| program(1.3) | |
| [buildInfo = dict<string, string>({{"coremlc-component-MIL", "3400.43.1"}, {"coremlc-version", "3400.58.2"}, {"coremltools-component-torch", "2.4.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] | |
| { | |
| func main<ios18>(tensor<fp16, [1, 80, 3000]> logmel_data) { | |
| string var_68_pad_type_0 = const()[name = string("op_68_pad_type_0"), val = string("custom")]; | |
| tensor<int32, [2]> var_68_pad_0 = const()[name = string("op_68_pad_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [1]> var_68_strides_0 = const()[name = string("op_68_strides_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<int32, [1]> var_68_dilations_0 = const()[name = string("op_68_dilations_0"), val = tensor<int32, [1]>([1])]; | |
| int32 var_68_groups_0 = const()[name = string("op_68_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1024, 80, 3]> weight_3_to_fp16 = const()[name = string("weight_3_to_fp16"), val = tensor<fp16, [1024, 80, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; | |
| tensor<fp16, [1024]> bias_3_to_fp16 = const()[name = string("bias_3_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(491648)))]; | |
| tensor<fp16, [1, 1024, 3000]> var_68_cast_fp16 = conv(bias = bias_3_to_fp16, dilations = var_68_dilations_0, groups = var_68_groups_0, pad = var_68_pad_0, pad_type = var_68_pad_type_0, strides = var_68_strides_0, weight = weight_3_to_fp16, x = logmel_data)[name = string("op_68_cast_fp16")]; | |
| string input_1_mode_0 = const()[name = string("input_1_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1024, 3000]> input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_68_cast_fp16)[name = string("input_1_cast_fp16")]; | |
| string var_86_pad_type_0 = const()[name = string("op_86_pad_type_0"), val = string("custom")]; | |
| tensor<int32, [2]> var_86_pad_0 = const()[name = string("op_86_pad_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [1]> var_86_strides_0 = const()[name = string("op_86_strides_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<int32, [1]> var_86_dilations_0 = const()[name = string("op_86_dilations_0"), val = tensor<int32, [1]>([1])]; | |
| int32 var_86_groups_0 = const()[name = string("op_86_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1024, 1024, 3]> weight_7_to_fp16 = const()[name = string("weight_7_to_fp16"), val = tensor<fp16, [1024, 1024, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(493760)))]; | |
| tensor<fp16, [1024]> bias_7_to_fp16 = const()[name = string("bias_7_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6785280)))]; | |
| tensor<fp16, [1, 1024, 1500]> var_86_cast_fp16 = conv(bias = bias_7_to_fp16, dilations = var_86_dilations_0, groups = var_86_groups_0, pad = var_86_pad_0, pad_type = var_86_pad_type_0, strides = var_86_strides_0, weight = weight_7_to_fp16, x = input_1_cast_fp16)[name = string("op_86_cast_fp16")]; | |
| string x_3_mode_0 = const()[name = string("x_3_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1024, 1500]> x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_86_cast_fp16)[name = string("x_3_cast_fp16")]; | |
| tensor<int32, [3]> var_92 = const()[name = string("op_92"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<fp16, [1500, 1024]> positional_embedding_to_fp16 = const()[name = string("positional_embedding_to_fp16"), val = tensor<fp16, [1500, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6787392)))]; | |
| tensor<fp16, [1, 1500, 1024]> x_5_cast_fp16 = transpose(perm = var_92, x = x_3_cast_fp16)[name = string("transpose_240")]; | |
| tensor<fp16, [1, 1500, 1024]> var_95_cast_fp16 = add(x = x_5_cast_fp16, y = positional_embedding_to_fp16)[name = string("op_95_cast_fp16")]; | |
| int32 var_108 = const()[name = string("op_108"), val = int32(-1)]; | |
| tensor<int32, [1]> var_124_axes_0 = const()[name = string("op_124_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_0_attn_ln_weight_to_fp16 = const()[name = string("blocks_0_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9859456)))]; | |
| tensor<fp16, [1024]> blocks_0_attn_ln_bias_to_fp16 = const()[name = string("blocks_0_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9861568)))]; | |
| fp16 var_114_to_fp16 = const()[name = string("op_114_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_124_cast_fp16 = layer_norm(axes = var_124_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_114_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_95_cast_fp16)[name = string("op_124_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_135_to_fp16 = const()[name = string("op_135_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9863680)))]; | |
| tensor<fp16, [1024]> var_136_to_fp16 = const()[name = string("op_136_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11960896)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_0_cast_fp16 = linear(bias = var_136_to_fp16, weight = var_135_to_fp16, x = var_124_cast_fp16)[name = string("linear_0_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_139_to_fp16 = const()[name = string("op_139_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11963008)))]; | |
| tensor<fp16, [1024]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14060224)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_139_to_fp16, x = var_124_cast_fp16)[name = string("linear_1_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_143_to_fp16 = const()[name = string("op_143_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14062336)))]; | |
| tensor<fp16, [1024]> var_144_to_fp16 = const()[name = string("op_144_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16159552)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_2_cast_fp16 = linear(bias = var_144_to_fp16, weight = var_143_to_fp16, x = var_124_cast_fp16)[name = string("linear_2_cast_fp16")]; | |
| tensor<int32, [4]> var_152 = const()[name = string("op_152"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_153_cast_fp16 = reshape(shape = var_152, x = linear_0_cast_fp16)[name = string("op_153_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_168_to_fp16 = const()[name = string("const_168_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_3_cast_fp16 = mul(x = var_153_cast_fp16, y = const_168_to_fp16)[name = string("q_3_cast_fp16")]; | |
| tensor<int32, [4]> var_159 = const()[name = string("op_159"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_160_cast_fp16 = reshape(shape = var_159, x = linear_1_cast_fp16)[name = string("op_160_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_169_to_fp16 = const()[name = string("const_169_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_3_cast_fp16 = mul(x = var_160_cast_fp16, y = const_169_to_fp16)[name = string("k_3_cast_fp16")]; | |
| tensor<int32, [4]> var_166 = const()[name = string("op_166"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_167_cast_fp16 = reshape(shape = var_166, x = linear_2_cast_fp16)[name = string("op_167_cast_fp16")]; | |
| tensor<int32, [4]> var_168 = const()[name = string("op_168"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_1_transpose_x_0 = const()[name = string("qk_1_transpose_x_0"), val = bool(false)]; | |
| bool qk_1_transpose_y_0 = const()[name = string("qk_1_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_96_perm_0 = const()[name = string("transpose_96_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_97_perm_0 = const()[name = string("transpose_97_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_97 = transpose(perm = transpose_97_perm_0, x = k_3_cast_fp16)[name = string("transpose_237")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_96 = transpose(perm = transpose_96_perm_0, x = q_3_cast_fp16)[name = string("transpose_238")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_1_cast_fp16 = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_96, y = transpose_97)[name = string("qk_1_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_172_cast_fp16 = softmax(axis = var_108, x = qk_1_cast_fp16)[name = string("op_172_cast_fp16")]; | |
| bool var_174_transpose_x_0 = const()[name = string("op_174_transpose_x_0"), val = bool(false)]; | |
| bool var_174_transpose_y_0 = const()[name = string("op_174_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_3_cast_fp16 = transpose(perm = var_168, x = var_167_cast_fp16)[name = string("transpose_239")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_174_cast_fp16 = matmul(transpose_x = var_174_transpose_x_0, transpose_y = var_174_transpose_y_0, x = var_172_cast_fp16, y = v_3_cast_fp16)[name = string("op_174_cast_fp16")]; | |
| tensor<int32, [4]> var_175 = const()[name = string("op_175"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_0 = const()[name = string("concat_0"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_176_cast_fp16 = transpose(perm = var_175, x = var_174_cast_fp16)[name = string("transpose_236")]; | |
| tensor<fp16, [1, 1500, 1024]> x_11_cast_fp16 = reshape(shape = concat_0, x = var_176_cast_fp16)[name = string("x_11_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_180_to_fp16 = const()[name = string("op_180_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16161664)))]; | |
| tensor<fp16, [1024]> var_181_to_fp16 = const()[name = string("op_181_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18258880)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_3_cast_fp16 = linear(bias = var_181_to_fp16, weight = var_180_to_fp16, x = x_11_cast_fp16)[name = string("linear_3_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_13_cast_fp16 = add(x = var_95_cast_fp16, y = linear_3_cast_fp16)[name = string("x_13_cast_fp16")]; | |
| tensor<int32, [1]> var_188_axes_0 = const()[name = string("op_188_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_0_mlp_ln_weight_to_fp16 = const()[name = string("blocks_0_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18260992)))]; | |
| tensor<fp16, [1024]> blocks_0_mlp_ln_bias_to_fp16 = const()[name = string("blocks_0_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18263104)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_188_cast_fp16 = layer_norm(axes = var_188_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_114_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = string("op_188_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_197_to_fp16 = const()[name = string("op_197_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18265216)))]; | |
| tensor<fp16, [4096]> var_198_to_fp16 = const()[name = string("op_198_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26653888)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_4_cast_fp16 = linear(bias = var_198_to_fp16, weight = var_197_to_fp16, x = var_188_cast_fp16)[name = string("linear_4_cast_fp16")]; | |
| string x_17_mode_0 = const()[name = string("x_17_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = string("x_17_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_203_to_fp16 = const()[name = string("op_203_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26662144)))]; | |
| tensor<fp16, [1024]> var_204_to_fp16 = const()[name = string("op_204_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35050816)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_5_cast_fp16 = linear(bias = var_204_to_fp16, weight = var_203_to_fp16, x = x_17_cast_fp16)[name = string("linear_5_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = string("x_19_cast_fp16")]; | |
| int32 var_214 = const()[name = string("op_214"), val = int32(-1)]; | |
| tensor<int32, [1]> var_230_axes_0 = const()[name = string("op_230_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_1_attn_ln_weight_to_fp16 = const()[name = string("blocks_1_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35052928)))]; | |
| tensor<fp16, [1024]> blocks_1_attn_ln_bias_to_fp16 = const()[name = string("blocks_1_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35055040)))]; | |
| fp16 var_220_to_fp16 = const()[name = string("op_220_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_230_cast_fp16 = layer_norm(axes = var_230_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_220_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = string("op_230_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_241_to_fp16 = const()[name = string("op_241_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35057152)))]; | |
| tensor<fp16, [1024]> var_242_to_fp16 = const()[name = string("op_242_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37154368)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_6_cast_fp16 = linear(bias = var_242_to_fp16, weight = var_241_to_fp16, x = var_230_cast_fp16)[name = string("linear_6_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_245_to_fp16 = const()[name = string("op_245_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37156480)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_245_to_fp16, x = var_230_cast_fp16)[name = string("linear_7_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_249_to_fp16 = const()[name = string("op_249_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39253696)))]; | |
| tensor<fp16, [1024]> var_250_to_fp16 = const()[name = string("op_250_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41350912)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_8_cast_fp16 = linear(bias = var_250_to_fp16, weight = var_249_to_fp16, x = var_230_cast_fp16)[name = string("linear_8_cast_fp16")]; | |
| tensor<int32, [4]> var_258 = const()[name = string("op_258"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_259_cast_fp16 = reshape(shape = var_258, x = linear_6_cast_fp16)[name = string("op_259_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_170_to_fp16 = const()[name = string("const_170_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_7_cast_fp16 = mul(x = var_259_cast_fp16, y = const_170_to_fp16)[name = string("q_7_cast_fp16")]; | |
| tensor<int32, [4]> var_265 = const()[name = string("op_265"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_266_cast_fp16 = reshape(shape = var_265, x = linear_7_cast_fp16)[name = string("op_266_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_171_to_fp16 = const()[name = string("const_171_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_7_cast_fp16 = mul(x = var_266_cast_fp16, y = const_171_to_fp16)[name = string("k_7_cast_fp16")]; | |
| tensor<int32, [4]> var_272 = const()[name = string("op_272"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_273_cast_fp16 = reshape(shape = var_272, x = linear_8_cast_fp16)[name = string("op_273_cast_fp16")]; | |
| tensor<int32, [4]> var_274 = const()[name = string("op_274"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_3_transpose_x_0 = const()[name = string("qk_3_transpose_x_0"), val = bool(false)]; | |
| bool qk_3_transpose_y_0 = const()[name = string("qk_3_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_98_perm_0 = const()[name = string("transpose_98_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_99_perm_0 = const()[name = string("transpose_99_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_99 = transpose(perm = transpose_99_perm_0, x = k_7_cast_fp16)[name = string("transpose_233")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_98 = transpose(perm = transpose_98_perm_0, x = q_7_cast_fp16)[name = string("transpose_234")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_3_cast_fp16 = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_98, y = transpose_99)[name = string("qk_3_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_278_cast_fp16 = softmax(axis = var_214, x = qk_3_cast_fp16)[name = string("op_278_cast_fp16")]; | |
| bool var_280_transpose_x_0 = const()[name = string("op_280_transpose_x_0"), val = bool(false)]; | |
| bool var_280_transpose_y_0 = const()[name = string("op_280_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_7_cast_fp16 = transpose(perm = var_274, x = var_273_cast_fp16)[name = string("transpose_235")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_280_cast_fp16 = matmul(transpose_x = var_280_transpose_x_0, transpose_y = var_280_transpose_y_0, x = var_278_cast_fp16, y = v_7_cast_fp16)[name = string("op_280_cast_fp16")]; | |
| tensor<int32, [4]> var_281 = const()[name = string("op_281"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_1 = const()[name = string("concat_1"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_282_cast_fp16 = transpose(perm = var_281, x = var_280_cast_fp16)[name = string("transpose_232")]; | |
| tensor<fp16, [1, 1500, 1024]> x_23_cast_fp16 = reshape(shape = concat_1, x = var_282_cast_fp16)[name = string("x_23_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_286_to_fp16 = const()[name = string("op_286_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41353024)))]; | |
| tensor<fp16, [1024]> var_287_to_fp16 = const()[name = string("op_287_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43450240)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_9_cast_fp16 = linear(bias = var_287_to_fp16, weight = var_286_to_fp16, x = x_23_cast_fp16)[name = string("linear_9_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = string("x_25_cast_fp16")]; | |
| tensor<int32, [1]> var_294_axes_0 = const()[name = string("op_294_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_1_mlp_ln_weight_to_fp16 = const()[name = string("blocks_1_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43452352)))]; | |
| tensor<fp16, [1024]> blocks_1_mlp_ln_bias_to_fp16 = const()[name = string("blocks_1_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43454464)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_294_cast_fp16 = layer_norm(axes = var_294_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_220_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = string("op_294_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_303_to_fp16 = const()[name = string("op_303_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43456576)))]; | |
| tensor<fp16, [4096]> var_304_to_fp16 = const()[name = string("op_304_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51845248)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_10_cast_fp16 = linear(bias = var_304_to_fp16, weight = var_303_to_fp16, x = var_294_cast_fp16)[name = string("linear_10_cast_fp16")]; | |
| string x_29_mode_0 = const()[name = string("x_29_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = string("x_29_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_309_to_fp16 = const()[name = string("op_309_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51853504)))]; | |
| tensor<fp16, [1024]> var_310_to_fp16 = const()[name = string("op_310_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60242176)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_11_cast_fp16 = linear(bias = var_310_to_fp16, weight = var_309_to_fp16, x = x_29_cast_fp16)[name = string("linear_11_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = string("x_31_cast_fp16")]; | |
| int32 var_320 = const()[name = string("op_320"), val = int32(-1)]; | |
| tensor<int32, [1]> var_336_axes_0 = const()[name = string("op_336_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_2_attn_ln_weight_to_fp16 = const()[name = string("blocks_2_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60244288)))]; | |
| tensor<fp16, [1024]> blocks_2_attn_ln_bias_to_fp16 = const()[name = string("blocks_2_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60246400)))]; | |
| fp16 var_326_to_fp16 = const()[name = string("op_326_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_336_cast_fp16 = layer_norm(axes = var_336_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_326_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = string("op_336_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_347_to_fp16 = const()[name = string("op_347_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60248512)))]; | |
| tensor<fp16, [1024]> var_348_to_fp16 = const()[name = string("op_348_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62345728)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_12_cast_fp16 = linear(bias = var_348_to_fp16, weight = var_347_to_fp16, x = var_336_cast_fp16)[name = string("linear_12_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_351_to_fp16 = const()[name = string("op_351_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62347840)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_351_to_fp16, x = var_336_cast_fp16)[name = string("linear_13_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_355_to_fp16 = const()[name = string("op_355_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64445056)))]; | |
| tensor<fp16, [1024]> var_356_to_fp16 = const()[name = string("op_356_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66542272)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_14_cast_fp16 = linear(bias = var_356_to_fp16, weight = var_355_to_fp16, x = var_336_cast_fp16)[name = string("linear_14_cast_fp16")]; | |
| tensor<int32, [4]> var_364 = const()[name = string("op_364"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_365_cast_fp16 = reshape(shape = var_364, x = linear_12_cast_fp16)[name = string("op_365_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_172_to_fp16 = const()[name = string("const_172_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_11_cast_fp16 = mul(x = var_365_cast_fp16, y = const_172_to_fp16)[name = string("q_11_cast_fp16")]; | |
| tensor<int32, [4]> var_371 = const()[name = string("op_371"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_372_cast_fp16 = reshape(shape = var_371, x = linear_13_cast_fp16)[name = string("op_372_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_173_to_fp16 = const()[name = string("const_173_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_11_cast_fp16 = mul(x = var_372_cast_fp16, y = const_173_to_fp16)[name = string("k_11_cast_fp16")]; | |
| tensor<int32, [4]> var_378 = const()[name = string("op_378"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_379_cast_fp16 = reshape(shape = var_378, x = linear_14_cast_fp16)[name = string("op_379_cast_fp16")]; | |
| tensor<int32, [4]> var_380 = const()[name = string("op_380"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_5_transpose_x_0 = const()[name = string("qk_5_transpose_x_0"), val = bool(false)]; | |
| bool qk_5_transpose_y_0 = const()[name = string("qk_5_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_100_perm_0 = const()[name = string("transpose_100_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_101_perm_0 = const()[name = string("transpose_101_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_101 = transpose(perm = transpose_101_perm_0, x = k_11_cast_fp16)[name = string("transpose_229")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_100 = transpose(perm = transpose_100_perm_0, x = q_11_cast_fp16)[name = string("transpose_230")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_5_cast_fp16 = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_100, y = transpose_101)[name = string("qk_5_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_384_cast_fp16 = softmax(axis = var_320, x = qk_5_cast_fp16)[name = string("op_384_cast_fp16")]; | |
| bool var_386_transpose_x_0 = const()[name = string("op_386_transpose_x_0"), val = bool(false)]; | |
| bool var_386_transpose_y_0 = const()[name = string("op_386_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_11_cast_fp16 = transpose(perm = var_380, x = var_379_cast_fp16)[name = string("transpose_231")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_386_cast_fp16 = matmul(transpose_x = var_386_transpose_x_0, transpose_y = var_386_transpose_y_0, x = var_384_cast_fp16, y = v_11_cast_fp16)[name = string("op_386_cast_fp16")]; | |
| tensor<int32, [4]> var_387 = const()[name = string("op_387"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_2 = const()[name = string("concat_2"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_388_cast_fp16 = transpose(perm = var_387, x = var_386_cast_fp16)[name = string("transpose_228")]; | |
| tensor<fp16, [1, 1500, 1024]> x_35_cast_fp16 = reshape(shape = concat_2, x = var_388_cast_fp16)[name = string("x_35_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_392_to_fp16 = const()[name = string("op_392_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66544384)))]; | |
