program(1.0) [buildInfo = dict, tensor>({{"coremlc-component-MIL", "3304.5.2"}, {"coremlc-version", "3304.6.2"}, {"coremltools-component-torch", "2.2.2"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "7.2"}})] { func main(tensor logmel_data) { tensor var_16 = const()[name = tensor("op_16"), val = tensor(1)]; tensor var_24 = const()[name = tensor("op_24"), val = tensor([1])]; tensor var_26 = const()[name = tensor("op_26"), val = tensor([1])]; tensor var_28_pad_type_0 = const()[name = tensor("op_28_pad_type_0"), val = tensor("custom")]; tensor var_28_pad_0 = const()[name = tensor("op_28_pad_0"), val = tensor([1, 1])]; tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184448)))]; tensor var_28_cast_fp16 = conv(bias = bias_3_to_fp16, dilations = var_26, groups = var_16, pad = var_28_pad_0, pad_type = var_28_pad_type_0, strides = var_24, weight = weight_3_to_fp16, x = logmel_data)[name = tensor("op_28_cast_fp16")]; tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; tensor input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_28_cast_fp16)[name = tensor("input_1_cast_fp16")]; tensor var_33 = const()[name = tensor("op_33"), val = tensor(1)]; tensor var_42 = const()[name = tensor("op_42"), val = tensor([2])]; tensor var_44 = const()[name = tensor("op_44"), val = tensor([1])]; tensor var_46_pad_type_0 = const()[name = tensor("op_46_pad_type_0"), val = tensor("custom")]; tensor var_46_pad_0 = const()[name = tensor("op_46_pad_0"), val = tensor([1, 1])]; tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185280)))]; tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1070080)))]; tensor var_46_cast_fp16 = conv(bias = bias_7_to_fp16, dilations = var_44, groups = var_33, pad = var_46_pad_0, pad_type = var_46_pad_type_0, strides = var_42, weight = weight_7_to_fp16, x = input_1_cast_fp16)[name = tensor("op_46_cast_fp16")]; tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; tensor x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_46_cast_fp16)[name = tensor("x_3_cast_fp16")]; tensor var_52 = const()[name = tensor("op_52"), val = tensor([0, 2, 1])]; tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1070912)))]; tensor transpose_40 = transpose(perm = var_52, x = x_3_cast_fp16)[name = tensor("transpose_40")]; tensor var_55_cast_fp16 = add(x = transpose_40, y = positional_embedding_to_fp16)[name = tensor("op_55_cast_fp16")]; tensor var_67 = const()[name = tensor("op_67"), val = tensor(-1)]; tensor var_83_axes_0 = const()[name = tensor("op_83_axes_0"), val = tensor([-1])]; tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2222976)))]; tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2223808)))]; tensor var_73_to_fp16 = const()[name = tensor("op_73_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_83_cast_fp16 = layer_norm(axes = var_83_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_73_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_55_cast_fp16)[name = tensor("op_83_cast_fp16")]; tensor var_94_to_fp16 = const()[name = tensor("op_94_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2224640)))]; tensor var_95_to_fp16 = const()[name = tensor("op_95_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2519616)))]; tensor linear_0_cast_fp16 = linear(bias = var_95_to_fp16, weight = var_94_to_fp16, x = var_83_cast_fp16)[name = tensor("linear_0_cast_fp16")]; tensor var_98_to_fp16 = const()[name = tensor("op_98_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2520448)))]; tensor linear_1_bias_0_to_fp16 = const()[name = tensor("linear_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2815424)))]; tensor linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_98_to_fp16, x = var_83_cast_fp16)[name = tensor("linear_1_cast_fp16")]; tensor var_102_to_fp16 = const()[name = tensor("op_102_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2816256)))]; tensor var_103_to_fp16 = const()[name = tensor("op_103_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3111232)))]; tensor linear_2_cast_fp16 = linear(bias = var_103_to_fp16, weight = var_102_to_fp16, x = var_83_cast_fp16)[name = tensor("linear_2_cast_fp16")]; tensor var_111 = const()[name = tensor("op_111"), val = tensor([1, 1500, 6, -1])]; tensor var_112_cast_fp16 = reshape(shape = var_111, x = linear_0_cast_fp16)[name = tensor("op_112_cast_fp16")]; tensor const_28_to_fp16 = const()[name = tensor("const_28_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor q_3_cast_fp16 = mul(x = var_112_cast_fp16, y = const_28_to_fp16)[name = tensor("q_3_cast_fp16")]; tensor var_118 = const()[name = tensor("op_118"), val = tensor([1, 1500, 6, -1])]; tensor var_119_cast_fp16 = reshape(shape = var_118, x = linear_1_cast_fp16)[name = tensor("op_119_cast_fp16")]; tensor const_29_to_fp16 = const()[name = tensor("const_29_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor k_3_cast_fp16 = mul(x = var_119_cast_fp16, y = const_29_to_fp16)[name = tensor("k_3_cast_fp16")]; tensor var_125 = const()[name = tensor("op_125"), val = tensor([1, 1500, 6, -1])]; tensor var_126_cast_fp16 = reshape(shape = var_125, x = linear_2_cast_fp16)[name = tensor("op_126_cast_fp16")]; tensor var_127 = const()[name = tensor("op_127"), val = tensor([0, 2, 1, 3])]; tensor qk_1_transpose_x_0 = const()[name = tensor("qk_1_transpose_x_0"), val = tensor(false)]; tensor qk_1_transpose_y_0 = const()[name = tensor("qk_1_transpose_y_0"), val = tensor(false)]; tensor transpose_16_perm_0 = const()[name = tensor("transpose_16_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_17_perm_0 = const()[name = tensor("transpose_17_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_37 = transpose(perm = transpose_17_perm_0, x = k_3_cast_fp16)[name = tensor("transpose_37")]; tensor transpose_38 = transpose(perm = transpose_16_perm_0, x = q_3_cast_fp16)[name = tensor("transpose_38")]; tensor qk_1_cast_fp16 = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_38, y = transpose_37)[name = tensor("qk_1_cast_fp16")]; tensor var_131_cast_fp16 = softmax(axis = var_67, x = qk_1_cast_fp16)[name = tensor("op_131_cast_fp16")]; tensor var_133_transpose_x_0 = const()[name = tensor("op_133_transpose_x_0"), val = tensor(false)]; tensor var_133_transpose_y_0 = const()[name = tensor("op_133_transpose_y_0"), val = tensor(false)]; tensor transpose_39 = transpose(perm = var_127, x = var_126_cast_fp16)[name = tensor("transpose_39")]; tensor var_133_cast_fp16 = matmul(transpose_x = var_133_transpose_x_0, transpose_y = var_133_transpose_y_0, x = var_131_cast_fp16, y = transpose_39)[name = tensor("op_133_cast_fp16")]; tensor var_134 = const()[name = tensor("op_134"), val = tensor([0, 2, 1, 3])]; tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 384])]; tensor transpose_36 = transpose(perm = var_134, x = var_133_cast_fp16)[name = tensor("transpose_36")]; tensor x_11_cast_fp16 = reshape(shape = concat_0, x = transpose_36)[name = tensor("x_11_cast_fp16")]; tensor var_139_to_fp16 = const()[name = tensor("op_139_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3112064)))]; tensor var_140_to_fp16 = const()[name = tensor("op_140_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3407040)))]; tensor linear_3_cast_fp16 = linear(bias = var_140_to_fp16, weight = var_139_to_fp16, x = x_11_cast_fp16)[name = tensor("linear_3_cast_fp16")]; tensor x_13_cast_fp16 = add(x = var_55_cast_fp16, y = linear_3_cast_fp16)[name = tensor("x_13_cast_fp16")]; tensor var_147_axes_0 = const()[name = tensor("op_147_axes_0"), val = tensor([-1])]; tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3407872)))]; tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3408704)))]; tensor var_147_cast_fp16 = layer_norm(axes = var_147_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_73_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = tensor("op_147_cast_fp16")]; tensor var_156_to_fp16 = const()[name = tensor("op_156_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3409536)))]; tensor var_157_to_fp16 = const()[name = tensor("op_157_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4589248)))]; tensor linear_4_cast_fp16 = linear(bias = var_157_to_fp16, weight = var_156_to_fp16, x = var_147_cast_fp16)[name = tensor("linear_4_cast_fp16")]; tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; tensor x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = tensor("x_17_cast_fp16")]; tensor var_162_to_fp16 = const()[name = tensor("op_162_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4592384)))]; tensor var_163_to_fp16 = const()[name = tensor("op_163_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5772096)))]; tensor linear_5_cast_fp16 = linear(bias = var_163_to_fp16, weight = var_162_to_fp16, x = x_17_cast_fp16)[name = tensor("linear_5_cast_fp16")]; tensor x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = tensor("x_19_cast_fp16")]; tensor var_172 = const()[name = tensor("op_172"), val = tensor(-1)]; tensor var_188_axes_0 = const()[name = tensor("op_188_axes_0"), val = tensor([-1])]; tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5772928)))]; tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5773760)))]; tensor var_178_to_fp16 = const()[name = tensor("op_178_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_188_cast_fp16 = layer_norm(axes = var_188_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_178_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = tensor("op_188_cast_fp16")]; tensor var_199_to_fp16 = const()[name = tensor("op_199_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5774592)))]; tensor var_200_to_fp16 = const()[name = tensor("op_200_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6069568)))]; tensor linear_6_cast_fp16 = linear(bias = var_200_to_fp16, weight = var_199_to_fp16, x = var_188_cast_fp16)[name = tensor("linear_6_cast_fp16")]; tensor var_203_to_fp16 = const()[name = tensor("op_203_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6070400)))]; tensor linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_203_to_fp16, x = var_188_cast_fp16)[name = tensor("linear_7_cast_fp16")]; tensor var_207_to_fp16 = const()[name = tensor("op_207_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6365376)))]; tensor var_208_to_fp16 = const()[name = tensor("op_208_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6660352)))]; tensor linear_8_cast_fp16 = linear(bias = var_208_to_fp16, weight = var_207_to_fp16, x = var_188_cast_fp16)[name = tensor("linear_8_cast_fp16")]; tensor var_216 = const()[name = tensor("op_216"), val = tensor([1, 1500, 6, -1])]; tensor var_217_cast_fp16 = reshape(shape = var_216, x = linear_6_cast_fp16)[name = tensor("op_217_cast_fp16")]; tensor const_30_to_fp16 = const()[name = tensor("const_30_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor q_7_cast_fp16 = mul(x = var_217_cast_fp16, y = const_30_to_fp16)[name = tensor("q_7_cast_fp16")]; tensor var_223 = const()[name = tensor("op_223"), val = tensor([1, 1500, 6, -1])]; tensor var_224_cast_fp16 = reshape(shape = var_223, x = linear_7_cast_fp16)[name = tensor("op_224_cast_fp16")]; tensor const_31_to_fp16 = const()[name = tensor("const_31_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor k_7_cast_fp16 = mul(x = var_224_cast_fp16, y = const_31_to_fp16)[name = tensor("k_7_cast_fp16")]; tensor var_230 = const()[name = tensor("op_230"), val = tensor([1, 1500, 6, -1])]; tensor var_231_cast_fp16 = reshape(shape = var_230, x = linear_8_cast_fp16)[name = tensor("op_231_cast_fp16")]; tensor var_232 = const()[name = tensor("op_232"), val = tensor([0, 2, 1, 3])]; tensor qk_3_transpose_x_0 = const()[name = tensor("qk_3_transpose_x_0"), val = tensor(false)]; tensor qk_3_transpose_y_0 = const()[name = tensor("qk_3_transpose_y_0"), val = tensor(false)]; tensor transpose_18_perm_0 = const()[name = tensor("transpose_18_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_19_perm_0 = const()[name = tensor("transpose_19_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_33 = transpose(perm = transpose_19_perm_0, x = k_7_cast_fp16)[name = tensor("transpose_33")]; tensor transpose_34 = transpose(perm = transpose_18_perm_0, x = q_7_cast_fp16)[name = tensor("transpose_34")]; tensor qk_3_cast_fp16 = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_34, y = transpose_33)[name = tensor("qk_3_cast_fp16")]; tensor var_236_cast_fp16 = softmax(axis = var_172, x = qk_3_cast_fp16)[name = tensor("op_236_cast_fp16")]; tensor var_238_transpose_x_0 = const()[name = tensor("op_238_transpose_x_0"), val = tensor(false)]; tensor var_238_transpose_y_0 = const()[name = tensor("op_238_transpose_y_0"), val = tensor(false)]; tensor transpose_35 = transpose(perm = var_232, x = var_231_cast_fp16)[name = tensor("transpose_35")]; tensor var_238_cast_fp16 = matmul(transpose_x = var_238_transpose_x_0, transpose_y = var_238_transpose_y_0, x = var_236_cast_fp16, y = transpose_35)[name = tensor("op_238_cast_fp16")]; tensor var_239 = const()[name = tensor("op_239"), val = tensor([0, 2, 1, 3])]; tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 384])]; tensor transpose_32 = transpose(perm = var_239, x = var_238_cast_fp16)[name = tensor("transpose_32")]; tensor x_23_cast_fp16 = reshape(shape = concat_1, x = transpose_32)[name = tensor("x_23_cast_fp16")]; tensor var_244_to_fp16 = const()[name = tensor("op_244_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6661184)))]; tensor var_245_to_fp16 = const()[name = tensor("op_245_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6956160)))]; tensor linear_9_cast_fp16 = linear(bias = var_245_to_fp16, weight = var_244_to_fp16, x = x_23_cast_fp16)[name = tensor("linear_9_cast_fp16")]; tensor x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = tensor("x_25_cast_fp16")]; tensor var_252_axes_0 = const()[name = tensor("op_252_axes_0"), val = tensor([-1])]; tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6956992)))]; tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6957824)))]; tensor var_252_cast_fp16 = layer_norm(axes = var_252_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_178_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = tensor("op_252_cast_fp16")]; tensor var_261_to_fp16 = const()[name = tensor("op_261_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6958656)))]; tensor var_262_to_fp16 = const()[name = tensor("op_262_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8138368)))]; tensor linear_10_cast_fp16 = linear(bias = var_262_to_fp16, weight = var_261_to_fp16, x = var_252_cast_fp16)[name = tensor("linear_10_cast_fp16")]; tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; tensor x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = tensor("x_29_cast_fp16")]; tensor var_267_to_fp16 = const()[name = tensor("op_267_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8141504)))]; tensor var_268_to_fp16 = const()[name = tensor("op_268_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9321216)))]; tensor linear_11_cast_fp16 = linear(bias = var_268_to_fp16, weight = var_267_to_fp16, x = x_29_cast_fp16)[name = tensor("linear_11_cast_fp16")]; tensor x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = tensor("x_31_cast_fp16")]; tensor var_277 = const()[name = tensor("op_277"), val = tensor(-1)]; tensor var_293_axes_0 = const()[name = tensor("op_293_axes_0"), val = tensor([-1])]; tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9322048)))]; tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9322880)))]; tensor var_283_to_fp16 = const()[name = tensor("op_283_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_293_cast_fp16 = layer_norm(axes = var_293_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_283_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = tensor("op_293_cast_fp16")]; tensor var_304_to_fp16 = const()[name = tensor("op_304_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9323712)))]; tensor var_305_to_fp16 = const()[name = tensor("op_305_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9618688)))]; tensor linear_12_cast_fp16 = linear(bias = var_305_to_fp16, weight = var_304_to_fp16, x = var_293_cast_fp16)[name = tensor("linear_12_cast_fp16")]; tensor var_308_to_fp16 = const()[name = tensor("op_308_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9619520)))]; tensor linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_308_to_fp16, x = var_293_cast_fp16)[name = tensor("linear_13_cast_fp16")]; tensor var_312_to_fp16 = const()[name = tensor("op_312_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9914496)))]; tensor var_313_to_fp16 = const()[name = tensor("op_313_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10209472)))]; tensor linear_14_cast_fp16 = linear(bias = var_313_to_fp16, weight = var_312_to_fp16, x = var_293_cast_fp16)[name = tensor("linear_14_cast_fp16")]; tensor var_321 = const()[name = tensor("op_321"), val = tensor([1, 1500, 6, -1])]; tensor var_322_cast_fp16 = reshape(shape = var_321, x = linear_12_cast_fp16)[name = tensor("op_322_cast_fp16")]; tensor const_32_to_fp16 = const()[name = tensor("const_32_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor q_11_cast_fp16 = mul(x = var_322_cast_fp16, y = const_32_to_fp16)[name = tensor("q_11_cast_fp16")]; tensor var_328 = const()[name = tensor("op_328"), val = tensor([1, 1500, 6, -1])]; tensor var_329_cast_fp16 = reshape(shape = var_328, x = linear_13_cast_fp16)[name = tensor("op_329_cast_fp16")]; tensor const_33_to_fp16 = const()[name = tensor("const_33_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor k_11_cast_fp16 = mul(x = var_329_cast_fp16, y = const_33_to_fp16)[name = tensor("k_11_cast_fp16")]; tensor var_335 = const()[name = tensor("op_335"), val = tensor([1, 1500, 6, -1])]; tensor var_336_cast_fp16 = reshape(shape = var_335, x = linear_14_cast_fp16)[name = tensor("op_336_cast_fp16")]; tensor var_337 = const()[name = tensor("op_337"), val = tensor([0, 2, 1, 3])]; tensor qk_5_transpose_x_0 = const()[name = tensor("qk_5_transpose_x_0"), val = tensor(false)]; tensor qk_5_transpose_y_0 = const()[name = tensor("qk_5_transpose_y_0"), val = tensor(false)]; tensor transpose_20_perm_0 = const()[name = tensor("transpose_20_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_21_perm_0 = const()[name = tensor("transpose_21_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_29 = transpose(perm = transpose_21_perm_0, x = k_11_cast_fp16)[name = tensor("transpose_29")]; tensor transpose_30 = transpose(perm = transpose_20_perm_0, x = q_11_cast_fp16)[name = tensor("transpose_30")]; tensor qk_5_cast_fp16 = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_30, y = transpose_29)[name = tensor("qk_5_cast_fp16")]; tensor var_341_cast_fp16 = softmax(axis = var_277, x = qk_5_cast_fp16)[name = tensor("op_341_cast_fp16")]; tensor var_343_transpose_x_0 = const()[name = tensor("op_343_transpose_x_0"), val = tensor(false)]; tensor var_343_transpose_y_0 = const()[name = tensor("op_343_transpose_y_0"), val = tensor(false)]; tensor transpose_31 = transpose(perm = var_337, x = var_336_cast_fp16)[name = tensor("transpose_31")]; tensor var_343_cast_fp16 = matmul(transpose_x = var_343_transpose_x_0, transpose_y = var_343_transpose_y_0, x = var_341_cast_fp16, y = transpose_31)[name = tensor("op_343_cast_fp16")]; tensor var_344 = const()[name = tensor("op_344"), val = tensor([0, 2, 1, 3])]; tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 384])]; tensor transpose_28 = transpose(perm = var_344, x = var_343_cast_fp16)[name = tensor("transpose_28")]; tensor x_35_cast_fp16 = reshape(shape = concat_2, x = transpose_28)[name = tensor("x_35_cast_fp16")]; tensor var_349_to_fp16 = const()[name = tensor("op_349_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10210304)))]; tensor var_350_to_fp16 = const()[name = tensor("op_350_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10505280)))]; tensor linear_15_cast_fp16 = linear(bias = var_350_to_fp16, weight = var_349_to_fp16, x = x_35_cast_fp16)[name = tensor("linear_15_cast_fp16")]; tensor x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = tensor("x_37_cast_fp16")]; tensor var_357_axes_0 = const()[name = tensor("op_357_axes_0"), val = tensor([-1])]; tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10506112)))]; tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10506944)))]; tensor