2670
     },
     {
       "epoch": 2.806282722513089,
-      "grad_norm": 1.269195039525123,
+      "grad_norm": 1.1592685502238653,
       "learning_rate": 6.326262252259463e-07,
-      "logits/chosen": 2.3125,
-      "logits/rejected": 3.671875,
-      "logps/chosen": -1432.0,
-      "logps/rejected": -2272.0,
-      "loss": 0.0228,
+      "logits/chosen": -2.5625,
+      "logits/rejected": -1.4765625,
+      "logps/chosen": -1440.0,
+      "logps/rejected": -2304.0,
+      "loss": 0.0261,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -10.0,
-      "rewards/margins": 8.125,
-      "rewards/rejected": -18.125,
+      "rewards/chosen": -10.125,
+      "rewards/margins": 8.3125,
+      "rewards/rejected": -18.375,
       "step": 2680
     },
     {
       "epoch": 2.816753926701571,
-      "grad_norm": 0.7063797264795104,
+      "grad_norm": 0.20377430933875343,
       "learning_rate": 5.66335087018871e-07,
-      "logits/chosen": 1.9609375,
-      "logits/rejected": 3.640625,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -2240.0,
+      "logits/chosen": -2.875,
+      "logits/rejected": -1.6640625,
+      "logps/chosen": -1440.0,
+      "logps/rejected": -2272.0,
       "loss": 0.011,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.75,
-      "rewards/margins": 8.25,
-      "rewards/rejected": -18.0,
+      "rewards/chosen": -10.125,
+      "rewards/margins": 8.125,
+      "rewards/rejected": -18.25,
       "step": 2690
     },
     {
       "epoch": 2.8272251308900525,
-      "grad_norm": 0.14570271564913057,
+      "grad_norm": 0.5578341695611294,
       "learning_rate": 5.036723655228226e-07,
-      "logits/chosen": 2.453125,
-      "logits/rejected": 3.453125,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -2192.0,
-      "loss": 0.0219,
+      "logits/chosen": -2.65625,
+      "logits/rejected": -1.796875,
+      "logps/chosen": -1368.0,
+      "logps/rejected": -2224.0,
+      "loss": 0.0234,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.6875,
-      "rewards/margins": 7.9375,
-      "rewards/rejected": -17.625,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 8.0,
+      "rewards/rejected": -18.0,
       "step": 2700
     },
     {
       "epoch": 2.837696335078534,
-      "grad_norm": 0.6246283290652587,
+      "grad_norm": 0.3194835286479539,
       "learning_rate": 4.446473661967432e-07,
-      "logits/chosen": 2.265625,
-      "logits/rejected": 4.25,
-      "logps/chosen": -1400.0,
-      "logps/rejected": -2288.0,
-      "loss": 0.0204,
+      "logits/chosen": -2.6875,
+      "logits/rejected": -1.2109375,
+      "logps/chosen": -1440.0,
+      "logps/rejected": -2336.0,
+      "loss": 0.0246,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.875,
-      "rewards/margins": 8.375,
-      "rewards/rejected": -18.25,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 8.4375,
+      "rewards/rejected": -18.75,
       "step": 2710
     },
     {
       "epoch": 2.8481675392670156,
-      "grad_norm": 1.0809276873421974,
+      "grad_norm": 0.5535851352363035,
       "learning_rate": 3.892688542952411e-07,
-      "logits/chosen": 3.125,
-      "logits/rejected": 4.09375,
-      "logps/chosen": -1352.0,
-      "logps/rejected": -2208.0,
-      "loss": 0.0194,
+      "logits/chosen": -2.203125,
+      "logits/rejected": -1.3984375,
+      "logps/chosen": -1384.0,
+      "logps/rejected": -2224.0,
+      "loss": 0.0221,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 7.96875,
-      "rewards/rejected": -17.5,
+      "rewards/chosen": -9.875,
+      "rewards/margins": 7.71875,
+      "rewards/rejected": -17.625,
       "step": 2720
     },
     {
       "epoch": 2.8481675392670156,
-      "eval_logits/chosen": 6.03125,
-      "eval_logits/rejected": 8.0625,
-      "eval_logps/chosen": -1640.0,
-      "eval_logps/rejected": -1760.0,
-      "eval_loss": 1.5687812566757202,
-      "eval_rewards/accuracies": 0.58984375,
-      "eval_rewards/chosen": -13.1875,
-      "eval_rewards/margins": 1.5,
-      "eval_rewards/rejected": -14.6875,