| tensor<fp16, [1024]> var_393_to_fp16 = const()[name = string("op_393_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68641600)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_15_cast_fp16 = linear(bias = var_393_to_fp16, weight = var_392_to_fp16, x = x_35_cast_fp16)[name = string("linear_15_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = string("x_37_cast_fp16")]; | |
| tensor<int32, [1]> var_400_axes_0 = const()[name = string("op_400_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_2_mlp_ln_weight_to_fp16 = const()[name = string("blocks_2_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68643712)))]; | |
| tensor<fp16, [1024]> blocks_2_mlp_ln_bias_to_fp16 = const()[name = string("blocks_2_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68645824)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_400_cast_fp16 = layer_norm(axes = var_400_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_326_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = string("op_400_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_409_to_fp16 = const()[name = string("op_409_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68647936)))]; | |
| tensor<fp16, [4096]> var_410_to_fp16 = const()[name = string("op_410_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77036608)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_16_cast_fp16 = linear(bias = var_410_to_fp16, weight = var_409_to_fp16, x = var_400_cast_fp16)[name = string("linear_16_cast_fp16")]; | |
| string x_41_mode_0 = const()[name = string("x_41_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = string("x_41_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_415_to_fp16 = const()[name = string("op_415_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77044864)))]; | |
| tensor<fp16, [1024]> var_416_to_fp16 = const()[name = string("op_416_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85433536)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_17_cast_fp16 = linear(bias = var_416_to_fp16, weight = var_415_to_fp16, x = x_41_cast_fp16)[name = string("linear_17_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = string("x_43_cast_fp16")]; | |
| int32 var_426 = const()[name = string("op_426"), val = int32(-1)]; | |
| tensor<int32, [1]> var_442_axes_0 = const()[name = string("op_442_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_3_attn_ln_weight_to_fp16 = const()[name = string("blocks_3_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85435648)))]; | |
| tensor<fp16, [1024]> blocks_3_attn_ln_bias_to_fp16 = const()[name = string("blocks_3_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85437760)))]; | |
| fp16 var_432_to_fp16 = const()[name = string("op_432_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_442_cast_fp16 = layer_norm(axes = var_442_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_432_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = string("op_442_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_453_to_fp16 = const()[name = string("op_453_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85439872)))]; | |
| tensor<fp16, [1024]> var_454_to_fp16 = const()[name = string("op_454_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87537088)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_18_cast_fp16 = linear(bias = var_454_to_fp16, weight = var_453_to_fp16, x = var_442_cast_fp16)[name = string("linear_18_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_457_to_fp16 = const()[name = string("op_457_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87539200)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_457_to_fp16, x = var_442_cast_fp16)[name = string("linear_19_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_461_to_fp16 = const()[name = string("op_461_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89636416)))]; | |
| tensor<fp16, [1024]> var_462_to_fp16 = const()[name = string("op_462_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91733632)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_20_cast_fp16 = linear(bias = var_462_to_fp16, weight = var_461_to_fp16, x = var_442_cast_fp16)[name = string("linear_20_cast_fp16")]; | |
| tensor<int32, [4]> var_470 = const()[name = string("op_470"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_471_cast_fp16 = reshape(shape = var_470, x = linear_18_cast_fp16)[name = string("op_471_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_174_to_fp16 = const()[name = string("const_174_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_15_cast_fp16 = mul(x = var_471_cast_fp16, y = const_174_to_fp16)[name = string("q_15_cast_fp16")]; | |
| tensor<int32, [4]> var_477 = const()[name = string("op_477"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_478_cast_fp16 = reshape(shape = var_477, x = linear_19_cast_fp16)[name = string("op_478_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_175_to_fp16 = const()[name = string("const_175_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_15_cast_fp16 = mul(x = var_478_cast_fp16, y = const_175_to_fp16)[name = string("k_15_cast_fp16")]; | |
| tensor<int32, [4]> var_484 = const()[name = string("op_484"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_485_cast_fp16 = reshape(shape = var_484, x = linear_20_cast_fp16)[name = string("op_485_cast_fp16")]; | |
| tensor<int32, [4]> var_486 = const()[name = string("op_486"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_7_transpose_x_0 = const()[name = string("qk_7_transpose_x_0"), val = bool(false)]; | |
| bool qk_7_transpose_y_0 = const()[name = string("qk_7_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_102_perm_0 = const()[name = string("transpose_102_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_103_perm_0 = const()[name = string("transpose_103_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_103 = transpose(perm = transpose_103_perm_0, x = k_15_cast_fp16)[name = string("transpose_225")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_102 = transpose(perm = transpose_102_perm_0, x = q_15_cast_fp16)[name = string("transpose_226")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_7_cast_fp16 = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_102, y = transpose_103)[name = string("qk_7_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_490_cast_fp16 = softmax(axis = var_426, x = qk_7_cast_fp16)[name = string("op_490_cast_fp16")]; | |
| bool var_492_transpose_x_0 = const()[name = string("op_492_transpose_x_0"), val = bool(false)]; | |
| bool var_492_transpose_y_0 = const()[name = string("op_492_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_15_cast_fp16 = transpose(perm = var_486, x = var_485_cast_fp16)[name = string("transpose_227")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_492_cast_fp16 = matmul(transpose_x = var_492_transpose_x_0, transpose_y = var_492_transpose_y_0, x = var_490_cast_fp16, y = v_15_cast_fp16)[name = string("op_492_cast_fp16")]; | |
| tensor<int32, [4]> var_493 = const()[name = string("op_493"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_3 = const()[name = string("concat_3"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_494_cast_fp16 = transpose(perm = var_493, x = var_492_cast_fp16)[name = string("transpose_224")]; | |
| tensor<fp16, [1, 1500, 1024]> x_47_cast_fp16 = reshape(shape = concat_3, x = var_494_cast_fp16)[name = string("x_47_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_498_to_fp16 = const()[name = string("op_498_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91735744)))]; | |
| tensor<fp16, [1024]> var_499_to_fp16 = const()[name = string("op_499_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93832960)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_21_cast_fp16 = linear(bias = var_499_to_fp16, weight = var_498_to_fp16, x = x_47_cast_fp16)[name = string("linear_21_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = string("x_49_cast_fp16")]; | |
| tensor<int32, [1]> var_506_axes_0 = const()[name = string("op_506_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_3_mlp_ln_weight_to_fp16 = const()[name = string("blocks_3_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93835072)))]; | |
| tensor<fp16, [1024]> blocks_3_mlp_ln_bias_to_fp16 = const()[name = string("blocks_3_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93837184)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_506_cast_fp16 = layer_norm(axes = var_506_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_432_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = string("op_506_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_515_to_fp16 = const()[name = string("op_515_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93839296)))]; | |
| tensor<fp16, [4096]> var_516_to_fp16 = const()[name = string("op_516_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102227968)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_22_cast_fp16 = linear(bias = var_516_to_fp16, weight = var_515_to_fp16, x = var_506_cast_fp16)[name = string("linear_22_cast_fp16")]; | |
| string x_53_mode_0 = const()[name = string("x_53_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = string("x_53_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_521_to_fp16 = const()[name = string("op_521_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102236224)))]; | |
| tensor<fp16, [1024]> var_522_to_fp16 = const()[name = string("op_522_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110624896)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_23_cast_fp16 = linear(bias = var_522_to_fp16, weight = var_521_to_fp16, x = x_53_cast_fp16)[name = string("linear_23_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_55_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = string("x_55_cast_fp16")]; | |
| int32 var_532 = const()[name = string("op_532"), val = int32(-1)]; | |
| tensor<int32, [1]> var_548_axes_0 = const()[name = string("op_548_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_4_attn_ln_weight_to_fp16 = const()[name = string("blocks_4_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110627008)))]; | |
| tensor<fp16, [1024]> blocks_4_attn_ln_bias_to_fp16 = const()[name = string("blocks_4_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110629120)))]; | |
| fp16 var_538_to_fp16 = const()[name = string("op_538_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_548_cast_fp16 = layer_norm(axes = var_548_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_538_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast_fp16)[name = string("op_548_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_559_to_fp16 = const()[name = string("op_559_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110631232)))]; | |
| tensor<fp16, [1024]> var_560_to_fp16 = const()[name = string("op_560_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112728448)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_24_cast_fp16 = linear(bias = var_560_to_fp16, weight = var_559_to_fp16, x = var_548_cast_fp16)[name = string("linear_24_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_563_to_fp16 = const()[name = string("op_563_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112730560)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_25_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_563_to_fp16, x = var_548_cast_fp16)[name = string("linear_25_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_567_to_fp16 = const()[name = string("op_567_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114827776)))]; | |
| tensor<fp16, [1024]> var_568_to_fp16 = const()[name = string("op_568_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116924992)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_26_cast_fp16 = linear(bias = var_568_to_fp16, weight = var_567_to_fp16, x = var_548_cast_fp16)[name = string("linear_26_cast_fp16")]; | |
| tensor<int32, [4]> var_576 = const()[name = string("op_576"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_577_cast_fp16 = reshape(shape = var_576, x = linear_24_cast_fp16)[name = string("op_577_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_176_to_fp16 = const()[name = string("const_176_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_19_cast_fp16 = mul(x = var_577_cast_fp16, y = const_176_to_fp16)[name = string("q_19_cast_fp16")]; | |
| tensor<int32, [4]> var_583 = const()[name = string("op_583"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_584_cast_fp16 = reshape(shape = var_583, x = linear_25_cast_fp16)[name = string("op_584_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_177_to_fp16 = const()[name = string("const_177_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_19_cast_fp16 = mul(x = var_584_cast_fp16, y = const_177_to_fp16)[name = string("k_19_cast_fp16")]; | |
| tensor<int32, [4]> var_590 = const()[name = string("op_590"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_591_cast_fp16 = reshape(shape = var_590, x = linear_26_cast_fp16)[name = string("op_591_cast_fp16")]; | |
| tensor<int32, [4]> var_592 = const()[name = string("op_592"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_9_transpose_x_0 = const()[name = string("qk_9_transpose_x_0"), val = bool(false)]; | |
| bool qk_9_transpose_y_0 = const()[name = string("qk_9_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_104_perm_0 = const()[name = string("transpose_104_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_105_perm_0 = const()[name = string("transpose_105_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_105 = transpose(perm = transpose_105_perm_0, x = k_19_cast_fp16)[name = string("transpose_221")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_104 = transpose(perm = transpose_104_perm_0, x = q_19_cast_fp16)[name = string("transpose_222")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_9_cast_fp16 = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_104, y = transpose_105)[name = string("qk_9_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_596_cast_fp16 = softmax(axis = var_532, x = qk_9_cast_fp16)[name = string("op_596_cast_fp16")]; | |
| bool var_598_transpose_x_0 = const()[name = string("op_598_transpose_x_0"), val = bool(false)]; | |
| bool var_598_transpose_y_0 = const()[name = string("op_598_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_19_cast_fp16 = transpose(perm = var_592, x = var_591_cast_fp16)[name = string("transpose_223")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_598_cast_fp16 = matmul(transpose_x = var_598_transpose_x_0, transpose_y = var_598_transpose_y_0, x = var_596_cast_fp16, y = v_19_cast_fp16)[name = string("op_598_cast_fp16")]; | |
| tensor<int32, [4]> var_599 = const()[name = string("op_599"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_4 = const()[name = string("concat_4"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_600_cast_fp16 = transpose(perm = var_599, x = var_598_cast_fp16)[name = string("transpose_220")]; | |
| tensor<fp16, [1, 1500, 1024]> x_59_cast_fp16 = reshape(shape = concat_4, x = var_600_cast_fp16)[name = string("x_59_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_604_to_fp16 = const()[name = string("op_604_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116927104)))]; | |
| tensor<fp16, [1024]> var_605_to_fp16 = const()[name = string("op_605_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119024320)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_27_cast_fp16 = linear(bias = var_605_to_fp16, weight = var_604_to_fp16, x = x_59_cast_fp16)[name = string("linear_27_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_61_cast_fp16 = add(x = x_55_cast_fp16, y = linear_27_cast_fp16)[name = string("x_61_cast_fp16")]; | |
| tensor<int32, [1]> var_612_axes_0 = const()[name = string("op_612_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_4_mlp_ln_weight_to_fp16 = const()[name = string("blocks_4_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119026432)))]; | |
| tensor<fp16, [1024]> blocks_4_mlp_ln_bias_to_fp16 = const()[name = string("blocks_4_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119028544)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_612_cast_fp16 = layer_norm(axes = var_612_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_538_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast_fp16)[name = string("op_612_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_621_to_fp16 = const()[name = string("op_621_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119030656)))]; | |
| tensor<fp16, [4096]> var_622_to_fp16 = const()[name = string("op_622_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127419328)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_28_cast_fp16 = linear(bias = var_622_to_fp16, weight = var_621_to_fp16, x = var_612_cast_fp16)[name = string("linear_28_cast_fp16")]; | |
| string x_65_mode_0 = const()[name = string("x_65_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_65_cast_fp16 = gelu(mode = x_65_mode_0, x = linear_28_cast_fp16)[name = string("x_65_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_627_to_fp16 = const()[name = string("op_627_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127427584)))]; | |
| tensor<fp16, [1024]> var_628_to_fp16 = const()[name = string("op_628_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135816256)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_29_cast_fp16 = linear(bias = var_628_to_fp16, weight = var_627_to_fp16, x = x_65_cast_fp16)[name = string("linear_29_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_67_cast_fp16 = add(x = x_61_cast_fp16, y = linear_29_cast_fp16)[name = string("x_67_cast_fp16")]; | |
| int32 var_638 = const()[name = string("op_638"), val = int32(-1)]; | |
| tensor<int32, [1]> var_654_axes_0 = const()[name = string("op_654_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_5_attn_ln_weight_to_fp16 = const()[name = string("blocks_5_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135818368)))]; | |
| tensor<fp16, [1024]> blocks_5_attn_ln_bias_to_fp16 = const()[name = string("blocks_5_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135820480)))]; | |
| fp16 var_644_to_fp16 = const()[name = string("op_644_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_654_cast_fp16 = layer_norm(axes = var_654_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_644_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast_fp16)[name = string("op_654_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_665_to_fp16 = const()[name = string("op_665_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135822592)))]; | |
| tensor<fp16, [1024]> var_666_to_fp16 = const()[name = string("op_666_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137919808)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_30_cast_fp16 = linear(bias = var_666_to_fp16, weight = var_665_to_fp16, x = var_654_cast_fp16)[name = string("linear_30_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_669_to_fp16 = const()[name = string("op_669_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137921920)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_31_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_669_to_fp16, x = var_654_cast_fp16)[name = string("linear_31_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_673_to_fp16 = const()[name = string("op_673_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140019136)))]; | |
| tensor<fp16, [1024]> var_674_to_fp16 = const()[name = string("op_674_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142116352)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_32_cast_fp16 = linear(bias = var_674_to_fp16, weight = var_673_to_fp16, x = var_654_cast_fp16)[name = string("linear_32_cast_fp16")]; | |
| tensor<int32, [4]> var_682 = const()[name = string("op_682"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_683_cast_fp16 = reshape(shape = var_682, x = linear_30_cast_fp16)[name = string("op_683_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_178_to_fp16 = const()[name = string("const_178_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_23_cast_fp16 = mul(x = var_683_cast_fp16, y = const_178_to_fp16)[name = string("q_23_cast_fp16")]; | |
| tensor<int32, [4]> var_689 = const()[name = string("op_689"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_690_cast_fp16 = reshape(shape = var_689, x = linear_31_cast_fp16)[name = string("op_690_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_179_to_fp16 = const()[name = string("const_179_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_23_cast_fp16 = mul(x = var_690_cast_fp16, y = const_179_to_fp16)[name = string("k_23_cast_fp16")]; | |
| tensor<int32, [4]> var_696 = const()[name = string("op_696"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_697_cast_fp16 = reshape(shape = var_696, x = linear_32_cast_fp16)[name = string("op_697_cast_fp16")]; | |
| tensor<int32, [4]> var_698 = const()[name = string("op_698"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_11_transpose_x_0 = const()[name = string("qk_11_transpose_x_0"), val = bool(false)]; | |
| bool qk_11_transpose_y_0 = const()[name = string("qk_11_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_106_perm_0 = const()[name = string("transpose_106_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_107_perm_0 = const()[name = string("transpose_107_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_107 = transpose(perm = transpose_107_perm_0, x = k_23_cast_fp16)[name = string("transpose_217")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_106 = transpose(perm = transpose_106_perm_0, x = q_23_cast_fp16)[name = string("transpose_218")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_11_cast_fp16 = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_106, y = transpose_107)[name = string("qk_11_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_702_cast_fp16 = softmax(axis = var_638, x = qk_11_cast_fp16)[name = string("op_702_cast_fp16")]; | |
| bool var_704_transpose_x_0 = const()[name = string("op_704_transpose_x_0"), val = bool(false)]; | |
| bool var_704_transpose_y_0 = const()[name = string("op_704_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_23_cast_fp16 = transpose(perm = var_698, x = var_697_cast_fp16)[name = string("transpose_219")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_704_cast_fp16 = matmul(transpose_x = var_704_transpose_x_0, transpose_y = var_704_transpose_y_0, x = var_702_cast_fp16, y = v_23_cast_fp16)[name = string("op_704_cast_fp16")]; | |
| tensor<int32, [4]> var_705 = const()[name = string("op_705"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_5 = const()[name = string("concat_5"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_706_cast_fp16 = transpose(perm = var_705, x = var_704_cast_fp16)[name = string("transpose_216")]; | |
| tensor<fp16, [1, 1500, 1024]> x_71_cast_fp16 = reshape(shape = concat_5, x = var_706_cast_fp16)[name = string("x_71_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_710_to_fp16 = const()[name = string("op_710_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142118464)))]; | |
| tensor<fp16, [1024]> var_711_to_fp16 = const()[name = string("op_711_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144215680)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_33_cast_fp16 = linear(bias = var_711_to_fp16, weight = var_710_to_fp16, x = x_71_cast_fp16)[name = string("linear_33_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_73_cast_fp16 = add(x = x_67_cast_fp16, y = linear_33_cast_fp16)[name = string("x_73_cast_fp16")]; | |
| tensor<int32, [1]> var_718_axes_0 = const()[name = string("op_718_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_5_mlp_ln_weight_to_fp16 = const()[name = string("blocks_5_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144217792)))]; | |
| tensor<fp16, [1024]> blocks_5_mlp_ln_bias_to_fp16 = const()[name = string("blocks_5_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144219904)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_718_cast_fp16 = layer_norm(axes = var_718_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_644_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast_fp16)[name = string("op_718_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_727_to_fp16 = const()[name = string("op_727_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144222016)))]; | |