var_357_cast_fp16 = layer_norm(axes = var_357_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_283_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = tensor("op_357_cast_fp16")]; tensor var_366_to_fp16 = const()[name = tensor("op_366_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10507776)))]; tensor var_367_to_fp16 = const()[name = tensor("op_367_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11687488)))]; tensor linear_16_cast_fp16 = linear(bias = var_367_to_fp16, weight = var_366_to_fp16, x = var_357_cast_fp16)[name = tensor("linear_16_cast_fp16")]; tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; tensor x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = tensor("x_41_cast_fp16")]; tensor var_372_to_fp16 = const()[name = tensor("op_372_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11690624)))]; tensor var_373_to_fp16 = const()[name = tensor("op_373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12870336)))]; tensor linear_17_cast_fp16 = linear(bias = var_373_to_fp16, weight = var_372_to_fp16, x = x_41_cast_fp16)[name = tensor("linear_17_cast_fp16")]; tensor x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = tensor("x_43_cast_fp16")]; tensor var_382 = const()[name = tensor("op_382"), val = tensor(-1)]; tensor var_398_axes_0 = const()[name = tensor("op_398_axes_0"), val = tensor([-1])]; tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12871168)))]; tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12872000)))]; tensor var_388_to_fp16 = const()[name = tensor("op_388_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_398_cast_fp16 = layer_norm(axes = var_398_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_388_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = tensor("op_398_cast_fp16")]; tensor var_409_to_fp16 = const()[name = tensor("op_409_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12872832)))]; tensor var_410_to_fp16 = const()[name = tensor("op_410_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13167808)))]; tensor linear_18_cast_fp16 = linear(bias = var_410_to_fp16, weight = var_409_to_fp16, x = var_398_cast_fp16)[name = tensor("linear_18_cast_fp16")]; tensor var_413_to_fp16 = const()[name = tensor("op_413_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13168640)))]; tensor linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_413_to_fp16, x = var_398_cast_fp16)[name = tensor("linear_19_cast_fp16")]; tensor var_417_to_fp16 = const()[name = tensor("op_417_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13463616)))]; tensor var_418_to_fp16 = const()[name = tensor("op_418_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13758592)))]; tensor linear_20_cast_fp16 = linear(bias = var_418_to_fp16, weight = var_417_to_fp16, x = var_398_cast_fp16)[name = tensor("linear_20_cast_fp16")]; tensor var_426 = const()[name = tensor("op_426"), val = tensor([1, 1500, 6, -1])]; tensor var_427_cast_fp16 = reshape(shape = var_426, x = linear_18_cast_fp16)[name = tensor("op_427_cast_fp16")]; tensor const_34_to_fp16 = const()[name = tensor("const_34_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor q_cast_fp16 = mul(x = var_427_cast_fp16, y = const_34_to_fp16)[name = tensor("q_cast_fp16")]; tensor var_433 = const()[name = tensor("op_433"), val = tensor([1, 1500, 6, -1])]; tensor var_434_cast_fp16 = reshape(shape = var_433, x = linear_19_cast_fp16)[name = tensor("op_434_cast_fp16")]; tensor const_35_to_fp16 = const()[name = tensor("const_35_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor k_cast_fp16 = mul(x = var_434_cast_fp16, y = const_35_to_fp16)[name = tensor("k_cast_fp16")]; tensor var_440 = const()[name = tensor("op_440"), val = tensor([1, 1500, 6, -1])]; tensor var_441_cast_fp16 = reshape(shape = var_440, x = linear_20_cast_fp16)[name = tensor("op_441_cast_fp16")]; tensor var_442 = const()[name = tensor("op_442"), val = tensor([0, 2, 1, 3])]; tensor qk_transpose_x_0 = const()[name = tensor("qk_transpose_x_0"), val = tensor(false)]; tensor qk_transpose_y_0 = const()[name = tensor("qk_transpose_y_0"), val = tensor(false)]; tensor transpose_22_perm_0 = const()[name = tensor("transpose_22_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_23_perm_0 = const()[name = tensor("transpose_23_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_25 = transpose(perm = transpose_23_perm_0, x = k_cast_fp16)[name = tensor("transpose_25")]; tensor transpose_26 = transpose(perm = transpose_22_perm_0, x = q_cast_fp16)[name = tensor("transpose_26")]; tensor qk_cast_fp16 = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_26, y = transpose_25)[name = tensor("qk_cast_fp16")]; tensor var_446_cast_fp16 = softmax(axis = var_382, x = qk_cast_fp16)[name = tensor("op_446_cast_fp16")]; tensor var_448_transpose_x_0 = const()[name = tensor("op_448_transpose_x_0"), val = tensor(false)]; tensor var_448_transpose_y_0 = const()[name = tensor("op_448_transpose_y_0"), val = tensor(false)]; tensor transpose_27 = transpose(perm = var_442, x = var_441_cast_fp16)[name = tensor("transpose_27")]; tensor var_448_cast_fp16 = matmul(transpose_x = var_448_transpose_x_0, transpose_y = var_448_transpose_y_0, x = var_446_cast_fp16, y = transpose_27)[name = tensor("op_448_cast_fp16")]; tensor var_449 = const()[name = tensor("op_449"), val = tensor([0, 2, 1, 3])]; tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 384])]; tensor transpose_24 = transpose(perm = var_449, x = var_448_cast_fp16)[name = tensor("transpose_24")]; tensor x_47_cast_fp16 = reshape(shape = concat_3, x = transpose_24)[name = tensor("x_47_cast_fp16")]; tensor var_454_to_fp16 = const()[name = tensor("op_454_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13759424)))]; tensor var_455_to_fp16 = const()[name = tensor("op_455_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14054400)))]; tensor linear_21_cast_fp16 = linear(bias = var_455_to_fp16, weight = var_454_to_fp16, x = x_47_cast_fp16)[name = tensor("linear_21_cast_fp16")]; tensor x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = tensor("x_49_cast_fp16")]; tensor var_462_axes_0 = const()[name = tensor("op_462_axes_0"), val = tensor([-1])]; tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14055232)))]; tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14056064)))]; tensor var_462_cast_fp16 = layer_norm(axes = var_462_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_388_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = tensor("op_462_cast_fp16")]; tensor var_471_to_fp16 = const()[name = tensor("op_471_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14056896)))]; tensor var_472_to_fp16 = const()[name = tensor("op_472_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15236608)))]; tensor linear_22_cast_fp16 = linear(bias = var_472_to_fp16, weight = var_471_to_fp16, x = var_462_cast_fp16)[name = tensor("linear_22_cast_fp16")]; tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; tensor x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = tensor("x_53_cast_fp16")]; tensor var_477_to_fp16 = const()[name = tensor("op_477_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15239744)))]; tensor var_478_to_fp16 = const()[name = tensor("op_478_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16419456)))]; tensor linear_23_cast_fp16 = linear(bias = var_478_to_fp16, weight = var_477_to_fp16, x = x_53_cast_fp16)[name = tensor("linear_23_cast_fp16")]; tensor x_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = tensor("x_cast_fp16")]; tensor var_491_axes_0 = const()[name = tensor("op_491_axes_0"), val = tensor([-1])]; tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16420288)))]; tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16421120)))]; tensor var_482_to_fp16 = const()[name = tensor("op_482_to_fp16"), val = tensor(0x1.5p-17)]; tensor output = layer_norm(axes = var_491_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_482_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = tensor("op_491_cast_fp16")]; } -> (output); }