-      "eval_runtime": 46.5497,
-      "eval_samples_per_second": 42.965,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -0.279296875,
+      "eval_logits/rejected": 1.046875,
+      "eval_logps/chosen": -1672.0,
+      "eval_logps/rejected": -1792.0,
+      "eval_loss": 1.6077030897140503,
+      "eval_rewards/accuracies": 0.603515625,
+      "eval_rewards/chosen": -13.5625,
+      "eval_rewards/margins": 1.4609375,
+      "eval_rewards/rejected": -15.0,
+      "eval_runtime": 58.4109,
+      "eval_samples_per_second": 34.24,
+      "eval_steps_per_second": 0.548,
       "step": 2720
     },
     {
       "epoch": 2.858638743455497,
-      "grad_norm": 1.2319422767300263,
+      "grad_norm": 4.624752075558184,
       "learning_rate": 3.3754505356694e-07,
-      "logits/chosen": 2.203125,
-      "logits/rejected": 3.3125,
+      "logits/chosen": -2.734375,
+      "logits/rejected": -1.8203125,
       "logps/chosen": -1456.0,
-      "logps/rejected": -2320.0,
-      "loss": 0.0203,
+      "logps/rejected": -2352.0,
+      "loss": 0.023,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.375,
-      "rewards/margins": 8.25,
-      "rewards/rejected": -18.625,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 8.5625,
+      "rewards/rejected": -18.875,
       "step": 2730
     },
     {
       "epoch": 2.869109947643979,
-      "grad_norm": 0.9262411571443954,
+      "grad_norm": 0.6922682830403168,
       "learning_rate": 2.8948364503322276e-07,
-      "logits/chosen": 2.46875,
-      "logits/rejected": 4.25,
-      "logps/chosen": -1368.0,
+      "logits/chosen": -2.6875,
+      "logits/rejected": -1.3125,
+      "logps/chosen": -1384.0,
       "logps/rejected": -2240.0,
-      "loss": 0.0252,
+      "loss": 0.0253,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.75,
-      "rewards/margins": 8.375,
+      "rewards/chosen": -9.9375,
+      "rewards/margins": 8.1875,
       "rewards/rejected": -18.125,
       "step": 2740
     },
     {
       "epoch": 2.8795811518324608,
-      "grad_norm": 0.5234867120285107,
+      "grad_norm": 0.2614533068110169,
       "learning_rate": 2.4509176584762706e-07,
-      "logits/chosen": 2.671875,
-      "logits/rejected": 4.4375,
-      "logps/chosen": -1368.0,
-      "logps/rejected": -2176.0,
-      "loss": 0.0171,
+      "logits/chosen": -2.4375,
+      "logits/rejected": -1.1484375,
+      "logps/chosen": -1400.0,
+      "logps/rejected": -2192.0,
+      "loss": 0.0162,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.6875,
-      "rewards/margins": 7.78125,
-      "rewards/rejected": -17.375,
+      "rewards/chosen": -9.9375,
+      "rewards/margins": 7.59375,
+      "rewards/rejected": -17.5,
       "step": 2750
     },
     {
       "epoch": 2.8900523560209423,
-      "grad_norm": 0.5506246909833438,
+      "grad_norm": 0.4866024472444901,
       "learning_rate": 2.0437600823595694e-07,
-      "logits/chosen": 1.8125,
-      "logits/rejected": 3.859375,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -2224.0,
-      "loss": 0.0205,
+      "logits/chosen": -3.109375,
+      "logits/rejected": -1.59375,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -2256.0,
+      "loss": 0.0222,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.6875,
+      "rewards/chosen": -9.875,
       "rewards/margins": 8.0,
-      "rewards/rejected": -17.625,
+      "rewards/rejected": -17.875,
       "step": 2760
     },
     {
       "epoch": 2.900523560209424,
-      "grad_norm": 1.7066608708074689,
+      "grad_norm": 1.0818670602133909,
       "learning_rate": 1.6734241851733e-07,
-      "logits/chosen": 2.234375,
-      "logits/rejected": 3.21875,
-      "logps/chosen": -1504.0,
-      "logps/rejected": -2400.0,
-      "loss": 0.0224,
+      "logits/chosen": -2.515625,
+      "logits/rejected": -1.796875,
+      "logps/chosen": -1552.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.0214,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.625,
-      "rewards/margins": 8.25,
-      "rewards/rejected": -18.875,