| tensor<fp16, [4096]> var_728_to_fp16 = const()[name = string("op_728_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152610688)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_34_cast_fp16 = linear(bias = var_728_to_fp16, weight = var_727_to_fp16, x = var_718_cast_fp16)[name = string("linear_34_cast_fp16")]; | |
| string x_77_mode_0 = const()[name = string("x_77_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_77_cast_fp16 = gelu(mode = x_77_mode_0, x = linear_34_cast_fp16)[name = string("x_77_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_733_to_fp16 = const()[name = string("op_733_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152618944)))]; | |
| tensor<fp16, [1024]> var_734_to_fp16 = const()[name = string("op_734_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161007616)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_35_cast_fp16 = linear(bias = var_734_to_fp16, weight = var_733_to_fp16, x = x_77_cast_fp16)[name = string("linear_35_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_79_cast_fp16 = add(x = x_73_cast_fp16, y = linear_35_cast_fp16)[name = string("x_79_cast_fp16")]; | |
| int32 var_744 = const()[name = string("op_744"), val = int32(-1)]; | |
| tensor<int32, [1]> var_760_axes_0 = const()[name = string("op_760_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_6_attn_ln_weight_to_fp16 = const()[name = string("blocks_6_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161009728)))]; | |
| tensor<fp16, [1024]> blocks_6_attn_ln_bias_to_fp16 = const()[name = string("blocks_6_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161011840)))]; | |
| fp16 var_750_to_fp16 = const()[name = string("op_750_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_760_cast_fp16 = layer_norm(axes = var_760_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_750_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast_fp16)[name = string("op_760_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_771_to_fp16 = const()[name = string("op_771_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161013952)))]; | |
| tensor<fp16, [1024]> var_772_to_fp16 = const()[name = string("op_772_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163111168)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_36_cast_fp16 = linear(bias = var_772_to_fp16, weight = var_771_to_fp16, x = var_760_cast_fp16)[name = string("linear_36_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_775_to_fp16 = const()[name = string("op_775_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163113280)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_775_to_fp16, x = var_760_cast_fp16)[name = string("linear_37_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_779_to_fp16 = const()[name = string("op_779_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165210496)))]; | |
| tensor<fp16, [1024]> var_780_to_fp16 = const()[name = string("op_780_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167307712)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_38_cast_fp16 = linear(bias = var_780_to_fp16, weight = var_779_to_fp16, x = var_760_cast_fp16)[name = string("linear_38_cast_fp16")]; | |
| tensor<int32, [4]> var_788 = const()[name = string("op_788"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_789_cast_fp16 = reshape(shape = var_788, x = linear_36_cast_fp16)[name = string("op_789_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_180_to_fp16 = const()[name = string("const_180_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_27_cast_fp16 = mul(x = var_789_cast_fp16, y = const_180_to_fp16)[name = string("q_27_cast_fp16")]; | |
| tensor<int32, [4]> var_795 = const()[name = string("op_795"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_796_cast_fp16 = reshape(shape = var_795, x = linear_37_cast_fp16)[name = string("op_796_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_181_to_fp16 = const()[name = string("const_181_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_27_cast_fp16 = mul(x = var_796_cast_fp16, y = const_181_to_fp16)[name = string("k_27_cast_fp16")]; | |
| tensor<int32, [4]> var_802 = const()[name = string("op_802"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_803_cast_fp16 = reshape(shape = var_802, x = linear_38_cast_fp16)[name = string("op_803_cast_fp16")]; | |
| tensor<int32, [4]> var_804 = const()[name = string("op_804"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_13_transpose_x_0 = const()[name = string("qk_13_transpose_x_0"), val = bool(false)]; | |
| bool qk_13_transpose_y_0 = const()[name = string("qk_13_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_108_perm_0 = const()[name = string("transpose_108_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_109_perm_0 = const()[name = string("transpose_109_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_109 = transpose(perm = transpose_109_perm_0, x = k_27_cast_fp16)[name = string("transpose_213")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_108 = transpose(perm = transpose_108_perm_0, x = q_27_cast_fp16)[name = string("transpose_214")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_13_cast_fp16 = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_108, y = transpose_109)[name = string("qk_13_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_808_cast_fp16 = softmax(axis = var_744, x = qk_13_cast_fp16)[name = string("op_808_cast_fp16")]; | |
| bool var_810_transpose_x_0 = const()[name = string("op_810_transpose_x_0"), val = bool(false)]; | |
| bool var_810_transpose_y_0 = const()[name = string("op_810_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_27_cast_fp16 = transpose(perm = var_804, x = var_803_cast_fp16)[name = string("transpose_215")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_810_cast_fp16 = matmul(transpose_x = var_810_transpose_x_0, transpose_y = var_810_transpose_y_0, x = var_808_cast_fp16, y = v_27_cast_fp16)[name = string("op_810_cast_fp16")]; | |
| tensor<int32, [4]> var_811 = const()[name = string("op_811"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_6 = const()[name = string("concat_6"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_812_cast_fp16 = transpose(perm = var_811, x = var_810_cast_fp16)[name = string("transpose_212")]; | |
| tensor<fp16, [1, 1500, 1024]> x_83_cast_fp16 = reshape(shape = concat_6, x = var_812_cast_fp16)[name = string("x_83_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_816_to_fp16 = const()[name = string("op_816_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167309824)))]; | |
| tensor<fp16, [1024]> var_817_to_fp16 = const()[name = string("op_817_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169407040)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_39_cast_fp16 = linear(bias = var_817_to_fp16, weight = var_816_to_fp16, x = x_83_cast_fp16)[name = string("linear_39_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_85_cast_fp16 = add(x = x_79_cast_fp16, y = linear_39_cast_fp16)[name = string("x_85_cast_fp16")]; | |
| tensor<int32, [1]> var_824_axes_0 = const()[name = string("op_824_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_6_mlp_ln_weight_to_fp16 = const()[name = string("blocks_6_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169409152)))]; | |
| tensor<fp16, [1024]> blocks_6_mlp_ln_bias_to_fp16 = const()[name = string("blocks_6_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169411264)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_824_cast_fp16 = layer_norm(axes = var_824_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_750_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast_fp16)[name = string("op_824_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_833_to_fp16 = const()[name = string("op_833_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169413376)))]; | |
| tensor<fp16, [4096]> var_834_to_fp16 = const()[name = string("op_834_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177802048)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_40_cast_fp16 = linear(bias = var_834_to_fp16, weight = var_833_to_fp16, x = var_824_cast_fp16)[name = string("linear_40_cast_fp16")]; | |
| string x_89_mode_0 = const()[name = string("x_89_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_89_cast_fp16 = gelu(mode = x_89_mode_0, x = linear_40_cast_fp16)[name = string("x_89_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_839_to_fp16 = const()[name = string("op_839_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177810304)))]; | |
| tensor<fp16, [1024]> var_840_to_fp16 = const()[name = string("op_840_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186198976)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_41_cast_fp16 = linear(bias = var_840_to_fp16, weight = var_839_to_fp16, x = x_89_cast_fp16)[name = string("linear_41_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_91_cast_fp16 = add(x = x_85_cast_fp16, y = linear_41_cast_fp16)[name = string("x_91_cast_fp16")]; | |
| int32 var_850 = const()[name = string("op_850"), val = int32(-1)]; | |
| tensor<int32, [1]> var_866_axes_0 = const()[name = string("op_866_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_7_attn_ln_weight_to_fp16 = const()[name = string("blocks_7_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186201088)))]; | |
| tensor<fp16, [1024]> blocks_7_attn_ln_bias_to_fp16 = const()[name = string("blocks_7_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186203200)))]; | |
| fp16 var_856_to_fp16 = const()[name = string("op_856_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_866_cast_fp16 = layer_norm(axes = var_866_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_856_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast_fp16)[name = string("op_866_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_877_to_fp16 = const()[name = string("op_877_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186205312)))]; | |
| tensor<fp16, [1024]> var_878_to_fp16 = const()[name = string("op_878_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188302528)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_42_cast_fp16 = linear(bias = var_878_to_fp16, weight = var_877_to_fp16, x = var_866_cast_fp16)[name = string("linear_42_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_881_to_fp16 = const()[name = string("op_881_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188304640)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_881_to_fp16, x = var_866_cast_fp16)[name = string("linear_43_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_885_to_fp16 = const()[name = string("op_885_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190401856)))]; | |
| tensor<fp16, [1024]> var_886_to_fp16 = const()[name = string("op_886_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192499072)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_44_cast_fp16 = linear(bias = var_886_to_fp16, weight = var_885_to_fp16, x = var_866_cast_fp16)[name = string("linear_44_cast_fp16")]; | |
| tensor<int32, [4]> var_894 = const()[name = string("op_894"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_895_cast_fp16 = reshape(shape = var_894, x = linear_42_cast_fp16)[name = string("op_895_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_182_to_fp16 = const()[name = string("const_182_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_31_cast_fp16 = mul(x = var_895_cast_fp16, y = const_182_to_fp16)[name = string("q_31_cast_fp16")]; | |
| tensor<int32, [4]> var_901 = const()[name = string("op_901"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_902_cast_fp16 = reshape(shape = var_901, x = linear_43_cast_fp16)[name = string("op_902_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_183_to_fp16 = const()[name = string("const_183_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_31_cast_fp16 = mul(x = var_902_cast_fp16, y = const_183_to_fp16)[name = string("k_31_cast_fp16")]; | |
| tensor<int32, [4]> var_908 = const()[name = string("op_908"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_909_cast_fp16 = reshape(shape = var_908, x = linear_44_cast_fp16)[name = string("op_909_cast_fp16")]; | |
| tensor<int32, [4]> var_910 = const()[name = string("op_910"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_15_transpose_x_0 = const()[name = string("qk_15_transpose_x_0"), val = bool(false)]; | |
| bool qk_15_transpose_y_0 = const()[name = string("qk_15_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_110_perm_0 = const()[name = string("transpose_110_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_111_perm_0 = const()[name = string("transpose_111_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_111 = transpose(perm = transpose_111_perm_0, x = k_31_cast_fp16)[name = string("transpose_209")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_110 = transpose(perm = transpose_110_perm_0, x = q_31_cast_fp16)[name = string("transpose_210")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_15_cast_fp16 = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_110, y = transpose_111)[name = string("qk_15_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_914_cast_fp16 = softmax(axis = var_850, x = qk_15_cast_fp16)[name = string("op_914_cast_fp16")]; | |
| bool var_916_transpose_x_0 = const()[name = string("op_916_transpose_x_0"), val = bool(false)]; | |
| bool var_916_transpose_y_0 = const()[name = string("op_916_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_31_cast_fp16 = transpose(perm = var_910, x = var_909_cast_fp16)[name = string("transpose_211")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_916_cast_fp16 = matmul(transpose_x = var_916_transpose_x_0, transpose_y = var_916_transpose_y_0, x = var_914_cast_fp16, y = v_31_cast_fp16)[name = string("op_916_cast_fp16")]; | |
| tensor<int32, [4]> var_917 = const()[name = string("op_917"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_7 = const()[name = string("concat_7"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_918_cast_fp16 = transpose(perm = var_917, x = var_916_cast_fp16)[name = string("transpose_208")]; | |
| tensor<fp16, [1, 1500, 1024]> x_95_cast_fp16 = reshape(shape = concat_7, x = var_918_cast_fp16)[name = string("x_95_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_922_to_fp16 = const()[name = string("op_922_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192501184)))]; | |
| tensor<fp16, [1024]> var_923_to_fp16 = const()[name = string("op_923_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194598400)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_45_cast_fp16 = linear(bias = var_923_to_fp16, weight = var_922_to_fp16, x = x_95_cast_fp16)[name = string("linear_45_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_97_cast_fp16 = add(x = x_91_cast_fp16, y = linear_45_cast_fp16)[name = string("x_97_cast_fp16")]; | |
| tensor<int32, [1]> var_930_axes_0 = const()[name = string("op_930_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_7_mlp_ln_weight_to_fp16 = const()[name = string("blocks_7_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194600512)))]; | |
| tensor<fp16, [1024]> blocks_7_mlp_ln_bias_to_fp16 = const()[name = string("blocks_7_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194602624)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_930_cast_fp16 = layer_norm(axes = var_930_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_856_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast_fp16)[name = string("op_930_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_939_to_fp16 = const()[name = string("op_939_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194604736)))]; | |
| tensor<fp16, [4096]> var_940_to_fp16 = const()[name = string("op_940_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202993408)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_46_cast_fp16 = linear(bias = var_940_to_fp16, weight = var_939_to_fp16, x = var_930_cast_fp16)[name = string("linear_46_cast_fp16")]; | |
| string x_101_mode_0 = const()[name = string("x_101_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_101_cast_fp16 = gelu(mode = x_101_mode_0, x = linear_46_cast_fp16)[name = string("x_101_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_945_to_fp16 = const()[name = string("op_945_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203001664)))]; | |
| tensor<fp16, [1024]> var_946_to_fp16 = const()[name = string("op_946_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211390336)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_47_cast_fp16 = linear(bias = var_946_to_fp16, weight = var_945_to_fp16, x = x_101_cast_fp16)[name = string("linear_47_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_103_cast_fp16 = add(x = x_97_cast_fp16, y = linear_47_cast_fp16)[name = string("x_103_cast_fp16")]; | |
| int32 var_956 = const()[name = string("op_956"), val = int32(-1)]; | |
| tensor<int32, [1]> var_972_axes_0 = const()[name = string("op_972_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_8_attn_ln_weight_to_fp16 = const()[name = string("blocks_8_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211392448)))]; | |
| tensor<fp16, [1024]> blocks_8_attn_ln_bias_to_fp16 = const()[name = string("blocks_8_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211394560)))]; | |
| fp16 var_962_to_fp16 = const()[name = string("op_962_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_972_cast_fp16 = layer_norm(axes = var_972_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_962_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast_fp16)[name = string("op_972_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_983_to_fp16 = const()[name = string("op_983_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211396672)))]; | |
| tensor<fp16, [1024]> var_984_to_fp16 = const()[name = string("op_984_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213493888)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_48_cast_fp16 = linear(bias = var_984_to_fp16, weight = var_983_to_fp16, x = var_972_cast_fp16)[name = string("linear_48_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_987_to_fp16 = const()[name = string("op_987_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213496000)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_49_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_987_to_fp16, x = var_972_cast_fp16)[name = string("linear_49_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_991_to_fp16 = const()[name = string("op_991_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215593216)))]; | |
| tensor<fp16, [1024]> var_992_to_fp16 = const()[name = string("op_992_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217690432)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_50_cast_fp16 = linear(bias = var_992_to_fp16, weight = var_991_to_fp16, x = var_972_cast_fp16)[name = string("linear_50_cast_fp16")]; | |
| tensor<int32, [4]> var_1000 = const()[name = string("op_1000"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1001_cast_fp16 = reshape(shape = var_1000, x = linear_48_cast_fp16)[name = string("op_1001_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_184_to_fp16 = const()[name = string("const_184_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_35_cast_fp16 = mul(x = var_1001_cast_fp16, y = const_184_to_fp16)[name = string("q_35_cast_fp16")]; | |
| tensor<int32, [4]> var_1007 = const()[name = string("op_1007"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1008_cast_fp16 = reshape(shape = var_1007, x = linear_49_cast_fp16)[name = string("op_1008_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_185_to_fp16 = const()[name = string("const_185_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_35_cast_fp16 = mul(x = var_1008_cast_fp16, y = const_185_to_fp16)[name = string("k_35_cast_fp16")]; | |
| tensor<int32, [4]> var_1014 = const()[name = string("op_1014"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1015_cast_fp16 = reshape(shape = var_1014, x = linear_50_cast_fp16)[name = string("op_1015_cast_fp16")]; | |
| tensor<int32, [4]> var_1016 = const()[name = string("op_1016"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_17_transpose_x_0 = const()[name = string("qk_17_transpose_x_0"), val = bool(false)]; | |
| bool qk_17_transpose_y_0 = const()[name = string("qk_17_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_112_perm_0 = const()[name = string("transpose_112_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_113_perm_0 = const()[name = string("transpose_113_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_113 = transpose(perm = transpose_113_perm_0, x = k_35_cast_fp16)[name = string("transpose_205")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_112 = transpose(perm = transpose_112_perm_0, x = q_35_cast_fp16)[name = string("transpose_206")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_17_cast_fp16 = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_112, y = transpose_113)[name = string("qk_17_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_1020_cast_fp16 = softmax(axis = var_956, x = qk_17_cast_fp16)[name = string("op_1020_cast_fp16")]; | |
| bool var_1022_transpose_x_0 = const()[name = string("op_1022_transpose_x_0"), val = bool(false)]; | |
| bool var_1022_transpose_y_0 = const()[name = string("op_1022_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_35_cast_fp16 = transpose(perm = var_1016, x = var_1015_cast_fp16)[name = string("transpose_207")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_1022_cast_fp16 = matmul(transpose_x = var_1022_transpose_x_0, transpose_y = var_1022_transpose_y_0, x = var_1020_cast_fp16, y = v_35_cast_fp16)[name = string("op_1022_cast_fp16")]; | |
| tensor<int32, [4]> var_1023 = const()[name = string("op_1023"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_8 = const()[name = string("concat_8"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1024_cast_fp16 = transpose(perm = var_1023, x = var_1022_cast_fp16)[name = string("transpose_204")]; | |
| tensor<fp16, [1, 1500, 1024]> x_107_cast_fp16 = reshape(shape = concat_8, x = var_1024_cast_fp16)[name = string("x_107_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1028_to_fp16 = const()[name = string("op_1028_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217692544)))]; | |
| tensor<fp16, [1024]> var_1029_to_fp16 = const()[name = string("op_1029_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219789760)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_51_cast_fp16 = linear(bias = var_1029_to_fp16, weight = var_1028_to_fp16, x = x_107_cast_fp16)[name = string("linear_51_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_109_cast_fp16 = add(x = x_103_cast_fp16, y = linear_51_cast_fp16)[name = string("x_109_cast_fp16")]; | |
| tensor<int32, [1]> var_1036_axes_0 = const()[name = string("op_1036_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_8_mlp_ln_weight_to_fp16 = const()[name = string("blocks_8_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219791872)))]; | |
| tensor<fp16, [1024]> blocks_8_mlp_ln_bias_to_fp16 = const()[name = string("blocks_8_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219793984)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_1036_cast_fp16 = layer_norm(axes = var_1036_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_962_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast_fp16)[name = string("op_1036_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_1045_to_fp16 = const()[name = string("op_1045_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219796096)))]; | |
| tensor<fp16, [4096]> var_1046_to_fp16 = const()[name = string("op_1046_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228184768)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_52_cast_fp16 = linear(bias = var_1046_to_fp16, weight = var_1045_to_fp16, x = var_1036_cast_fp16)[name = string("linear_52_cast_fp16")]; | |
| string x_113_mode_0 = const()[name = string("x_113_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_113_cast_fp16 = gelu(mode = x_113_mode_0, x = linear_52_cast_fp16)[name = string("x_113_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_1051_to_fp16 = const()[name = string("op_1051_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228193024)))]; | |
| tensor<fp16, [1024]> var_1052_to_fp16 = const()[name = string("op_1052_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236581696)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_53_cast_fp16 = linear(bias = var_1052_to_fp16, weight = var_1051_to_fp16, x = x_113_cast_fp16)[name = string("linear_53_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_115_cast_fp16 = add(x = x_109_cast_fp16, y = linear_53_cast_fp16)[name = string("x_115_cast_fp16")]; | |
| int32 var_1062 = const()[name = string("op_1062"), val = int32(-1)]; | |
| tensor<int32, [1]> var_1078_axes_0 = const()[name = string("op_1078_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_9_attn_ln_weight_to_fp16 = const()[name = string("blocks_9_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236583808)))]; | |
| tensor<fp16, [1024]> blocks_9_attn_ln_bias_to_fp16 = const()[name = string("blocks_9_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236585920)))]; | |
| fp16 var_1068_to_fp16 = const()[name = string("op_1068_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_1078_cast_fp16 = layer_norm(axes = var_1078_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1068_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast_fp16)[name = string("op_1078_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1089_to_fp16 = const()[name = string("op_1089_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236588032)))]; | |
| tensor<fp16, [1024]> var_1090_to_fp16 = const()[name = string("op_1090_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238685248)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_54_cast_fp16 = linear(bias = var_1090_to_fp16, weight = var_1089_to_fp16, x = var_1078_cast_fp16)[name = string("linear_54_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1093_to_fp16 = const()[name = string("op_1093_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238687360)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1093_to_fp16, x = var_1078_cast_fp16)[name = string("linear_55_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1097_to_fp16 = const()[name = string("op_1097_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240784576)))]; | |
| tensor<fp16, [1024]> var_1098_to_fp16 = const()[name = string("op_1098_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242881792)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_56_cast_fp16 = linear(bias = var_1098_to_fp16, weight = var_1097_to_fp16, x = var_1078_cast_fp16)[name = string("linear_56_cast_fp16")]; | |
| tensor<int32, [4]> var_1106 = const()[name = string("op_1106"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1107_cast_fp16 = reshape(shape = var_1106, x = linear_54_cast_fp16)[name = string("op_1107_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_186_to_fp16 = const()[name = string("const_186_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_39_cast_fp16 = mul(x = var_1107_cast_fp16, y = const_186_to_fp16)[name = string("q_39_cast_fp16")]; | |
| tensor<int32, [4]> var_1113 = const()[name = string("op_1113"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1114_cast_fp16 = reshape(shape = var_1113, x = linear_55_cast_fp16)[name = string("op_1114_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_187_to_fp16 = const()[name = string("const_187_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_39_cast_fp16 = mul(x = var_1114_cast_fp16, y = const_187_to_fp16)[name = string("k_39_cast_fp16")]; | |
| tensor<int32, [4]> var_1120 = const()[name = string("op_1120"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1121_cast_fp16 = reshape(shape = var_1120, x = linear_56_cast_fp16)[name = string("op_1121_cast_fp16")]; | |
| tensor<int32, [4]> var_1122 = const()[name = string("op_1122"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_19_transpose_x_0 = const()[name = string("qk_19_transpose_x_0"), val = bool(false)]; | |
| bool qk_19_transpose_y_0 = const()[name = string("qk_19_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_114_perm_0 = const()[name = string("transpose_114_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_115_perm_0 = const()[name = string("transpose_115_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_115 = transpose(perm = transpose_115_perm_0, x = k_39_cast_fp16)[name = string("transpose_201")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_114 = transpose(perm = transpose_114_perm_0, x = q_39_cast_fp16)[name = string("transpose_202")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_19_cast_fp16 = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_114, y = transpose_115)[name = string("qk_19_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_1126_cast_fp16 = softmax(axis = var_1062, x = qk_19_cast_fp16)[name = string("op_1126_cast_fp16")]; | |
| bool var_1128_transpose_x_0 = const()[name = string("op_1128_transpose_x_0"), val = bool(false)]; | |
| bool var_1128_transpose_y_0 = const()[name = string("op_1128_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_39_cast_fp16 = transpose(perm = var_1122, x = var_1121_cast_fp16)[name = string("transpose_203")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_1128_cast_fp16 = matmul(transpose_x = var_1128_transpose_x_0, transpose_y = var_1128_transpose_y_0, x = var_1126_cast_fp16, y = v_39_cast_fp16)[name = string("op_1128_cast_fp16")]; | |
| tensor<int32, [4]> var_1129 = const()[name = string("op_1129"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_9 = const()[name = string("concat_9"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1130_cast_fp16 = transpose(perm = var_1129, x = var_1128_cast_fp16)[name = string("transpose_200")]; | |
| tensor<fp16, [1, 1500, 1024]> x_119_cast_fp16 = reshape(shape = concat_9, x = var_1130_cast_fp16)[name = string("x_119_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1134_to_fp16 = const()[name = string("op_1134_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242883904)))]; | |
| tensor<fp16, [1024]> var_1135_to_fp16 = const()[name = string("op_1135_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244981120)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_57_cast_fp16 = linear(bias = var_1135_to_fp16, weight = var_1134_to_fp16, x = x_119_cast_fp16)[name = string("linear_57_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_121_cast_fp16 = add(x = x_115_cast_fp16, y = linear_57_cast_fp16)[name = string("x_121_cast_fp16")]; | |
| tensor<int32, [1]> var_1142_axes_0 = const()[name = string("op_1142_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_9_mlp_ln_weight_to_fp16 = const()[name = string("blocks_9_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244983232)))]; | |
| tensor<fp16, [1024]> blocks_9_mlp_ln_bias_to_fp16 = const()[name = string("blocks_9_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244985344)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_1142_cast_fp16 = layer_norm(axes = var_1142_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1068_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast_fp16)[name = string("op_1142_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_1151_to_fp16 = const()[name = string("op_1151_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244987456)))]; | |
| tensor<fp16, [4096]> var_1152_to_fp16 = const()[name = string("op_1152_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253376128)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_58_cast_fp16 = linear(bias = var_1152_to_fp16, weight = var_1151_to_fp16, x = var_1142_cast_fp16)[name = string("linear_58_cast_fp16")]; | |
| string x_125_mode_0 = const()[name = string("x_125_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_125_cast_fp16 = gelu(mode = x_125_mode_0, x = linear_58_cast_fp16)[name = string("x_125_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_1157_to_fp16 = const()[name = string("op_1157_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253384384)))]; | |
| tensor<fp16, [1024]> var_1158_to_fp16 = const()[name = string("op_1158_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261773056)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_59_cast_fp16 = linear(bias = var_1158_to_fp16, weight = var_1157_to_fp16, x = x_125_cast_fp16)[name = string("linear_59_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_127_cast_fp16 = add(x = x_121_cast_fp16, y = linear_59_cast_fp16)[name = string("x_127_cast_fp16")]; | |
| int32 var_1168 = const()[name = string("op_1168"), val = int32(-1)]; | |
| tensor<int32, [1]> var_1184_axes_0 = const()[name = string("op_1184_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_10_attn_ln_weight_to_fp16 = const()[name = string("blocks_10_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261775168)))]; | |
| tensor<fp16, [1024]> blocks_10_attn_ln_bias_to_fp16 = const()[name = string("blocks_10_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261777280)))]; | |
| fp16 var_1174_to_fp16 = const()[name = string("op_1174_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_1184_cast_fp16 = layer_norm(axes = var_1184_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1174_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast_fp16)[name = string("op_1184_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1195_to_fp16 = const()[name = string("op_1195_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261779392)))]; | |
| tensor<fp16, [1024]> var_1196_to_fp16 = const()[name = string("op_1196_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263876608)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_60_cast_fp16 = linear(bias = var_1196_to_fp16, weight = var_1195_to_fp16, x = var_1184_cast_fp16)[name = string("linear_60_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1199_to_fp16 = const()[name = string("op_1199_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263878720)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_61_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1199_to_fp16, x = var_1184_cast_fp16)[name = string("linear_61_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1203_to_fp16 = const()[name = string("op_1203_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265975936)))]; | |
| tensor<fp16, [1024]> var_1204_to_fp16 = const()[name = string("op_1204_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268073152)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_62_cast_fp16 = linear(bias = var_1204_to_fp16, weight = var_1203_to_fp16, x = var_1184_cast_fp16)[name = string("linear_62_cast_fp16")]; | |
| tensor<int32, [4]> var_1212 = const()[name = string("op_1212"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1213_cast_fp16 = reshape(shape = var_1212, x = linear_60_cast_fp16)[name = string("op_1213_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_188_to_fp16 = const()[name = string("const_188_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_43_cast_fp16 = mul(x = var_1213_cast_fp16, y = const_188_to_fp16)[name = string("q_43_cast_fp16")]; | |
| tensor<int32, [4]> var_1219 = const()[name = string("op_1219"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1220_cast_fp16 = reshape(shape = var_1219, x = linear_61_cast_fp16)[name = string("op_1220_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_189_to_fp16 = const()[name = string("const_189_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_43_cast_fp16 = mul(x = var_1220_cast_fp16, y = const_189_to_fp16)[name = string("k_43_cast_fp16")]; | |
| tensor<int32, [4]> var_1226 = const()[name = string("op_1226"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1227_cast_fp16 = reshape(shape = var_1226, x = linear_62_cast_fp16)[name = string("op_1227_cast_fp16")]; | |
| tensor<int32, [4]> var_1228 = const()[name = string("op_1228"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_21_transpose_x_0 = const()[name = string("qk_21_transpose_x_0"), val = bool(false)]; | |
| bool qk_21_transpose_y_0 = const()[name = string("qk_21_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_116_perm_0 = const()[name = string("transpose_116_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_117_perm_0 = const()[name = string("transpose_117_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_117 = transpose(perm = transpose_117_perm_0, x = k_43_cast_fp16)[name = string("transpose_197")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_116 = transpose(perm = transpose_116_perm_0, x = q_43_cast_fp16)[name = string("transpose_198")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_21_cast_fp16 = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_116, y = transpose_117)[name = string("qk_21_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_1232_cast_fp16 = softmax(axis = var_1168, x = qk_21_cast_fp16)[name = string("op_1232_cast_fp16")]; | |
| bool var_1234_transpose_x_0 = const()[name = string("op_1234_transpose_x_0"), val = bool(false)]; | |
| bool var_1234_transpose_y_0 = const()[name = string("op_1234_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_43_cast_fp16 = transpose(perm = var_1228, x = var_1227_cast_fp16)[name = string("transpose_199")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_1234_cast_fp16 = matmul(transpose_x = var_1234_transpose_x_0, transpose_y = var_1234_transpose_y_0, x = var_1232_cast_fp16, y = v_43_cast_fp16)[name = string("op_1234_cast_fp16")]; | |
| tensor<int32, [4]> var_1235 = const()[name = string("op_1235"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_10 = const()[name = string("concat_10"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1236_cast_fp16 = transpose(perm = var_1235, x = var_1234_cast_fp16)[name = string("transpose_196")]; | |
| tensor<fp16, [1, 1500, 1024]> x_131_cast_fp16 = reshape(shape = concat_10, x = var_1236_cast_fp16)[name = string("x_131_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1240_to_fp16 = const()[name = string("op_1240_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268075264)))]; | |
| tensor<fp16, [1024]> var_1241_to_fp16 = const()[name = string("op_1241_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270172480)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_63_cast_fp16 = linear(bias = var_1241_to_fp16, weight = var_1240_to_fp16, x = x_131_cast_fp16)[name = string("linear_63_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_133_cast_fp16 = add(x = x_127_cast_fp16, y = linear_63_cast_fp16)[name = string("x_133_cast_fp16")]; | |
| tensor<int32, [1]> var_1248_axes_0 = const()[name = string("op_1248_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_10_mlp_ln_weight_to_fp16 = const()[name = string("blocks_10_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270174592)))]; | |
| tensor<fp16, [1024]> blocks_10_mlp_ln_bias_to_fp16 = const()[name = string("blocks_10_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270176704)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_1248_cast_fp16 = layer_norm(axes = var_1248_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1174_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast_fp16)[name = string("op_1248_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_1257_to_fp16 = const()[name = string("op_1257_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270178816)))]; | |
| tensor<fp16, [4096]> var_1258_to_fp16 = const()[name = string("op_1258_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278567488)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_64_cast_fp16 = linear(bias = var_1258_to_fp16, weight = var_1257_to_fp16, x = var_1248_cast_fp16)[name = string("linear_64_cast_fp16")]; | |
| string x_137_mode_0 = const()[name = string("x_137_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_137_cast_fp16 = gelu(mode = x_137_mode_0, x = linear_64_cast_fp16)[name = string("x_137_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_1263_to_fp16 = const()[name = string("op_1263_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278575744)))]; | |
| tensor<fp16, [1024]> var_1264_to_fp16 = const()[name = string("op_1264_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286964416)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_65_cast_fp16 = linear(bias = var_1264_to_fp16, weight = var_1263_to_fp16, x = x_137_cast_fp16)[name = string("linear_65_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_139_cast_fp16 = add(x = x_133_cast_fp16, y = linear_65_cast_fp16)[name = string("x_139_cast_fp16")]; | |
| int32 var_1274 = const()[name = string("op_1274"), val = int32(-1)]; | |
| tensor<int32, [1]> var_1290_axes_0 = const()[name = string("op_1290_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_11_attn_ln_weight_to_fp16 = const()[name = string("blocks_11_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286966528)))]; | |
| tensor<fp16, [1024]> blocks_11_attn_ln_bias_to_fp16 = const()[name = string("blocks_11_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286968640)))]; | |
| fp16 var_1280_to_fp16 = const()[name = string("op_1280_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_1290_cast_fp16 = layer_norm(axes = var_1290_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1280_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast_fp16)[name = string("op_1290_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1301_to_fp16 = const()[name = string("op_1301_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286970752)))]; | |
| tensor<fp16, [1024]> var_1302_to_fp16 = const()[name = string("op_1302_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289067968)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_66_cast_fp16 = linear(bias = var_1302_to_fp16, weight = var_1301_to_fp16, x = var_1290_cast_fp16)[name = string("linear_66_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1305_to_fp16 = const()[name = string("op_1305_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289070080)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_67_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1305_to_fp16, x = var_1290_cast_fp16)[name = string("linear_67_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1309_to_fp16 = const()[name = string("op_1309_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291167296)))]; | |
| tensor<fp16, [1024]> var_1310_to_fp16 = const()[name = string("op_1310_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293264512)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_68_cast_fp16 = linear(bias = var_1310_to_fp16, weight = var_1309_to_fp16, x = var_1290_cast_fp16)[name = string("linear_68_cast_fp16")]; | |
| tensor<int32, [4]> var_1318 = const()[name = string("op_1318"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1319_cast_fp16 = reshape(shape = var_1318, x = linear_66_cast_fp16)[name = string("op_1319_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_190_to_fp16 = const()[name = string("const_190_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_47_cast_fp16 = mul(x = var_1319_cast_fp16, y = const_190_to_fp16)[name = string("q_47_cast_fp16")]; | |
| tensor<int32, [4]> var_1325 = const()[name = string("op_1325"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1326_cast_fp16 = reshape(shape = var_1325, x = linear_67_cast_fp16)[name = string("op_1326_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_191_to_fp16 = const()[name = string("const_191_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_47_cast_fp16 = mul(x = var_1326_cast_fp16, y = const_191_to_fp16)[name = string("k_47_cast_fp16")]; | |
| tensor<int32, [4]> var_1332 = const()[name = string("op_1332"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1333_cast_fp16 = reshape(shape = var_1332, x = linear_68_cast_fp16)[name = string("op_1333_cast_fp16")]; | |
| tensor<int32, [4]> var_1334 = const()[name = string("op_1334"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_23_transpose_x_0 = const()[name = string("qk_23_transpose_x_0"), val = bool(false)]; | |
| bool qk_23_transpose_y_0 = const()[name = string("qk_23_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_118_perm_0 = const()[name = string("transpose_118_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_119_perm_0 = const()[name = string("transpose_119_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_119 = transpose(perm = transpose_119_perm_0, x = k_47_cast_fp16)[name = string("transpose_193")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_118 = transpose(perm = transpose_118_perm_0, x = q_47_cast_fp16)[name = string("transpose_194")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_23_cast_fp16 = matmul(transpose_x = qk_23_transpose_x_0, transpose_y = qk_23_transpose_y_0, x = transpose_118, y = transpose_119)[name = string("qk_23_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_1338_cast_fp16 = softmax(axis = var_1274, x = qk_23_cast_fp16)[name = string("op_1338_cast_fp16")]; | |
| bool var_1340_transpose_x_0 = const()[name = string("op_1340_transpose_x_0"), val = bool(false)]; | |
| bool var_1340_transpose_y_0 = const()[name = string("op_1340_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_47_cast_fp16 = transpose(perm = var_1334, x = var_1333_cast_fp16)[name = string("transpose_195")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_1340_cast_fp16 = matmul(transpose_x = var_1340_transpose_x_0, transpose_y = var_1340_transpose_y_0, x = var_1338_cast_fp16, y = v_47_cast_fp16)[name = string("op_1340_cast_fp16")]; | |
| tensor<int32, [4]> var_1341 = const()[name = string("op_1341"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_11 = const()[name = string("concat_11"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1342_cast_fp16 = transpose(perm = var_1341, x = var_1340_cast_fp16)[name = string("transpose_192")]; | |
| tensor<fp16, [1, 1500, 1024]> x_143_cast_fp16 = reshape(shape = concat_11, x = var_1342_cast_fp16)[name = string("x_143_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1346_to_fp16 = const()[name = string("op_1346_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293266624)))]; | |
| tensor<fp16, [1024]> var_1347_to_fp16 = const()[name = string("op_1347_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295363840)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_69_cast_fp16 = linear(bias = var_1347_to_fp16, weight = var_1346_to_fp16, x = x_143_cast_fp16)[name = string("linear_69_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_145_cast_fp16 = add(x = x_139_cast_fp16, y = linear_69_cast_fp16)[name = string("x_145_cast_fp16")]; | |
| tensor<int32, [1]> var_1354_axes_0 = const()[name = string("op_1354_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_11_mlp_ln_weight_to_fp16 = const()[name = string("blocks_11_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295365952)))]; | |
| tensor<fp16, [1024]> blocks_11_mlp_ln_bias_to_fp16 = const()[name = string("blocks_11_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295368064)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_1354_cast_fp16 = layer_norm(axes = var_1354_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1280_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast_fp16)[name = string("op_1354_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_1363_to_fp16 = const()[name = string("op_1363_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295370176)))]; | |
| tensor<fp16, [4096]> var_1364_to_fp16 = const()[name = string("op_1364_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303758848)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_70_cast_fp16 = linear(bias = var_1364_to_fp16, weight = var_1363_to_fp16, x = var_1354_cast_fp16)[name = string("linear_70_cast_fp16")]; | |
| string x_149_mode_0 = const()[name = string("x_149_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_149_cast_fp16 = gelu(mode = x_149_mode_0, x = linear_70_cast_fp16)[name = string("x_149_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_1369_to_fp16 = const()[name = string("op_1369_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303767104)))]; | |
| tensor<fp16, [1024]> var_1370_to_fp16 = const()[name = string("op_1370_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312155776)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_71_cast_fp16 = linear(bias = var_1370_to_fp16, weight = var_1369_to_fp16, x = x_149_cast_fp16)[name = string("linear_71_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_151_cast_fp16 = add(x = x_145_cast_fp16, y = linear_71_cast_fp16)[name = string("x_151_cast_fp16")]; | |