+      "rewards/chosen": -11.0,
+      "rewards/margins": 8.0,
+      "rewards/rejected": -19.0,
       "step": 2770
     },
     {
       "epoch": 2.9109947643979055,
-      "grad_norm": 0.9362487737833018,
+      "grad_norm": 0.596416979638384,
       "learning_rate": 1.339964962062984e-07,
-      "logits/chosen": 1.90625,
-      "logits/rejected": 2.78125,
-      "logps/chosen": -1352.0,
-      "logps/rejected": -2288.0,
-      "loss": 0.0193,
+      "logits/chosen": -2.875,
+      "logits/rejected": -2.15625,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -2320.0,
+      "loss": 0.0201,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.75,
-      "rewards/margins": 8.5625,
-      "rewards/rejected": -18.25,
+      "rewards/chosen": -10.125,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -18.625,
       "step": 2780
     },
     {
       "epoch": 2.9214659685863875,
-      "grad_norm": 0.23617984510661233,
+      "grad_norm": 0.7483209911533596,
       "learning_rate": 1.0434319319617136e-07,
-      "logits/chosen": 2.0625,
-      "logits/rejected": 4.09375,
-      "logps/chosen": -1360.0,
-      "logps/rejected": -2256.0,
-      "loss": 0.0209,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.5625,
-      "rewards/margins": 8.625,
-      "rewards/rejected": -18.25,
+      "logits/chosen": -2.734375,
+      "logits/rejected": -1.328125,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -2288.0,
+      "loss": 0.0241,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -10.1875,
+      "rewards/margins": 8.3125,
+      "rewards/rejected": -18.5,
       "step": 2790
     },
     {
       "epoch": 2.931937172774869,
-      "grad_norm": 0.12596497828930828,
+      "grad_norm": 0.3184435737137817,
       "learning_rate": 7.838691302365363e-08,
-      "logits/chosen": 2.046875,
-      "logits/rejected": 3.609375,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -2272.0,
-      "loss": 0.0163,
+      "logits/chosen": -2.640625,
+      "logits/rejected": -1.515625,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -2304.0,
+      "loss": 0.0182,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.875,
-      "rewards/margins": 8.1875,
-      "rewards/rejected": -18.125,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 8.25,
+      "rewards/rejected": -18.5,
       "step": 2800
     },
     {
       "epoch": 2.931937172774869,
-      "eval_logits/chosen": 6.03125,
-      "eval_logits/rejected": 8.0,
-      "eval_logps/chosen": -1640.0,
-      "eval_logps/rejected": -1760.0,
-      "eval_loss": 1.5703437328338623,
-      "eval_rewards/accuracies": 0.58984375,
-      "eval_rewards/chosen": -13.1875,
-      "eval_rewards/margins": 1.5,
-      "eval_rewards/rejected": -14.6875,
-      "eval_runtime": 46.5112,
-      "eval_samples_per_second": 43.0,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -0.26953125,
+      "eval_logits/rejected": 1.0625,
+      "eval_logps/chosen": -1680.0,
+      "eval_logps/rejected": -1800.0,
+      "eval_loss": 1.619046926498413,
+      "eval_rewards/accuracies": 0.599609375,
+      "eval_rewards/chosen": -13.625,
+      "eval_rewards/margins": 1.46875,
+      "eval_rewards/rejected": -15.0625,
+      "eval_runtime": 58.4474,
+      "eval_samples_per_second": 34.219,
+      "eval_steps_per_second": 0.548,
       "step": 2800
     },
     {
       "epoch": 2.9424083769633507,
-      "grad_norm": 0.24650208252783026,
+      "grad_norm": 0.3178605488558762,
       "learning_rate": 5.613151021490459e-08,
-      "logits/chosen": 2.96875,
-      "logits/rejected": 4.21875,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -2144.0,
-      "loss": 0.0169,
+      "logits/chosen": -2.15625,
+      "logits/rejected": -1.1796875,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -2208.0,
+      "loss": 0.022,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.6875,
-      "rewards/margins": 7.9375,
-      "rewards/rejected": -17.625,
+      "rewards/chosen": -10.125,
+      "rewards/margins": 8.0,
+      "rewards/rejected": -18.125,