| int32 var_1380 = const()[name = string("op_1380"), val = int32(-1)]; | |
| tensor<int32, [1]> var_1396_axes_0 = const()[name = string("op_1396_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_12_attn_ln_weight_to_fp16 = const()[name = string("blocks_12_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312157888)))]; | |
| tensor<fp16, [1024]> blocks_12_attn_ln_bias_to_fp16 = const()[name = string("blocks_12_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312160000)))]; | |
| fp16 var_1386_to_fp16 = const()[name = string("op_1386_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_1396_cast_fp16 = layer_norm(axes = var_1396_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1386_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast_fp16)[name = string("op_1396_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1407_to_fp16 = const()[name = string("op_1407_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312162112)))]; | |
| tensor<fp16, [1024]> var_1408_to_fp16 = const()[name = string("op_1408_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314259328)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_72_cast_fp16 = linear(bias = var_1408_to_fp16, weight = var_1407_to_fp16, x = var_1396_cast_fp16)[name = string("linear_72_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1411_to_fp16 = const()[name = string("op_1411_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314261440)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1411_to_fp16, x = var_1396_cast_fp16)[name = string("linear_73_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1415_to_fp16 = const()[name = string("op_1415_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316358656)))]; | |
| tensor<fp16, [1024]> var_1416_to_fp16 = const()[name = string("op_1416_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318455872)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_74_cast_fp16 = linear(bias = var_1416_to_fp16, weight = var_1415_to_fp16, x = var_1396_cast_fp16)[name = string("linear_74_cast_fp16")]; | |
| tensor<int32, [4]> var_1424 = const()[name = string("op_1424"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1425_cast_fp16 = reshape(shape = var_1424, x = linear_72_cast_fp16)[name = string("op_1425_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_192_to_fp16 = const()[name = string("const_192_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_51_cast_fp16 = mul(x = var_1425_cast_fp16, y = const_192_to_fp16)[name = string("q_51_cast_fp16")]; | |
| tensor<int32, [4]> var_1431 = const()[name = string("op_1431"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1432_cast_fp16 = reshape(shape = var_1431, x = linear_73_cast_fp16)[name = string("op_1432_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_193_to_fp16 = const()[name = string("const_193_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_51_cast_fp16 = mul(x = var_1432_cast_fp16, y = const_193_to_fp16)[name = string("k_51_cast_fp16")]; | |
| tensor<int32, [4]> var_1438 = const()[name = string("op_1438"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1439_cast_fp16 = reshape(shape = var_1438, x = linear_74_cast_fp16)[name = string("op_1439_cast_fp16")]; | |
| tensor<int32, [4]> var_1440 = const()[name = string("op_1440"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_25_transpose_x_0 = const()[name = string("qk_25_transpose_x_0"), val = bool(false)]; | |
| bool qk_25_transpose_y_0 = const()[name = string("qk_25_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_120_perm_0 = const()[name = string("transpose_120_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_121_perm_0 = const()[name = string("transpose_121_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_121 = transpose(perm = transpose_121_perm_0, x = k_51_cast_fp16)[name = string("transpose_189")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_120 = transpose(perm = transpose_120_perm_0, x = q_51_cast_fp16)[name = string("transpose_190")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_25_cast_fp16 = matmul(transpose_x = qk_25_transpose_x_0, transpose_y = qk_25_transpose_y_0, x = transpose_120, y = transpose_121)[name = string("qk_25_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_1444_cast_fp16 = softmax(axis = var_1380, x = qk_25_cast_fp16)[name = string("op_1444_cast_fp16")]; | |
| bool var_1446_transpose_x_0 = const()[name = string("op_1446_transpose_x_0"), val = bool(false)]; | |
| bool var_1446_transpose_y_0 = const()[name = string("op_1446_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_51_cast_fp16 = transpose(perm = var_1440, x = var_1439_cast_fp16)[name = string("transpose_191")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_1446_cast_fp16 = matmul(transpose_x = var_1446_transpose_x_0, transpose_y = var_1446_transpose_y_0, x = var_1444_cast_fp16, y = v_51_cast_fp16)[name = string("op_1446_cast_fp16")]; | |
| tensor<int32, [4]> var_1447 = const()[name = string("op_1447"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_12 = const()[name = string("concat_12"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1448_cast_fp16 = transpose(perm = var_1447, x = var_1446_cast_fp16)[name = string("transpose_188")]; | |
| tensor<fp16, [1, 1500, 1024]> x_155_cast_fp16 = reshape(shape = concat_12, x = var_1448_cast_fp16)[name = string("x_155_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1452_to_fp16 = const()[name = string("op_1452_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318457984)))]; | |
| tensor<fp16, [1024]> var_1453_to_fp16 = const()[name = string("op_1453_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320555200)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_75_cast_fp16 = linear(bias = var_1453_to_fp16, weight = var_1452_to_fp16, x = x_155_cast_fp16)[name = string("linear_75_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_157_cast_fp16 = add(x = x_151_cast_fp16, y = linear_75_cast_fp16)[name = string("x_157_cast_fp16")]; | |
| tensor<int32, [1]> var_1460_axes_0 = const()[name = string("op_1460_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_12_mlp_ln_weight_to_fp16 = const()[name = string("blocks_12_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320557312)))]; | |
| tensor<fp16, [1024]> blocks_12_mlp_ln_bias_to_fp16 = const()[name = string("blocks_12_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320559424)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_1460_cast_fp16 = layer_norm(axes = var_1460_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1386_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast_fp16)[name = string("op_1460_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_1469_to_fp16 = const()[name = string("op_1469_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320561536)))]; | |
| tensor<fp16, [4096]> var_1470_to_fp16 = const()[name = string("op_1470_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(328950208)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_76_cast_fp16 = linear(bias = var_1470_to_fp16, weight = var_1469_to_fp16, x = var_1460_cast_fp16)[name = string("linear_76_cast_fp16")]; | |
| string x_161_mode_0 = const()[name = string("x_161_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_161_cast_fp16 = gelu(mode = x_161_mode_0, x = linear_76_cast_fp16)[name = string("x_161_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_1475_to_fp16 = const()[name = string("op_1475_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(328958464)))]; | |
| tensor<fp16, [1024]> var_1476_to_fp16 = const()[name = string("op_1476_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337347136)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_77_cast_fp16 = linear(bias = var_1476_to_fp16, weight = var_1475_to_fp16, x = x_161_cast_fp16)[name = string("linear_77_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_163_cast_fp16 = add(x = x_157_cast_fp16, y = linear_77_cast_fp16)[name = string("x_163_cast_fp16")]; | |
| int32 var_1486 = const()[name = string("op_1486"), val = int32(-1)]; | |
| tensor<int32, [1]> var_1502_axes_0 = const()[name = string("op_1502_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_13_attn_ln_weight_to_fp16 = const()[name = string("blocks_13_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337349248)))]; | |
| tensor<fp16, [1024]> blocks_13_attn_ln_bias_to_fp16 = const()[name = string("blocks_13_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337351360)))]; | |
| fp16 var_1492_to_fp16 = const()[name = string("op_1492_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_1502_cast_fp16 = layer_norm(axes = var_1502_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1492_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast_fp16)[name = string("op_1502_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1513_to_fp16 = const()[name = string("op_1513_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337353472)))]; | |
| tensor<fp16, [1024]> var_1514_to_fp16 = const()[name = string("op_1514_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339450688)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_78_cast_fp16 = linear(bias = var_1514_to_fp16, weight = var_1513_to_fp16, x = var_1502_cast_fp16)[name = string("linear_78_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1517_to_fp16 = const()[name = string("op_1517_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339452800)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_79_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1517_to_fp16, x = var_1502_cast_fp16)[name = string("linear_79_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1521_to_fp16 = const()[name = string("op_1521_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341550016)))]; | |
| tensor<fp16, [1024]> var_1522_to_fp16 = const()[name = string("op_1522_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343647232)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_80_cast_fp16 = linear(bias = var_1522_to_fp16, weight = var_1521_to_fp16, x = var_1502_cast_fp16)[name = string("linear_80_cast_fp16")]; | |
| tensor<int32, [4]> var_1530 = const()[name = string("op_1530"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1531_cast_fp16 = reshape(shape = var_1530, x = linear_78_cast_fp16)[name = string("op_1531_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_194_to_fp16 = const()[name = string("const_194_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_55_cast_fp16 = mul(x = var_1531_cast_fp16, y = const_194_to_fp16)[name = string("q_55_cast_fp16")]; | |
| tensor<int32, [4]> var_1537 = const()[name = string("op_1537"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1538_cast_fp16 = reshape(shape = var_1537, x = linear_79_cast_fp16)[name = string("op_1538_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_195_to_fp16 = const()[name = string("const_195_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_55_cast_fp16 = mul(x = var_1538_cast_fp16, y = const_195_to_fp16)[name = string("k_55_cast_fp16")]; | |
| tensor<int32, [4]> var_1544 = const()[name = string("op_1544"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1545_cast_fp16 = reshape(shape = var_1544, x = linear_80_cast_fp16)[name = string("op_1545_cast_fp16")]; | |
| tensor<int32, [4]> var_1546 = const()[name = string("op_1546"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_27_transpose_x_0 = const()[name = string("qk_27_transpose_x_0"), val = bool(false)]; | |
| bool qk_27_transpose_y_0 = const()[name = string("qk_27_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_122_perm_0 = const()[name = string("transpose_122_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_123_perm_0 = const()[name = string("transpose_123_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_123 = transpose(perm = transpose_123_perm_0, x = k_55_cast_fp16)[name = string("transpose_185")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_122 = transpose(perm = transpose_122_perm_0, x = q_55_cast_fp16)[name = string("transpose_186")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_27_cast_fp16 = matmul(transpose_x = qk_27_transpose_x_0, transpose_y = qk_27_transpose_y_0, x = transpose_122, y = transpose_123)[name = string("qk_27_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_1550_cast_fp16 = softmax(axis = var_1486, x = qk_27_cast_fp16)[name = string("op_1550_cast_fp16")]; | |
| bool var_1552_transpose_x_0 = const()[name = string("op_1552_transpose_x_0"), val = bool(false)]; | |
| bool var_1552_transpose_y_0 = const()[name = string("op_1552_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_55_cast_fp16 = transpose(perm = var_1546, x = var_1545_cast_fp16)[name = string("transpose_187")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_1552_cast_fp16 = matmul(transpose_x = var_1552_transpose_x_0, transpose_y = var_1552_transpose_y_0, x = var_1550_cast_fp16, y = v_55_cast_fp16)[name = string("op_1552_cast_fp16")]; | |
| tensor<int32, [4]> var_1553 = const()[name = string("op_1553"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_13 = const()[name = string("concat_13"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1554_cast_fp16 = transpose(perm = var_1553, x = var_1552_cast_fp16)[name = string("transpose_184")]; | |
| tensor<fp16, [1, 1500, 1024]> x_167_cast_fp16 = reshape(shape = concat_13, x = var_1554_cast_fp16)[name = string("x_167_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1558_to_fp16 = const()[name = string("op_1558_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343649344)))]; | |
| tensor<fp16, [1024]> var_1559_to_fp16 = const()[name = string("op_1559_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345746560)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_81_cast_fp16 = linear(bias = var_1559_to_fp16, weight = var_1558_to_fp16, x = x_167_cast_fp16)[name = string("linear_81_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_169_cast_fp16 = add(x = x_163_cast_fp16, y = linear_81_cast_fp16)[name = string("x_169_cast_fp16")]; | |
| tensor<int32, [1]> var_1566_axes_0 = const()[name = string("op_1566_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_13_mlp_ln_weight_to_fp16 = const()[name = string("blocks_13_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345748672)))]; | |
| tensor<fp16, [1024]> blocks_13_mlp_ln_bias_to_fp16 = const()[name = string("blocks_13_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345750784)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_1566_cast_fp16 = layer_norm(axes = var_1566_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1492_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast_fp16)[name = string("op_1566_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_1575_to_fp16 = const()[name = string("op_1575_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345752896)))]; | |
| tensor<fp16, [4096]> var_1576_to_fp16 = const()[name = string("op_1576_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354141568)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_82_cast_fp16 = linear(bias = var_1576_to_fp16, weight = var_1575_to_fp16, x = var_1566_cast_fp16)[name = string("linear_82_cast_fp16")]; | |
| string x_173_mode_0 = const()[name = string("x_173_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_173_cast_fp16 = gelu(mode = x_173_mode_0, x = linear_82_cast_fp16)[name = string("x_173_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_1581_to_fp16 = const()[name = string("op_1581_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354149824)))]; | |
| tensor<fp16, [1024]> var_1582_to_fp16 = const()[name = string("op_1582_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362538496)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_83_cast_fp16 = linear(bias = var_1582_to_fp16, weight = var_1581_to_fp16, x = x_173_cast_fp16)[name = string("linear_83_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_175_cast_fp16 = add(x = x_169_cast_fp16, y = linear_83_cast_fp16)[name = string("x_175_cast_fp16")]; | |
| int32 var_1592 = const()[name = string("op_1592"), val = int32(-1)]; | |
| tensor<int32, [1]> var_1608_axes_0 = const()[name = string("op_1608_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_14_attn_ln_weight_to_fp16 = const()[name = string("blocks_14_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362540608)))]; | |
| tensor<fp16, [1024]> blocks_14_attn_ln_bias_to_fp16 = const()[name = string("blocks_14_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362542720)))]; | |
| fp16 var_1598_to_fp16 = const()[name = string("op_1598_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_1608_cast_fp16 = layer_norm(axes = var_1608_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1598_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast_fp16)[name = string("op_1608_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1619_to_fp16 = const()[name = string("op_1619_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362544832)))]; | |
| tensor<fp16, [1024]> var_1620_to_fp16 = const()[name = string("op_1620_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364642048)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_84_cast_fp16 = linear(bias = var_1620_to_fp16, weight = var_1619_to_fp16, x = var_1608_cast_fp16)[name = string("linear_84_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1623_to_fp16 = const()[name = string("op_1623_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364644160)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_85_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1623_to_fp16, x = var_1608_cast_fp16)[name = string("linear_85_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1627_to_fp16 = const()[name = string("op_1627_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(366741376)))]; | |
| tensor<fp16, [1024]> var_1628_to_fp16 = const()[name = string("op_1628_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368838592)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_86_cast_fp16 = linear(bias = var_1628_to_fp16, weight = var_1627_to_fp16, x = var_1608_cast_fp16)[name = string("linear_86_cast_fp16")]; | |
| tensor<int32, [4]> var_1636 = const()[name = string("op_1636"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1637_cast_fp16 = reshape(shape = var_1636, x = linear_84_cast_fp16)[name = string("op_1637_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_196_to_fp16 = const()[name = string("const_196_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_59_cast_fp16 = mul(x = var_1637_cast_fp16, y = const_196_to_fp16)[name = string("q_59_cast_fp16")]; | |
| tensor<int32, [4]> var_1643 = const()[name = string("op_1643"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1644_cast_fp16 = reshape(shape = var_1643, x = linear_85_cast_fp16)[name = string("op_1644_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_197_to_fp16 = const()[name = string("const_197_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_59_cast_fp16 = mul(x = var_1644_cast_fp16, y = const_197_to_fp16)[name = string("k_59_cast_fp16")]; | |
| tensor<int32, [4]> var_1650 = const()[name = string("op_1650"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1651_cast_fp16 = reshape(shape = var_1650, x = linear_86_cast_fp16)[name = string("op_1651_cast_fp16")]; | |
| tensor<int32, [4]> var_1652 = const()[name = string("op_1652"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_29_transpose_x_0 = const()[name = string("qk_29_transpose_x_0"), val = bool(false)]; | |
| bool qk_29_transpose_y_0 = const()[name = string("qk_29_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_124_perm_0 = const()[name = string("transpose_124_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_125_perm_0 = const()[name = string("transpose_125_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_125 = transpose(perm = transpose_125_perm_0, x = k_59_cast_fp16)[name = string("transpose_181")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_124 = transpose(perm = transpose_124_perm_0, x = q_59_cast_fp16)[name = string("transpose_182")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_29_cast_fp16 = matmul(transpose_x = qk_29_transpose_x_0, transpose_y = qk_29_transpose_y_0, x = transpose_124, y = transpose_125)[name = string("qk_29_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_1656_cast_fp16 = softmax(axis = var_1592, x = qk_29_cast_fp16)[name = string("op_1656_cast_fp16")]; | |
| bool var_1658_transpose_x_0 = const()[name = string("op_1658_transpose_x_0"), val = bool(false)]; | |
| bool var_1658_transpose_y_0 = const()[name = string("op_1658_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_59_cast_fp16 = transpose(perm = var_1652, x = var_1651_cast_fp16)[name = string("transpose_183")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_1658_cast_fp16 = matmul(transpose_x = var_1658_transpose_x_0, transpose_y = var_1658_transpose_y_0, x = var_1656_cast_fp16, y = v_59_cast_fp16)[name = string("op_1658_cast_fp16")]; | |
| tensor<int32, [4]> var_1659 = const()[name = string("op_1659"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_14 = const()[name = string("concat_14"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1660_cast_fp16 = transpose(perm = var_1659, x = var_1658_cast_fp16)[name = string("transpose_180")]; | |
| tensor<fp16, [1, 1500, 1024]> x_179_cast_fp16 = reshape(shape = concat_14, x = var_1660_cast_fp16)[name = string("x_179_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1664_to_fp16 = const()[name = string("op_1664_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368840704)))]; | |
| tensor<fp16, [1024]> var_1665_to_fp16 = const()[name = string("op_1665_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370937920)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_87_cast_fp16 = linear(bias = var_1665_to_fp16, weight = var_1664_to_fp16, x = x_179_cast_fp16)[name = string("linear_87_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_181_cast_fp16 = add(x = x_175_cast_fp16, y = linear_87_cast_fp16)[name = string("x_181_cast_fp16")]; | |
| tensor<int32, [1]> var_1672_axes_0 = const()[name = string("op_1672_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_14_mlp_ln_weight_to_fp16 = const()[name = string("blocks_14_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370940032)))]; | |
| tensor<fp16, [1024]> blocks_14_mlp_ln_bias_to_fp16 = const()[name = string("blocks_14_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370942144)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_1672_cast_fp16 = layer_norm(axes = var_1672_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1598_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast_fp16)[name = string("op_1672_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_1681_to_fp16 = const()[name = string("op_1681_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370944256)))]; | |
| tensor<fp16, [4096]> var_1682_to_fp16 = const()[name = string("op_1682_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(379332928)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_88_cast_fp16 = linear(bias = var_1682_to_fp16, weight = var_1681_to_fp16, x = var_1672_cast_fp16)[name = string("linear_88_cast_fp16")]; | |
| string x_185_mode_0 = const()[name = string("x_185_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_185_cast_fp16 = gelu(mode = x_185_mode_0, x = linear_88_cast_fp16)[name = string("x_185_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_1687_to_fp16 = const()[name = string("op_1687_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(379341184)))]; | |
| tensor<fp16, [1024]> var_1688_to_fp16 = const()[name = string("op_1688_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387729856)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_89_cast_fp16 = linear(bias = var_1688_to_fp16, weight = var_1687_to_fp16, x = x_185_cast_fp16)[name = string("linear_89_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_187_cast_fp16 = add(x = x_181_cast_fp16, y = linear_89_cast_fp16)[name = string("x_187_cast_fp16")]; | |
| int32 var_1698 = const()[name = string("op_1698"), val = int32(-1)]; | |
| tensor<int32, [1]> var_1714_axes_0 = const()[name = string("op_1714_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_15_attn_ln_weight_to_fp16 = const()[name = string("blocks_15_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387731968)))]; | |
| tensor<fp16, [1024]> blocks_15_attn_ln_bias_to_fp16 = const()[name = string("blocks_15_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387734080)))]; | |
| fp16 var_1704_to_fp16 = const()[name = string("op_1704_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_1714_cast_fp16 = layer_norm(axes = var_1714_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1704_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast_fp16)[name = string("op_1714_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1725_to_fp16 = const()[name = string("op_1725_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387736192)))]; | |
| tensor<fp16, [1024]> var_1726_to_fp16 = const()[name = string("op_1726_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(389833408)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_90_cast_fp16 = linear(bias = var_1726_to_fp16, weight = var_1725_to_fp16, x = var_1714_cast_fp16)[name = string("linear_90_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1729_to_fp16 = const()[name = string("op_1729_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(389835520)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1729_to_fp16, x = var_1714_cast_fp16)[name = string("linear_91_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1733_to_fp16 = const()[name = string("op_1733_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(391932736)))]; | |
| tensor<fp16, [1024]> var_1734_to_fp16 = const()[name = string("op_1734_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394029952)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_92_cast_fp16 = linear(bias = var_1734_to_fp16, weight = var_1733_to_fp16, x = var_1714_cast_fp16)[name = string("linear_92_cast_fp16")]; | |
| tensor<int32, [4]> var_1742 = const()[name = string("op_1742"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1743_cast_fp16 = reshape(shape = var_1742, x = linear_90_cast_fp16)[name = string("op_1743_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_198_to_fp16 = const()[name = string("const_198_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_63_cast_fp16 = mul(x = var_1743_cast_fp16, y = const_198_to_fp16)[name = string("q_63_cast_fp16")]; | |
| tensor<int32, [4]> var_1749 = const()[name = string("op_1749"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1750_cast_fp16 = reshape(shape = var_1749, x = linear_91_cast_fp16)[name = string("op_1750_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_199_to_fp16 = const()[name = string("const_199_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_63_cast_fp16 = mul(x = var_1750_cast_fp16, y = const_199_to_fp16)[name = string("k_63_cast_fp16")]; | |
| tensor<int32, [4]> var_1756 = const()[name = string("op_1756"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1757_cast_fp16 = reshape(shape = var_1756, x = linear_92_cast_fp16)[name = string("op_1757_cast_fp16")]; | |
| tensor<int32, [4]> var_1758 = const()[name = string("op_1758"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_31_transpose_x_0 = const()[name = string("qk_31_transpose_x_0"), val = bool(false)]; | |
| bool qk_31_transpose_y_0 = const()[name = string("qk_31_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_126_perm_0 = const()[name = string("transpose_126_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_127_perm_0 = const()[name = string("transpose_127_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_127 = transpose(perm = transpose_127_perm_0, x = k_63_cast_fp16)[name = string("transpose_177")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_126 = transpose(perm = transpose_126_perm_0, x = q_63_cast_fp16)[name = string("transpose_178")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_31_cast_fp16 = matmul(transpose_x = qk_31_transpose_x_0, transpose_y = qk_31_transpose_y_0, x = transpose_126, y = transpose_127)[name = string("qk_31_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_1762_cast_fp16 = softmax(axis = var_1698, x = qk_31_cast_fp16)[name = string("op_1762_cast_fp16")]; | |
| bool var_1764_transpose_x_0 = const()[name = string("op_1764_transpose_x_0"), val = bool(false)]; | |
| bool var_1764_transpose_y_0 = const()[name = string("op_1764_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_63_cast_fp16 = transpose(perm = var_1758, x = var_1757_cast_fp16)[name = string("transpose_179")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_1764_cast_fp16 = matmul(transpose_x = var_1764_transpose_x_0, transpose_y = var_1764_transpose_y_0, x = var_1762_cast_fp16, y = v_63_cast_fp16)[name = string("op_1764_cast_fp16")]; | |
| tensor<int32, [4]> var_1765 = const()[name = string("op_1765"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_15 = const()[name = string("concat_15"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1766_cast_fp16 = transpose(perm = var_1765, x = var_1764_cast_fp16)[name = string("transpose_176")]; | |
| tensor<fp16, [1, 1500, 1024]> x_191_cast_fp16 = reshape(shape = concat_15, x = var_1766_cast_fp16)[name = string("x_191_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1770_to_fp16 = const()[name = string("op_1770_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394032064)))]; | |
| tensor<fp16, [1024]> var_1771_to_fp16 = const()[name = string("op_1771_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396129280)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_93_cast_fp16 = linear(bias = var_1771_to_fp16, weight = var_1770_to_fp16, x = x_191_cast_fp16)[name = string("linear_93_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_193_cast_fp16 = add(x = x_187_cast_fp16, y = linear_93_cast_fp16)[name = string("x_193_cast_fp16")]; | |
| tensor<int32, [1]> var_1778_axes_0 = const()[name = string("op_1778_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_15_mlp_ln_weight_to_fp16 = const()[name = string("blocks_15_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396131392)))]; | |
| tensor<fp16, [1024]> blocks_15_mlp_ln_bias_to_fp16 = const()[name = string("blocks_15_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396133504)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_1778_cast_fp16 = layer_norm(axes = var_1778_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1704_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast_fp16)[name = string("op_1778_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_1787_to_fp16 = const()[name = string("op_1787_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396135616)))]; | |
| tensor<fp16, [4096]> var_1788_to_fp16 = const()[name = string("op_1788_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404524288)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_94_cast_fp16 = linear(bias = var_1788_to_fp16, weight = var_1787_to_fp16, x = var_1778_cast_fp16)[name = string("linear_94_cast_fp16")]; | |
| string x_197_mode_0 = const()[name = string("x_197_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_197_cast_fp16 = gelu(mode = x_197_mode_0, x = linear_94_cast_fp16)[name = string("x_197_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_1793_to_fp16 = const()[name = string("op_1793_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404532544)))]; | |
| tensor<fp16, [1024]> var_1794_to_fp16 = const()[name = string("op_1794_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412921216)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_95_cast_fp16 = linear(bias = var_1794_to_fp16, weight = var_1793_to_fp16, x = x_197_cast_fp16)[name = string("linear_95_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_199_cast_fp16 = add(x = x_193_cast_fp16, y = linear_95_cast_fp16)[name = string("x_199_cast_fp16")]; | |
| int32 var_1804 = const()[name = string("op_1804"), val = int32(-1)]; | |
| tensor<int32, [1]> var_1820_axes_0 = const()[name = string("op_1820_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_16_attn_ln_weight_to_fp16 = const()[name = string("blocks_16_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412923328)))]; | |
| tensor<fp16, [1024]> blocks_16_attn_ln_bias_to_fp16 = const()[name = string("blocks_16_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412925440)))]; | |
| fp16 var_1810_to_fp16 = const()[name = string("op_1810_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_1820_cast_fp16 = layer_norm(axes = var_1820_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1810_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast_fp16)[name = string("op_1820_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1831_to_fp16 = const()[name = string("op_1831_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412927552)))]; | |
| tensor<fp16, [1024]> var_1832_to_fp16 = const()[name = string("op_1832_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415024768)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_96_cast_fp16 = linear(bias = var_1832_to_fp16, weight = var_1831_to_fp16, x = var_1820_cast_fp16)[name = string("linear_96_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1835_to_fp16 = const()[name = string("op_1835_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415026880)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_97_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1835_to_fp16, x = var_1820_cast_fp16)[name = string("linear_97_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1839_to_fp16 = const()[name = string("op_1839_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(417124096)))]; | |
| tensor<fp16, [1024]> var_1840_to_fp16 = const()[name = string("op_1840_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(419221312)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_98_cast_fp16 = linear(bias = var_1840_to_fp16, weight = var_1839_to_fp16, x = var_1820_cast_fp16)[name = string("linear_98_cast_fp16")]; | |
| tensor<int32, [4]> var_1848 = const()[name = string("op_1848"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1849_cast_fp16 = reshape(shape = var_1848, x = linear_96_cast_fp16)[name = string("op_1849_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_200_to_fp16 = const()[name = string("const_200_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_67_cast_fp16 = mul(x = var_1849_cast_fp16, y = const_200_to_fp16)[name = string("q_67_cast_fp16")]; | |
| tensor<int32, [4]> var_1855 = const()[name = string("op_1855"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1856_cast_fp16 = reshape(shape = var_1855, x = linear_97_cast_fp16)[name = string("op_1856_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_201_to_fp16 = const()[name = string("const_201_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_67_cast_fp16 = mul(x = var_1856_cast_fp16, y = const_201_to_fp16)[name = string("k_67_cast_fp16")]; | |
| tensor<int32, [4]> var_1862 = const()[name = string("op_1862"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1863_cast_fp16 = reshape(shape = var_1862, x = linear_98_cast_fp16)[name = string("op_1863_cast_fp16")]; | |
| tensor<int32, [4]> var_1864 = const()[name = string("op_1864"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_33_transpose_x_0 = const()[name = string("qk_33_transpose_x_0"), val = bool(false)]; | |
| bool qk_33_transpose_y_0 = const()[name = string("qk_33_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_128_perm_0 = const()[name = string("transpose_128_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_129_perm_0 = const()[name = string("transpose_129_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_129 = transpose(perm = transpose_129_perm_0, x = k_67_cast_fp16)[name = string("transpose_173")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_128 = transpose(perm = transpose_128_perm_0, x = q_67_cast_fp16)[name = string("transpose_174")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_33_cast_fp16 = matmul(transpose_x = qk_33_transpose_x_0, transpose_y = qk_33_transpose_y_0, x = transpose_128, y = transpose_129)[name = string("qk_33_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_1868_cast_fp16 = softmax(axis = var_1804, x = qk_33_cast_fp16)[name = string("op_1868_cast_fp16")]; | |
| bool var_1870_transpose_x_0 = const()[name = string("op_1870_transpose_x_0"), val = bool(false)]; | |
| bool var_1870_transpose_y_0 = const()[name = string("op_1870_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_67_cast_fp16 = transpose(perm = var_1864, x = var_1863_cast_fp16)[name = string("transpose_175")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_1870_cast_fp16 = matmul(transpose_x = var_1870_transpose_x_0, transpose_y = var_1870_transpose_y_0, x = var_1868_cast_fp16, y = v_67_cast_fp16)[name = string("op_1870_cast_fp16")]; | |
| tensor<int32, [4]> var_1871 = const()[name = string("op_1871"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_16 = const()[name = string("concat_16"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1872_cast_fp16 = transpose(perm = var_1871, x = var_1870_cast_fp16)[name = string("transpose_172")]; | |
| tensor<fp16, [1, 1500, 1024]> x_203_cast_fp16 = reshape(shape = concat_16, x = var_1872_cast_fp16)[name = string("x_203_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1876_to_fp16 = const()[name = string("op_1876_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(419223424)))]; | |
| tensor<fp16, [1024]> var_1877_to_fp16 = const()[name = string("op_1877_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421320640)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_99_cast_fp16 = linear(bias = var_1877_to_fp16, weight = var_1876_to_fp16, x = x_203_cast_fp16)[name = string("linear_99_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_205_cast_fp16 = add(x = x_199_cast_fp16, y = linear_99_cast_fp16)[name = string("x_205_cast_fp16")]; | |
| tensor<int32, [1]> var_1884_axes_0 = const()[name = string("op_1884_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_16_mlp_ln_weight_to_fp16 = const()[name = string("blocks_16_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421322752)))]; | |
| tensor<fp16, [1024]> blocks_16_mlp_ln_bias_to_fp16 = const()[name = string("blocks_16_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421324864)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_1884_cast_fp16 = layer_norm(axes = var_1884_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1810_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast_fp16)[name = string("op_1884_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_1893_to_fp16 = const()[name = string("op_1893_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421326976)))]; | |
| tensor<fp16, [4096]> var_1894_to_fp16 = const()[name = string("op_1894_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(429715648)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_100_cast_fp16 = linear(bias = var_1894_to_fp16, weight = var_1893_to_fp16, x = var_1884_cast_fp16)[name = string("linear_100_cast_fp16")]; | |
| string x_209_mode_0 = const()[name = string("x_209_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_209_cast_fp16 = gelu(mode = x_209_mode_0, x = linear_100_cast_fp16)[name = string("x_209_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_1899_to_fp16 = const()[name = string("op_1899_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(429723904)))]; | |
| tensor<fp16, [1024]> var_1900_to_fp16 = const()[name = string("op_1900_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438112576)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_101_cast_fp16 = linear(bias = var_1900_to_fp16, weight = var_1899_to_fp16, x = x_209_cast_fp16)[name = string("linear_101_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_211_cast_fp16 = add(x = x_205_cast_fp16, y = linear_101_cast_fp16)[name = string("x_211_cast_fp16")]; | |
| int32 var_1910 = const()[name = string("op_1910"), val = int32(-1)]; | |
| tensor<int32, [1]> var_1926_axes_0 = const()[name = string("op_1926_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_17_attn_ln_weight_to_fp16 = const()[name = string("blocks_17_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438114688)))]; | |
| tensor<fp16, [1024]> blocks_17_attn_ln_bias_to_fp16 = const()[name = string("blocks_17_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438116800)))]; | |
| fp16 var_1916_to_fp16 = const()[name = string("op_1916_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_1926_cast_fp16 = layer_norm(axes = var_1926_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1916_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast_fp16)[name = string("op_1926_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1937_to_fp16 = const()[name = string("op_1937_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438118912)))]; | |
| tensor<fp16, [1024]> var_1938_to_fp16 = const()[name = string("op_1938_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(440216128)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_102_cast_fp16 = linear(bias = var_1938_to_fp16, weight = var_1937_to_fp16, x = var_1926_cast_fp16)[name = string("linear_102_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1941_to_fp16 = const()[name = string("op_1941_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(440218240)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_103_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1941_to_fp16, x = var_1926_cast_fp16)[name = string("linear_103_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1945_to_fp16 = const()[name = string("op_1945_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(442315456)))]; | |
| tensor<fp16, [1024]> var_1946_to_fp16 = const()[name = string("op_1946_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(444412672)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_104_cast_fp16 = linear(bias = var_1946_to_fp16, weight = var_1945_to_fp16, x = var_1926_cast_fp16)[name = string("linear_104_cast_fp16")]; | |
| tensor<int32, [4]> var_1954 = const()[name = string("op_1954"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1955_cast_fp16 = reshape(shape = var_1954, x = linear_102_cast_fp16)[name = string("op_1955_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_202_to_fp16 = const()[name = string("const_202_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_71_cast_fp16 = mul(x = var_1955_cast_fp16, y = const_202_to_fp16)[name = string("q_71_cast_fp16")]; | |
| tensor<int32, [4]> var_1961 = const()[name = string("op_1961"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1962_cast_fp16 = reshape(shape = var_1961, x = linear_103_cast_fp16)[name = string("op_1962_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_203_to_fp16 = const()[name = string("const_203_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_71_cast_fp16 = mul(x = var_1962_cast_fp16, y = const_203_to_fp16)[name = string("k_71_cast_fp16")]; | |
| tensor<int32, [4]> var_1968 = const()[name = string("op_1968"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1969_cast_fp16 = reshape(shape = var_1968, x = linear_104_cast_fp16)[name = string("op_1969_cast_fp16")]; | |
| tensor<int32, [4]> var_1970 = const()[name = string("op_1970"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_35_transpose_x_0 = const()[name = string("qk_35_transpose_x_0"), val = bool(false)]; | |
| bool qk_35_transpose_y_0 = const()[name = string("qk_35_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_130_perm_0 = const()[name = string("transpose_130_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_131_perm_0 = const()[name = string("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_131 = transpose(perm = transpose_131_perm_0, x = k_71_cast_fp16)[name = string("transpose_169")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_130 = transpose(perm = transpose_130_perm_0, x = q_71_cast_fp16)[name = string("transpose_170")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_35_cast_fp16 = matmul(transpose_x = qk_35_transpose_x_0, transpose_y = qk_35_transpose_y_0, x = transpose_130, y = transpose_131)[name = string("qk_35_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_1974_cast_fp16 = softmax(axis = var_1910, x = qk_35_cast_fp16)[name = string("op_1974_cast_fp16")]; | |
| bool var_1976_transpose_x_0 = const()[name = string("op_1976_transpose_x_0"), val = bool(false)]; | |
| bool var_1976_transpose_y_0 = const()[name = string("op_1976_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_71_cast_fp16 = transpose(perm = var_1970, x = var_1969_cast_fp16)[name = string("transpose_171")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_1976_cast_fp16 = matmul(transpose_x = var_1976_transpose_x_0, transpose_y = var_1976_transpose_y_0, x = var_1974_cast_fp16, y = v_71_cast_fp16)[name = string("op_1976_cast_fp16")]; | |
| tensor<int32, [4]> var_1977 = const()[name = string("op_1977"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_17 = const()[name = string("concat_17"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_1978_cast_fp16 = transpose(perm = var_1977, x = var_1976_cast_fp16)[name = string("transpose_168")]; | |
| tensor<fp16, [1, 1500, 1024]> x_215_cast_fp16 = reshape(shape = concat_17, x = var_1978_cast_fp16)[name = string("x_215_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_1982_to_fp16 = const()[name = string("op_1982_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(444414784)))]; | |
| tensor<fp16, [1024]> var_1983_to_fp16 = const()[name = string("op_1983_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(446512000)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_105_cast_fp16 = linear(bias = var_1983_to_fp16, weight = var_1982_to_fp16, x = x_215_cast_fp16)[name = string("linear_105_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_217_cast_fp16 = add(x = x_211_cast_fp16, y = linear_105_cast_fp16)[name = string("x_217_cast_fp16")]; | |
| tensor<int32, [1]> var_1990_axes_0 = const()[name = string("op_1990_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_17_mlp_ln_weight_to_fp16 = const()[name = string("blocks_17_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(446514112)))]; | |
| tensor<fp16, [1024]> blocks_17_mlp_ln_bias_to_fp16 = const()[name = string("blocks_17_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(446516224)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_1990_cast_fp16 = layer_norm(axes = var_1990_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1916_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast_fp16)[name = string("op_1990_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_1999_to_fp16 = const()[name = string("op_1999_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(446518336)))]; | |
| tensor<fp16, [4096]> var_2000_to_fp16 = const()[name = string("op_2000_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(454907008)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_106_cast_fp16 = linear(bias = var_2000_to_fp16, weight = var_1999_to_fp16, x = var_1990_cast_fp16)[name = string("linear_106_cast_fp16")]; | |
| string x_221_mode_0 = const()[name = string("x_221_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_221_cast_fp16 = gelu(mode = x_221_mode_0, x = linear_106_cast_fp16)[name = string("x_221_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_2005_to_fp16 = const()[name = string("op_2005_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(454915264)))]; | |
| tensor<fp16, [1024]> var_2006_to_fp16 = const()[name = string("op_2006_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(463303936)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_107_cast_fp16 = linear(bias = var_2006_to_fp16, weight = var_2005_to_fp16, x = x_221_cast_fp16)[name = string("linear_107_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_223_cast_fp16 = add(x = x_217_cast_fp16, y = linear_107_cast_fp16)[name = string("x_223_cast_fp16")]; | |
| int32 var_2016 = const()[name = string("op_2016"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2032_axes_0 = const()[name = string("op_2032_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_18_attn_ln_weight_to_fp16 = const()[name = string("blocks_18_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(463306048)))]; | |