       "step": 2810
     },
     {
       "epoch": 2.9528795811518327,
-      "grad_norm": 1.7673911296436624,
+      "grad_norm": 0.3067095117894361,
       "learning_rate": 3.7580289713159965e-08,
-      "logits/chosen": 3.171875,
-      "logits/rejected": 4.53125,
-      "logps/chosen": -1448.0,
-      "logps/rejected": -2432.0,
-      "loss": 0.0196,
+      "logits/chosen": -2.125,
+      "logits/rejected": -1.1171875,
+      "logps/chosen": -1480.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.0161,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.25,
+      "rewards/chosen": -10.5625,
       "rewards/margins": 8.8125,
-      "rewards/rejected": -19.125,
+      "rewards/rejected": -19.375,
       "step": 2820
     },
     {
       "epoch": 2.9633507853403143,
-      "grad_norm": 0.45031960083606704,
+      "grad_norm": 0.9782976608525737,
       "learning_rate": 2.2736006387927277e-08,
-      "logits/chosen": 2.25,
-      "logits/rejected": 3.890625,
-      "logps/chosen": -1440.0,
-      "logps/rejected": -2288.0,
-      "loss": 0.0177,
+      "logits/chosen": -2.640625,
+      "logits/rejected": -1.515625,
+      "logps/chosen": -1456.0,
+      "logps/rejected": -2336.0,
+      "loss": 0.0198,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.125,
-      "rewards/margins": 8.25,
-      "rewards/rejected": -18.375,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 8.5625,
+      "rewards/rejected": -18.875,
       "step": 2830
     },
     {
       "epoch": 2.973821989528796,
-      "grad_norm": 0.29120827619366724,
+      "grad_norm": 0.5545881251789558,
       "learning_rate": 1.1600864625893604e-08,
-      "logits/chosen": 1.9609375,
-      "logits/rejected": 3.234375,
-      "logps/chosen": -1416.0,
-      "logps/rejected": -2272.0,
-      "loss": 0.0291,
+      "logits/chosen": -2.90625,
+      "logits/rejected": -1.9140625,
+      "logps/chosen": -1456.0,
+      "logps/rejected": -2320.0,
+      "loss": 0.0279,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.9375,
+      "rewards/chosen": -10.4375,
       "rewards/margins": 8.25,
-      "rewards/rejected": -18.25,
+      "rewards/rejected": -18.625,
       "step": 2840
     },
     {
       "epoch": 2.9842931937172774,
-      "grad_norm": 0.4964477815174809,
+      "grad_norm": 0.2234873699250944,
       "learning_rate": 4.176518003573549e-09,
-      "logits/chosen": 2.421875,
-      "logits/rejected": 4.28125,
-      "logps/chosen": -1424.0,
-      "logps/rejected": -2240.0,
-      "loss": 0.0283,
+      "logits/chosen": -2.640625,
+      "logits/rejected": -1.25,
+      "logps/chosen": -1456.0,
+      "logps/rejected": -2304.0,
+      "loss": 0.0297,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.3125,
-      "rewards/margins": 7.46875,
-      "rewards/rejected": -17.75,
+      "rewards/chosen": -10.5625,
+      "rewards/margins": 7.8125,
+      "rewards/rejected": -18.375,
       "step": 2850
     },
     {
       "epoch": 2.994764397905759,
-      "grad_norm": 0.43937926995656784,
+      "grad_norm": 0.30254200867746056,
       "learning_rate": 4.640690417528903e-10,
-      "logits/chosen": 2.171875,
-      "logits/rejected": 3.71875,
-      "logps/chosen": -1344.0,
+      "logits/chosen": -2.796875,
+      "logits/rejected": -1.6953125,
+      "logps/chosen": -1376.0,
       "logps/rejected": -2192.0,
-      "loss": 0.0266,
+      "loss": 0.0243,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 8.125,
+      "rewards/chosen": -9.9375,
+      "rewards/margins": 7.8125,
       "rewards/rejected": -17.75,
       "step": 2860
     },
@@ -4877,10 +4877,10 @@
       "epoch": 3.0,
       "step": 2865,
       "total_flos": 0.0,
-      "train_loss": 0.27230105622692674,
-      "train_runtime": 12616.2292,
-      "train_samples_per_second": 14.533,
-      "train_steps_per_second": 0.227
+      "train_loss": 0.2725709742172434,
+      "train_runtime": 12848.3898,
+      "train_samples_per_second": 14.271,
+      "train_steps_per_second": 0.223
     }
   ],
   "logging_steps": 10,