| tensor<fp16, [1024]> blocks_18_attn_ln_bias_to_fp16 = const()[name = string("blocks_18_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(463308160)))]; | |
| fp16 var_2022_to_fp16 = const()[name = string("op_2022_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_2032_cast_fp16 = layer_norm(axes = var_2032_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_2022_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast_fp16)[name = string("op_2032_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2043_to_fp16 = const()[name = string("op_2043_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(463310272)))]; | |
| tensor<fp16, [1024]> var_2044_to_fp16 = const()[name = string("op_2044_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(465407488)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_108_cast_fp16 = linear(bias = var_2044_to_fp16, weight = var_2043_to_fp16, x = var_2032_cast_fp16)[name = string("linear_108_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2047_to_fp16 = const()[name = string("op_2047_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(465409600)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2047_to_fp16, x = var_2032_cast_fp16)[name = string("linear_109_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2051_to_fp16 = const()[name = string("op_2051_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(467506816)))]; | |
| tensor<fp16, [1024]> var_2052_to_fp16 = const()[name = string("op_2052_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(469604032)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_110_cast_fp16 = linear(bias = var_2052_to_fp16, weight = var_2051_to_fp16, x = var_2032_cast_fp16)[name = string("linear_110_cast_fp16")]; | |
| tensor<int32, [4]> var_2060 = const()[name = string("op_2060"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2061_cast_fp16 = reshape(shape = var_2060, x = linear_108_cast_fp16)[name = string("op_2061_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_204_to_fp16 = const()[name = string("const_204_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_75_cast_fp16 = mul(x = var_2061_cast_fp16, y = const_204_to_fp16)[name = string("q_75_cast_fp16")]; | |
| tensor<int32, [4]> var_2067 = const()[name = string("op_2067"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2068_cast_fp16 = reshape(shape = var_2067, x = linear_109_cast_fp16)[name = string("op_2068_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_205_to_fp16 = const()[name = string("const_205_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_75_cast_fp16 = mul(x = var_2068_cast_fp16, y = const_205_to_fp16)[name = string("k_75_cast_fp16")]; | |
| tensor<int32, [4]> var_2074 = const()[name = string("op_2074"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2075_cast_fp16 = reshape(shape = var_2074, x = linear_110_cast_fp16)[name = string("op_2075_cast_fp16")]; | |
| tensor<int32, [4]> var_2076 = const()[name = string("op_2076"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_37_transpose_x_0 = const()[name = string("qk_37_transpose_x_0"), val = bool(false)]; | |
| bool qk_37_transpose_y_0 = const()[name = string("qk_37_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_132_perm_0 = const()[name = string("transpose_132_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_133_perm_0 = const()[name = string("transpose_133_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_133 = transpose(perm = transpose_133_perm_0, x = k_75_cast_fp16)[name = string("transpose_165")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_132 = transpose(perm = transpose_132_perm_0, x = q_75_cast_fp16)[name = string("transpose_166")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_37_cast_fp16 = matmul(transpose_x = qk_37_transpose_x_0, transpose_y = qk_37_transpose_y_0, x = transpose_132, y = transpose_133)[name = string("qk_37_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_2080_cast_fp16 = softmax(axis = var_2016, x = qk_37_cast_fp16)[name = string("op_2080_cast_fp16")]; | |
| bool var_2082_transpose_x_0 = const()[name = string("op_2082_transpose_x_0"), val = bool(false)]; | |
| bool var_2082_transpose_y_0 = const()[name = string("op_2082_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_75_cast_fp16 = transpose(perm = var_2076, x = var_2075_cast_fp16)[name = string("transpose_167")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_2082_cast_fp16 = matmul(transpose_x = var_2082_transpose_x_0, transpose_y = var_2082_transpose_y_0, x = var_2080_cast_fp16, y = v_75_cast_fp16)[name = string("op_2082_cast_fp16")]; | |
| tensor<int32, [4]> var_2083 = const()[name = string("op_2083"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_18 = const()[name = string("concat_18"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2084_cast_fp16 = transpose(perm = var_2083, x = var_2082_cast_fp16)[name = string("transpose_164")]; | |
| tensor<fp16, [1, 1500, 1024]> x_227_cast_fp16 = reshape(shape = concat_18, x = var_2084_cast_fp16)[name = string("x_227_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2088_to_fp16 = const()[name = string("op_2088_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(469606144)))]; | |
| tensor<fp16, [1024]> var_2089_to_fp16 = const()[name = string("op_2089_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(471703360)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_111_cast_fp16 = linear(bias = var_2089_to_fp16, weight = var_2088_to_fp16, x = x_227_cast_fp16)[name = string("linear_111_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_229_cast_fp16 = add(x = x_223_cast_fp16, y = linear_111_cast_fp16)[name = string("x_229_cast_fp16")]; | |
| tensor<int32, [1]> var_2096_axes_0 = const()[name = string("op_2096_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_18_mlp_ln_weight_to_fp16 = const()[name = string("blocks_18_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(471705472)))]; | |
| tensor<fp16, [1024]> blocks_18_mlp_ln_bias_to_fp16 = const()[name = string("blocks_18_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(471707584)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_2096_cast_fp16 = layer_norm(axes = var_2096_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_2022_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast_fp16)[name = string("op_2096_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_2105_to_fp16 = const()[name = string("op_2105_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(471709696)))]; | |
| tensor<fp16, [4096]> var_2106_to_fp16 = const()[name = string("op_2106_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(480098368)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_112_cast_fp16 = linear(bias = var_2106_to_fp16, weight = var_2105_to_fp16, x = var_2096_cast_fp16)[name = string("linear_112_cast_fp16")]; | |
| string x_233_mode_0 = const()[name = string("x_233_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_233_cast_fp16 = gelu(mode = x_233_mode_0, x = linear_112_cast_fp16)[name = string("x_233_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_2111_to_fp16 = const()[name = string("op_2111_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(480106624)))]; | |
| tensor<fp16, [1024]> var_2112_to_fp16 = const()[name = string("op_2112_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(488495296)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_113_cast_fp16 = linear(bias = var_2112_to_fp16, weight = var_2111_to_fp16, x = x_233_cast_fp16)[name = string("linear_113_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_235_cast_fp16 = add(x = x_229_cast_fp16, y = linear_113_cast_fp16)[name = string("x_235_cast_fp16")]; | |
| int32 var_2122 = const()[name = string("op_2122"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2138_axes_0 = const()[name = string("op_2138_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_19_attn_ln_weight_to_fp16 = const()[name = string("blocks_19_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(488497408)))]; | |
| tensor<fp16, [1024]> blocks_19_attn_ln_bias_to_fp16 = const()[name = string("blocks_19_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(488499520)))]; | |
| fp16 var_2128_to_fp16 = const()[name = string("op_2128_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_2138_cast_fp16 = layer_norm(axes = var_2138_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_2128_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast_fp16)[name = string("op_2138_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2149_to_fp16 = const()[name = string("op_2149_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(488501632)))]; | |
| tensor<fp16, [1024]> var_2150_to_fp16 = const()[name = string("op_2150_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(490598848)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_114_cast_fp16 = linear(bias = var_2150_to_fp16, weight = var_2149_to_fp16, x = var_2138_cast_fp16)[name = string("linear_114_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2153_to_fp16 = const()[name = string("op_2153_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(490600960)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_115_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2153_to_fp16, x = var_2138_cast_fp16)[name = string("linear_115_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2157_to_fp16 = const()[name = string("op_2157_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(492698176)))]; | |
| tensor<fp16, [1024]> var_2158_to_fp16 = const()[name = string("op_2158_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(494795392)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_116_cast_fp16 = linear(bias = var_2158_to_fp16, weight = var_2157_to_fp16, x = var_2138_cast_fp16)[name = string("linear_116_cast_fp16")]; | |
| tensor<int32, [4]> var_2166 = const()[name = string("op_2166"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2167_cast_fp16 = reshape(shape = var_2166, x = linear_114_cast_fp16)[name = string("op_2167_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_206_to_fp16 = const()[name = string("const_206_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_79_cast_fp16 = mul(x = var_2167_cast_fp16, y = const_206_to_fp16)[name = string("q_79_cast_fp16")]; | |
| tensor<int32, [4]> var_2173 = const()[name = string("op_2173"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2174_cast_fp16 = reshape(shape = var_2173, x = linear_115_cast_fp16)[name = string("op_2174_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_207_to_fp16 = const()[name = string("const_207_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_79_cast_fp16 = mul(x = var_2174_cast_fp16, y = const_207_to_fp16)[name = string("k_79_cast_fp16")]; | |
| tensor<int32, [4]> var_2180 = const()[name = string("op_2180"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2181_cast_fp16 = reshape(shape = var_2180, x = linear_116_cast_fp16)[name = string("op_2181_cast_fp16")]; | |
| tensor<int32, [4]> var_2182 = const()[name = string("op_2182"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_39_transpose_x_0 = const()[name = string("qk_39_transpose_x_0"), val = bool(false)]; | |
| bool qk_39_transpose_y_0 = const()[name = string("qk_39_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_134_perm_0 = const()[name = string("transpose_134_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_135_perm_0 = const()[name = string("transpose_135_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_135 = transpose(perm = transpose_135_perm_0, x = k_79_cast_fp16)[name = string("transpose_161")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_134 = transpose(perm = transpose_134_perm_0, x = q_79_cast_fp16)[name = string("transpose_162")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_39_cast_fp16 = matmul(transpose_x = qk_39_transpose_x_0, transpose_y = qk_39_transpose_y_0, x = transpose_134, y = transpose_135)[name = string("qk_39_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_2186_cast_fp16 = softmax(axis = var_2122, x = qk_39_cast_fp16)[name = string("op_2186_cast_fp16")]; | |
| bool var_2188_transpose_x_0 = const()[name = string("op_2188_transpose_x_0"), val = bool(false)]; | |
| bool var_2188_transpose_y_0 = const()[name = string("op_2188_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_79_cast_fp16 = transpose(perm = var_2182, x = var_2181_cast_fp16)[name = string("transpose_163")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_2188_cast_fp16 = matmul(transpose_x = var_2188_transpose_x_0, transpose_y = var_2188_transpose_y_0, x = var_2186_cast_fp16, y = v_79_cast_fp16)[name = string("op_2188_cast_fp16")]; | |
| tensor<int32, [4]> var_2189 = const()[name = string("op_2189"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_19 = const()[name = string("concat_19"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2190_cast_fp16 = transpose(perm = var_2189, x = var_2188_cast_fp16)[name = string("transpose_160")]; | |
| tensor<fp16, [1, 1500, 1024]> x_239_cast_fp16 = reshape(shape = concat_19, x = var_2190_cast_fp16)[name = string("x_239_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2194_to_fp16 = const()[name = string("op_2194_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(494797504)))]; | |
| tensor<fp16, [1024]> var_2195_to_fp16 = const()[name = string("op_2195_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496894720)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_117_cast_fp16 = linear(bias = var_2195_to_fp16, weight = var_2194_to_fp16, x = x_239_cast_fp16)[name = string("linear_117_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_241_cast_fp16 = add(x = x_235_cast_fp16, y = linear_117_cast_fp16)[name = string("x_241_cast_fp16")]; | |
| tensor<int32, [1]> var_2202_axes_0 = const()[name = string("op_2202_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_19_mlp_ln_weight_to_fp16 = const()[name = string("blocks_19_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496896832)))]; | |
| tensor<fp16, [1024]> blocks_19_mlp_ln_bias_to_fp16 = const()[name = string("blocks_19_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496898944)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_2202_cast_fp16 = layer_norm(axes = var_2202_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_2128_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast_fp16)[name = string("op_2202_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_2211_to_fp16 = const()[name = string("op_2211_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496901056)))]; | |
| tensor<fp16, [4096]> var_2212_to_fp16 = const()[name = string("op_2212_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(505289728)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_118_cast_fp16 = linear(bias = var_2212_to_fp16, weight = var_2211_to_fp16, x = var_2202_cast_fp16)[name = string("linear_118_cast_fp16")]; | |
| string x_245_mode_0 = const()[name = string("x_245_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_245_cast_fp16 = gelu(mode = x_245_mode_0, x = linear_118_cast_fp16)[name = string("x_245_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_2217_to_fp16 = const()[name = string("op_2217_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(505297984)))]; | |
| tensor<fp16, [1024]> var_2218_to_fp16 = const()[name = string("op_2218_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513686656)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_119_cast_fp16 = linear(bias = var_2218_to_fp16, weight = var_2217_to_fp16, x = x_245_cast_fp16)[name = string("linear_119_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_247_cast_fp16 = add(x = x_241_cast_fp16, y = linear_119_cast_fp16)[name = string("x_247_cast_fp16")]; | |
| int32 var_2228 = const()[name = string("op_2228"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2244_axes_0 = const()[name = string("op_2244_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_20_attn_ln_weight_to_fp16 = const()[name = string("blocks_20_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513688768)))]; | |
| tensor<fp16, [1024]> blocks_20_attn_ln_bias_to_fp16 = const()[name = string("blocks_20_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513690880)))]; | |
| fp16 var_2234_to_fp16 = const()[name = string("op_2234_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_2244_cast_fp16 = layer_norm(axes = var_2244_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2234_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast_fp16)[name = string("op_2244_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2255_to_fp16 = const()[name = string("op_2255_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513692992)))]; | |
| tensor<fp16, [1024]> var_2256_to_fp16 = const()[name = string("op_2256_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(515790208)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_120_cast_fp16 = linear(bias = var_2256_to_fp16, weight = var_2255_to_fp16, x = var_2244_cast_fp16)[name = string("linear_120_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2259_to_fp16 = const()[name = string("op_2259_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(515792320)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_121_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2259_to_fp16, x = var_2244_cast_fp16)[name = string("linear_121_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2263_to_fp16 = const()[name = string("op_2263_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(517889536)))]; | |
| tensor<fp16, [1024]> var_2264_to_fp16 = const()[name = string("op_2264_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(519986752)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_122_cast_fp16 = linear(bias = var_2264_to_fp16, weight = var_2263_to_fp16, x = var_2244_cast_fp16)[name = string("linear_122_cast_fp16")]; | |
| tensor<int32, [4]> var_2272 = const()[name = string("op_2272"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2273_cast_fp16 = reshape(shape = var_2272, x = linear_120_cast_fp16)[name = string("op_2273_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_208_to_fp16 = const()[name = string("const_208_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_83_cast_fp16 = mul(x = var_2273_cast_fp16, y = const_208_to_fp16)[name = string("q_83_cast_fp16")]; | |
| tensor<int32, [4]> var_2279 = const()[name = string("op_2279"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2280_cast_fp16 = reshape(shape = var_2279, x = linear_121_cast_fp16)[name = string("op_2280_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_209_to_fp16 = const()[name = string("const_209_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_83_cast_fp16 = mul(x = var_2280_cast_fp16, y = const_209_to_fp16)[name = string("k_83_cast_fp16")]; | |
| tensor<int32, [4]> var_2286 = const()[name = string("op_2286"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2287_cast_fp16 = reshape(shape = var_2286, x = linear_122_cast_fp16)[name = string("op_2287_cast_fp16")]; | |
| tensor<int32, [4]> var_2288 = const()[name = string("op_2288"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_41_transpose_x_0 = const()[name = string("qk_41_transpose_x_0"), val = bool(false)]; | |
| bool qk_41_transpose_y_0 = const()[name = string("qk_41_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_136_perm_0 = const()[name = string("transpose_136_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_137_perm_0 = const()[name = string("transpose_137_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_137 = transpose(perm = transpose_137_perm_0, x = k_83_cast_fp16)[name = string("transpose_157")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_136 = transpose(perm = transpose_136_perm_0, x = q_83_cast_fp16)[name = string("transpose_158")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_41_cast_fp16 = matmul(transpose_x = qk_41_transpose_x_0, transpose_y = qk_41_transpose_y_0, x = transpose_136, y = transpose_137)[name = string("qk_41_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_2292_cast_fp16 = softmax(axis = var_2228, x = qk_41_cast_fp16)[name = string("op_2292_cast_fp16")]; | |
| bool var_2294_transpose_x_0 = const()[name = string("op_2294_transpose_x_0"), val = bool(false)]; | |
| bool var_2294_transpose_y_0 = const()[name = string("op_2294_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_83_cast_fp16 = transpose(perm = var_2288, x = var_2287_cast_fp16)[name = string("transpose_159")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_2294_cast_fp16 = matmul(transpose_x = var_2294_transpose_x_0, transpose_y = var_2294_transpose_y_0, x = var_2292_cast_fp16, y = v_83_cast_fp16)[name = string("op_2294_cast_fp16")]; | |
| tensor<int32, [4]> var_2295 = const()[name = string("op_2295"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_20 = const()[name = string("concat_20"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2296_cast_fp16 = transpose(perm = var_2295, x = var_2294_cast_fp16)[name = string("transpose_156")]; | |
| tensor<fp16, [1, 1500, 1024]> x_251_cast_fp16 = reshape(shape = concat_20, x = var_2296_cast_fp16)[name = string("x_251_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2300_to_fp16 = const()[name = string("op_2300_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(519988864)))]; | |
| tensor<fp16, [1024]> var_2301_to_fp16 = const()[name = string("op_2301_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(522086080)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_123_cast_fp16 = linear(bias = var_2301_to_fp16, weight = var_2300_to_fp16, x = x_251_cast_fp16)[name = string("linear_123_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_253_cast_fp16 = add(x = x_247_cast_fp16, y = linear_123_cast_fp16)[name = string("x_253_cast_fp16")]; | |
| tensor<int32, [1]> var_2308_axes_0 = const()[name = string("op_2308_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_20_mlp_ln_weight_to_fp16 = const()[name = string("blocks_20_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(522088192)))]; | |
| tensor<fp16, [1024]> blocks_20_mlp_ln_bias_to_fp16 = const()[name = string("blocks_20_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(522090304)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_2308_cast_fp16 = layer_norm(axes = var_2308_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2234_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast_fp16)[name = string("op_2308_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_2317_to_fp16 = const()[name = string("op_2317_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(522092416)))]; | |
| tensor<fp16, [4096]> var_2318_to_fp16 = const()[name = string("op_2318_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(530481088)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_124_cast_fp16 = linear(bias = var_2318_to_fp16, weight = var_2317_to_fp16, x = var_2308_cast_fp16)[name = string("linear_124_cast_fp16")]; | |
| string x_257_mode_0 = const()[name = string("x_257_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_257_cast_fp16 = gelu(mode = x_257_mode_0, x = linear_124_cast_fp16)[name = string("x_257_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_2323_to_fp16 = const()[name = string("op_2323_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(530489344)))]; | |
| tensor<fp16, [1024]> var_2324_to_fp16 = const()[name = string("op_2324_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(538878016)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_125_cast_fp16 = linear(bias = var_2324_to_fp16, weight = var_2323_to_fp16, x = x_257_cast_fp16)[name = string("linear_125_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_259_cast_fp16 = add(x = x_253_cast_fp16, y = linear_125_cast_fp16)[name = string("x_259_cast_fp16")]; | |
| int32 var_2334 = const()[name = string("op_2334"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2350_axes_0 = const()[name = string("op_2350_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_21_attn_ln_weight_to_fp16 = const()[name = string("blocks_21_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(538880128)))]; | |
| tensor<fp16, [1024]> blocks_21_attn_ln_bias_to_fp16 = const()[name = string("blocks_21_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(538882240)))]; | |
| fp16 var_2340_to_fp16 = const()[name = string("op_2340_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_2350_cast_fp16 = layer_norm(axes = var_2350_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2340_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast_fp16)[name = string("op_2350_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2361_to_fp16 = const()[name = string("op_2361_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(538884352)))]; | |
| tensor<fp16, [1024]> var_2362_to_fp16 = const()[name = string("op_2362_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(540981568)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_126_cast_fp16 = linear(bias = var_2362_to_fp16, weight = var_2361_to_fp16, x = var_2350_cast_fp16)[name = string("linear_126_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2365_to_fp16 = const()[name = string("op_2365_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(540983680)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2365_to_fp16, x = var_2350_cast_fp16)[name = string("linear_127_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2369_to_fp16 = const()[name = string("op_2369_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(543080896)))]; | |
| tensor<fp16, [1024]> var_2370_to_fp16 = const()[name = string("op_2370_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(545178112)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_128_cast_fp16 = linear(bias = var_2370_to_fp16, weight = var_2369_to_fp16, x = var_2350_cast_fp16)[name = string("linear_128_cast_fp16")]; | |
| tensor<int32, [4]> var_2378 = const()[name = string("op_2378"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2379_cast_fp16 = reshape(shape = var_2378, x = linear_126_cast_fp16)[name = string("op_2379_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_210_to_fp16 = const()[name = string("const_210_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_87_cast_fp16 = mul(x = var_2379_cast_fp16, y = const_210_to_fp16)[name = string("q_87_cast_fp16")]; | |
| tensor<int32, [4]> var_2385 = const()[name = string("op_2385"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2386_cast_fp16 = reshape(shape = var_2385, x = linear_127_cast_fp16)[name = string("op_2386_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_211_to_fp16 = const()[name = string("const_211_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_87_cast_fp16 = mul(x = var_2386_cast_fp16, y = const_211_to_fp16)[name = string("k_87_cast_fp16")]; | |
| tensor<int32, [4]> var_2392 = const()[name = string("op_2392"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2393_cast_fp16 = reshape(shape = var_2392, x = linear_128_cast_fp16)[name = string("op_2393_cast_fp16")]; | |
| tensor<int32, [4]> var_2394 = const()[name = string("op_2394"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_43_transpose_x_0 = const()[name = string("qk_43_transpose_x_0"), val = bool(false)]; | |
| bool qk_43_transpose_y_0 = const()[name = string("qk_43_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_138_perm_0 = const()[name = string("transpose_138_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_139_perm_0 = const()[name = string("transpose_139_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_139 = transpose(perm = transpose_139_perm_0, x = k_87_cast_fp16)[name = string("transpose_153")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_138 = transpose(perm = transpose_138_perm_0, x = q_87_cast_fp16)[name = string("transpose_154")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_43_cast_fp16 = matmul(transpose_x = qk_43_transpose_x_0, transpose_y = qk_43_transpose_y_0, x = transpose_138, y = transpose_139)[name = string("qk_43_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_2398_cast_fp16 = softmax(axis = var_2334, x = qk_43_cast_fp16)[name = string("op_2398_cast_fp16")]; | |
| bool var_2400_transpose_x_0 = const()[name = string("op_2400_transpose_x_0"), val = bool(false)]; | |
| bool var_2400_transpose_y_0 = const()[name = string("op_2400_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_87_cast_fp16 = transpose(perm = var_2394, x = var_2393_cast_fp16)[name = string("transpose_155")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_2400_cast_fp16 = matmul(transpose_x = var_2400_transpose_x_0, transpose_y = var_2400_transpose_y_0, x = var_2398_cast_fp16, y = v_87_cast_fp16)[name = string("op_2400_cast_fp16")]; | |
| tensor<int32, [4]> var_2401 = const()[name = string("op_2401"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_21 = const()[name = string("concat_21"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2402_cast_fp16 = transpose(perm = var_2401, x = var_2400_cast_fp16)[name = string("transpose_152")]; | |
| tensor<fp16, [1, 1500, 1024]> x_263_cast_fp16 = reshape(shape = concat_21, x = var_2402_cast_fp16)[name = string("x_263_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2406_to_fp16 = const()[name = string("op_2406_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(545180224)))]; | |
| tensor<fp16, [1024]> var_2407_to_fp16 = const()[name = string("op_2407_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(547277440)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_129_cast_fp16 = linear(bias = var_2407_to_fp16, weight = var_2406_to_fp16, x = x_263_cast_fp16)[name = string("linear_129_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_265_cast_fp16 = add(x = x_259_cast_fp16, y = linear_129_cast_fp16)[name = string("x_265_cast_fp16")]; | |
| tensor<int32, [1]> var_2414_axes_0 = const()[name = string("op_2414_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_21_mlp_ln_weight_to_fp16 = const()[name = string("blocks_21_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(547279552)))]; | |
| tensor<fp16, [1024]> blocks_21_mlp_ln_bias_to_fp16 = const()[name = string("blocks_21_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(547281664)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_2414_cast_fp16 = layer_norm(axes = var_2414_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2340_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast_fp16)[name = string("op_2414_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_2423_to_fp16 = const()[name = string("op_2423_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(547283776)))]; | |
| tensor<fp16, [4096]> var_2424_to_fp16 = const()[name = string("op_2424_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(555672448)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_130_cast_fp16 = linear(bias = var_2424_to_fp16, weight = var_2423_to_fp16, x = var_2414_cast_fp16)[name = string("linear_130_cast_fp16")]; | |
| string x_269_mode_0 = const()[name = string("x_269_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_269_cast_fp16 = gelu(mode = x_269_mode_0, x = linear_130_cast_fp16)[name = string("x_269_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_2429_to_fp16 = const()[name = string("op_2429_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(555680704)))]; | |
| tensor<fp16, [1024]> var_2430_to_fp16 = const()[name = string("op_2430_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(564069376)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_131_cast_fp16 = linear(bias = var_2430_to_fp16, weight = var_2429_to_fp16, x = x_269_cast_fp16)[name = string("linear_131_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_271_cast_fp16 = add(x = x_265_cast_fp16, y = linear_131_cast_fp16)[name = string("x_271_cast_fp16")]; | |
| int32 var_2440 = const()[name = string("op_2440"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2456_axes_0 = const()[name = string("op_2456_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_22_attn_ln_weight_to_fp16 = const()[name = string("blocks_22_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(564071488)))]; | |
| tensor<fp16, [1024]> blocks_22_attn_ln_bias_to_fp16 = const()[name = string("blocks_22_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(564073600)))]; | |
| fp16 var_2446_to_fp16 = const()[name = string("op_2446_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_2456_cast_fp16 = layer_norm(axes = var_2456_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2446_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast_fp16)[name = string("op_2456_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2467_to_fp16 = const()[name = string("op_2467_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(564075712)))]; | |
| tensor<fp16, [1024]> var_2468_to_fp16 = const()[name = string("op_2468_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(566172928)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_132_cast_fp16 = linear(bias = var_2468_to_fp16, weight = var_2467_to_fp16, x = var_2456_cast_fp16)[name = string("linear_132_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2471_to_fp16 = const()[name = string("op_2471_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(566175040)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_133_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2471_to_fp16, x = var_2456_cast_fp16)[name = string("linear_133_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2475_to_fp16 = const()[name = string("op_2475_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(568272256)))]; | |
| tensor<fp16, [1024]> var_2476_to_fp16 = const()[name = string("op_2476_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(570369472)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_134_cast_fp16 = linear(bias = var_2476_to_fp16, weight = var_2475_to_fp16, x = var_2456_cast_fp16)[name = string("linear_134_cast_fp16")]; | |
| tensor<int32, [4]> var_2484 = const()[name = string("op_2484"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2485_cast_fp16 = reshape(shape = var_2484, x = linear_132_cast_fp16)[name = string("op_2485_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_212_to_fp16 = const()[name = string("const_212_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_91_cast_fp16 = mul(x = var_2485_cast_fp16, y = const_212_to_fp16)[name = string("q_91_cast_fp16")]; | |
| tensor<int32, [4]> var_2491 = const()[name = string("op_2491"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2492_cast_fp16 = reshape(shape = var_2491, x = linear_133_cast_fp16)[name = string("op_2492_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_213_to_fp16 = const()[name = string("const_213_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_91_cast_fp16 = mul(x = var_2492_cast_fp16, y = const_213_to_fp16)[name = string("k_91_cast_fp16")]; | |
| tensor<int32, [4]> var_2498 = const()[name = string("op_2498"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2499_cast_fp16 = reshape(shape = var_2498, x = linear_134_cast_fp16)[name = string("op_2499_cast_fp16")]; | |
| tensor<int32, [4]> var_2500 = const()[name = string("op_2500"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_45_transpose_x_0 = const()[name = string("qk_45_transpose_x_0"), val = bool(false)]; | |
| bool qk_45_transpose_y_0 = const()[name = string("qk_45_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_140_perm_0 = const()[name = string("transpose_140_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_141_perm_0 = const()[name = string("transpose_141_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_141 = transpose(perm = transpose_141_perm_0, x = k_91_cast_fp16)[name = string("transpose_149")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_140 = transpose(perm = transpose_140_perm_0, x = q_91_cast_fp16)[name = string("transpose_150")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_45_cast_fp16 = matmul(transpose_x = qk_45_transpose_x_0, transpose_y = qk_45_transpose_y_0, x = transpose_140, y = transpose_141)[name = string("qk_45_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_2504_cast_fp16 = softmax(axis = var_2440, x = qk_45_cast_fp16)[name = string("op_2504_cast_fp16")]; | |
| bool var_2506_transpose_x_0 = const()[name = string("op_2506_transpose_x_0"), val = bool(false)]; | |
| bool var_2506_transpose_y_0 = const()[name = string("op_2506_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_91_cast_fp16 = transpose(perm = var_2500, x = var_2499_cast_fp16)[name = string("transpose_151")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_2506_cast_fp16 = matmul(transpose_x = var_2506_transpose_x_0, transpose_y = var_2506_transpose_y_0, x = var_2504_cast_fp16, y = v_91_cast_fp16)[name = string("op_2506_cast_fp16")]; | |
| tensor<int32, [4]> var_2507 = const()[name = string("op_2507"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_22 = const()[name = string("concat_22"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2508_cast_fp16 = transpose(perm = var_2507, x = var_2506_cast_fp16)[name = string("transpose_148")]; | |
| tensor<fp16, [1, 1500, 1024]> x_275_cast_fp16 = reshape(shape = concat_22, x = var_2508_cast_fp16)[name = string("x_275_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2512_to_fp16 = const()[name = string("op_2512_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(570371584)))]; | |
| tensor<fp16, [1024]> var_2513_to_fp16 = const()[name = string("op_2513_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(572468800)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_135_cast_fp16 = linear(bias = var_2513_to_fp16, weight = var_2512_to_fp16, x = x_275_cast_fp16)[name = string("linear_135_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_277_cast_fp16 = add(x = x_271_cast_fp16, y = linear_135_cast_fp16)[name = string("x_277_cast_fp16")]; | |
| tensor<int32, [1]> var_2520_axes_0 = const()[name = string("op_2520_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_22_mlp_ln_weight_to_fp16 = const()[name = string("blocks_22_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(572470912)))]; | |
| tensor<fp16, [1024]> blocks_22_mlp_ln_bias_to_fp16 = const()[name = string("blocks_22_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(572473024)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_2520_cast_fp16 = layer_norm(axes = var_2520_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2446_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast_fp16)[name = string("op_2520_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_2529_to_fp16 = const()[name = string("op_2529_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(572475136)))]; | |
| tensor<fp16, [4096]> var_2530_to_fp16 = const()[name = string("op_2530_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(580863808)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_136_cast_fp16 = linear(bias = var_2530_to_fp16, weight = var_2529_to_fp16, x = var_2520_cast_fp16)[name = string("linear_136_cast_fp16")]; | |
| string x_281_mode_0 = const()[name = string("x_281_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_281_cast_fp16 = gelu(mode = x_281_mode_0, x = linear_136_cast_fp16)[name = string("x_281_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_2535_to_fp16 = const()[name = string("op_2535_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(580872064)))]; | |
| tensor<fp16, [1024]> var_2536_to_fp16 = const()[name = string("op_2536_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(589260736)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_137_cast_fp16 = linear(bias = var_2536_to_fp16, weight = var_2535_to_fp16, x = x_281_cast_fp16)[name = string("linear_137_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_283_cast_fp16 = add(x = x_277_cast_fp16, y = linear_137_cast_fp16)[name = string("x_283_cast_fp16")]; | |
| int32 var_2546 = const()[name = string("op_2546"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2562_axes_0 = const()[name = string("op_2562_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_23_attn_ln_weight_to_fp16 = const()[name = string("blocks_23_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(589262848)))]; | |
| tensor<fp16, [1024]> blocks_23_attn_ln_bias_to_fp16 = const()[name = string("blocks_23_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(589264960)))]; | |
| fp16 var_2552_to_fp16 = const()[name = string("op_2552_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> var_2562_cast_fp16 = layer_norm(axes = var_2562_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2552_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast_fp16)[name = string("op_2562_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2573_to_fp16 = const()[name = string("op_2573_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(589267072)))]; | |
| tensor<fp16, [1024]> var_2574_to_fp16 = const()[name = string("op_2574_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(591364288)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_138_cast_fp16 = linear(bias = var_2574_to_fp16, weight = var_2573_to_fp16, x = var_2562_cast_fp16)[name = string("linear_138_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2577_to_fp16 = const()[name = string("op_2577_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(591366400)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_139_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2577_to_fp16, x = var_2562_cast_fp16)[name = string("linear_139_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2581_to_fp16 = const()[name = string("op_2581_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(593463616)))]; | |
| tensor<fp16, [1024]> var_2582_to_fp16 = const()[name = string("op_2582_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(595560832)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_140_cast_fp16 = linear(bias = var_2582_to_fp16, weight = var_2581_to_fp16, x = var_2562_cast_fp16)[name = string("linear_140_cast_fp16")]; | |
| tensor<int32, [4]> var_2590 = const()[name = string("op_2590"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2591_cast_fp16 = reshape(shape = var_2590, x = linear_138_cast_fp16)[name = string("op_2591_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_214_to_fp16 = const()[name = string("const_214_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> q_cast_fp16 = mul(x = var_2591_cast_fp16, y = const_214_to_fp16)[name = string("q_cast_fp16")]; | |
| tensor<int32, [4]> var_2597 = const()[name = string("op_2597"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2598_cast_fp16 = reshape(shape = var_2597, x = linear_139_cast_fp16)[name = string("op_2598_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_215_to_fp16 = const()[name = string("const_215_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 16, 64]> k_cast_fp16 = mul(x = var_2598_cast_fp16, y = const_215_to_fp16)[name = string("k_cast_fp16")]; | |
| tensor<int32, [4]> var_2604 = const()[name = string("op_2604"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2605_cast_fp16 = reshape(shape = var_2604, x = linear_140_cast_fp16)[name = string("op_2605_cast_fp16")]; | |
| tensor<int32, [4]> var_2606 = const()[name = string("op_2606"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_transpose_x_0 = const()[name = string("qk_transpose_x_0"), val = bool(false)]; | |
| bool qk_transpose_y_0 = const()[name = string("qk_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_142_perm_0 = const()[name = string("transpose_142_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_143_perm_0 = const()[name = string("transpose_143_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 16, 64, 1500]> transpose_143 = transpose(perm = transpose_143_perm_0, x = k_cast_fp16)[name = string("transpose_145")]; | |
| tensor<fp16, [1, 16, 1500, 64]> transpose_142 = transpose(perm = transpose_142_perm_0, x = q_cast_fp16)[name = string("transpose_146")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> qk_cast_fp16 = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_142, y = transpose_143)[name = string("qk_cast_fp16")]; | |
| tensor<fp16, [1, 16, 1500, 1500]> var_2610_cast_fp16 = softmax(axis = var_2546, x = qk_cast_fp16)[name = string("op_2610_cast_fp16")]; | |
| bool var_2612_transpose_x_0 = const()[name = string("op_2612_transpose_x_0"), val = bool(false)]; | |
| bool var_2612_transpose_y_0 = const()[name = string("op_2612_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 16, 1500, 64]> v_cast_fp16 = transpose(perm = var_2606, x = var_2605_cast_fp16)[name = string("transpose_147")]; | |
| tensor<fp16, [1, 16, 1500, 64]> var_2612_cast_fp16 = matmul(transpose_x = var_2612_transpose_x_0, transpose_y = var_2612_transpose_y_0, x = var_2610_cast_fp16, y = v_cast_fp16)[name = string("op_2612_cast_fp16")]; | |
| tensor<int32, [4]> var_2613 = const()[name = string("op_2613"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_23 = const()[name = string("concat_23"), val = tensor<int32, [3]>([1, 1500, 1024])]; | |
| tensor<fp16, [1, 1500, 16, 64]> var_2614_cast_fp16 = transpose(perm = var_2613, x = var_2612_cast_fp16)[name = string("transpose_144")]; | |
| tensor<fp16, [1, 1500, 1024]> x_287_cast_fp16 = reshape(shape = concat_23, x = var_2614_cast_fp16)[name = string("x_287_cast_fp16")]; | |
| tensor<fp16, [1024, 1024]> var_2618_to_fp16 = const()[name = string("op_2618_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(595562944)))]; | |
| tensor<fp16, [1024]> var_2619_to_fp16 = const()[name = string("op_2619_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(597660160)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_141_cast_fp16 = linear(bias = var_2619_to_fp16, weight = var_2618_to_fp16, x = x_287_cast_fp16)[name = string("linear_141_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_289_cast_fp16 = add(x = x_283_cast_fp16, y = linear_141_cast_fp16)[name = string("x_289_cast_fp16")]; | |
| tensor<int32, [1]> var_2626_axes_0 = const()[name = string("op_2626_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> blocks_23_mlp_ln_weight_to_fp16 = const()[name = string("blocks_23_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(597662272)))]; | |
| tensor<fp16, [1024]> blocks_23_mlp_ln_bias_to_fp16 = const()[name = string("blocks_23_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(597664384)))]; | |
| tensor<fp16, [1, 1500, 1024]> var_2626_cast_fp16 = layer_norm(axes = var_2626_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2552_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast_fp16)[name = string("op_2626_cast_fp16")]; | |
| tensor<fp16, [4096, 1024]> var_2635_to_fp16 = const()[name = string("op_2635_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(597666496)))]; | |
| tensor<fp16, [4096]> var_2636_to_fp16 = const()[name = string("op_2636_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(606055168)))]; | |
| tensor<fp16, [1, 1500, 4096]> linear_142_cast_fp16 = linear(bias = var_2636_to_fp16, weight = var_2635_to_fp16, x = var_2626_cast_fp16)[name = string("linear_142_cast_fp16")]; | |
| string x_293_mode_0 = const()[name = string("x_293_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 4096]> x_293_cast_fp16 = gelu(mode = x_293_mode_0, x = linear_142_cast_fp16)[name = string("x_293_cast_fp16")]; | |
| tensor<fp16, [1024, 4096]> var_2641_to_fp16 = const()[name = string("op_2641_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(606063424)))]; | |
| tensor<fp16, [1024]> var_2642_to_fp16 = const()[name = string("op_2642_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614452096)))]; | |
| tensor<fp16, [1, 1500, 1024]> linear_143_cast_fp16 = linear(bias = var_2642_to_fp16, weight = var_2641_to_fp16, x = x_293_cast_fp16)[name = string("linear_143_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1024]> x_cast_fp16 = add(x = x_289_cast_fp16, y = linear_143_cast_fp16)[name = string("x_cast_fp16")]; | |
| tensor<int32, [1]> var_2655_axes_0 = const()[name = string("op_2655_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1024]> ln_post_weight_to_fp16 = const()[name = string("ln_post_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614454208)))]; | |
| tensor<fp16, [1024]> ln_post_bias_to_fp16 = const()[name = string("ln_post_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614456320)))]; | |
| fp16 var_2646_to_fp16 = const()[name = string("op_2646_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1024]> output = layer_norm(axes = var_2655_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_2646_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = string("op_2655_cast_fp16")]; | |
| } -> (output); | |
| } |