program(1.3) [buildInfo = dict({{"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(tensor logmel_data) { string var_44_pad_type_0 = const()[name = string("op_44_pad_type_0"), val = string("custom")]; tensor var_44_pad_0 = const()[name = string("op_44_pad_0"), val = tensor([1, 1])]; tensor var_44_strides_0 = const()[name = string("op_44_strides_0"), val = tensor([1])]; tensor var_44_dilations_0 = const()[name = string("op_44_dilations_0"), val = tensor([1])]; int32 var_44_groups_0 = const()[name = string("op_44_groups_0"), val = int32(1)]; tensor weight_3_to_fp16 = const()[name = string("weight_3_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; tensor bias_3_to_fp16 = const()[name = string("bias_3_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368768)))]; tensor var_44_cast_fp16 = conv(bias = bias_3_to_fp16, dilations = var_44_dilations_0, groups = var_44_groups_0, pad = var_44_pad_0, pad_type = var_44_pad_type_0, strides = var_44_strides_0, weight = weight_3_to_fp16, x = logmel_data)[name = string("op_44_cast_fp16")]; string input_1_mode_0 = const()[name = string("input_1_mode_0"), val = string("EXACT")]; tensor input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_44_cast_fp16)[name = string("input_1_cast_fp16")]; string var_62_pad_type_0 = const()[name = string("op_62_pad_type_0"), val = string("custom")]; tensor var_62_pad_0 = const()[name = string("op_62_pad_0"), val = tensor([1, 1])]; tensor var_62_strides_0 = const()[name = string("op_62_strides_0"), val = tensor([2])]; tensor var_62_dilations_0 = const()[name = string("op_62_dilations_0"), val = tensor([1])]; int32 var_62_groups_0 = const()[name = string("op_62_groups_0"), val = int32(1)]; tensor weight_7_to_fp16 = const()[name = string("weight_7_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370368)))]; tensor bias_7_to_fp16 = const()[name = string("bias_7_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3909376)))]; tensor var_62_cast_fp16 = conv(bias = bias_7_to_fp16, dilations = var_62_dilations_0, groups = var_62_groups_0, pad = var_62_pad_0, pad_type = var_62_pad_type_0, strides = var_62_strides_0, weight = weight_7_to_fp16, x = input_1_cast_fp16)[name = string("op_62_cast_fp16")]; string x_3_mode_0 = const()[name = string("x_3_mode_0"), val = string("EXACT")]; tensor x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_62_cast_fp16)[name = string("x_3_cast_fp16")]; tensor var_68 = const()[name = string("op_68"), val = tensor([0, 2, 1])]; tensor positional_embedding_to_fp16 = const()[name = string("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3910976)))]; tensor x_5_cast_fp16 = transpose(perm = var_68, x = x_3_cast_fp16)[name = string("transpose_120")]; tensor var_71_cast_fp16 = add(x = x_5_cast_fp16, y = positional_embedding_to_fp16)[name = string("op_71_cast_fp16")]; int32 var_84 = const()[name = string("op_84"), val = int32(-1)]; tensor var_100_axes_0 = const()[name = string("op_100_axes_0"), val = tensor([-1])]; tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = string("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6215040)))]; tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = string("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6216640)))]; fp16 var_90_to_fp16 = const()[name = string("op_90_to_fp16"), val = fp16(0x1.5p-17)]; tensor var_100_cast_fp16 = layer_norm(axes = var_100_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_90_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_71_cast_fp16)[name = string("op_100_cast_fp16")]; tensor var_111_to_fp16 = const()[name = string("op_111_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6218240)))]; tensor var_112_to_fp16 = const()[name = string("op_112_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7397952)))]; tensor linear_0_cast_fp16 = linear(bias = var_112_to_fp16, weight = var_111_to_fp16, x = var_100_cast_fp16)[name = string("linear_0_cast_fp16")]; tensor var_115_to_fp16 = const()[name = string("op_115_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7399552)))]; tensor linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8579264)))]; tensor linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_115_to_fp16, x = var_100_cast_fp16)[name = string("linear_1_cast_fp16")]; tensor var_119_to_fp16 = const()[name = string("op_119_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8580864)))]; tensor var_120_to_fp16 = const()[name = string("op_120_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9760576)))]; tensor linear_2_cast_fp16 = linear(bias = var_120_to_fp16, weight = var_119_to_fp16, x = var_100_cast_fp16)[name = string("linear_2_cast_fp16")]; tensor var_128 = const()[name = string("op_128"), val = tensor([1, 1500, 12, -1])]; tensor var_129_cast_fp16 = reshape(shape = var_128, x = linear_0_cast_fp16)[name = string("op_129_cast_fp16")]; tensor const_84_to_fp16 = const()[name = string("const_84_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor q_3_cast_fp16 = mul(x = var_129_cast_fp16, y = const_84_to_fp16)[name = string("q_3_cast_fp16")]; tensor var_135 = const()[name = string("op_135"), val = tensor([1, 1500, 12, -1])]; tensor var_136_cast_fp16 = reshape(shape = var_135, x = linear_1_cast_fp16)[name = string("op_136_cast_fp16")]; tensor const_85_to_fp16 = const()[name = string("const_85_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor k_3_cast_fp16 = mul(x = var_136_cast_fp16, y = const_85_to_fp16)[name = string("k_3_cast_fp16")]; tensor var_142 = const()[name = string("op_142"), val = tensor([1, 1500, 12, -1])]; tensor var_143_cast_fp16 = reshape(shape = var_142, x = linear_2_cast_fp16)[name = string("op_143_cast_fp16")]; tensor var_144 = const()[name = string("op_144"), val = tensor([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 transpose_48_perm_0 = const()[name = string("transpose_48_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_49_perm_0 = const()[name = string("transpose_49_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_49 = transpose(perm = transpose_49_perm_0, x = k_3_cast_fp16)[name = string("transpose_117")]; tensor transpose_48 = transpose(perm = transpose_48_perm_0, x = q_3_cast_fp16)[name = string("transpose_118")]; tensor qk_1_cast_fp16 = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_48, y = transpose_49)[name = string("qk_1_cast_fp16")]; tensor var_148_cast_fp16 = softmax(axis = var_84, x = qk_1_cast_fp16)[name = string("op_148_cast_fp16")]; bool var_150_transpose_x_0 = const()[name = string("op_150_transpose_x_0"), val = bool(false)]; bool var_150_transpose_y_0 = const()[name = string("op_150_transpose_y_0"), val = bool(false)]; tensor v_3_cast_fp16 = transpose(perm = var_144, x = var_143_cast_fp16)[name = string("transpose_119")]; tensor var_150_cast_fp16 = matmul(transpose_x = var_150_transpose_x_0, transpose_y = var_150_transpose_y_0, x = var_148_cast_fp16, y = v_3_cast_fp16)[name = string("op_150_cast_fp16")]; tensor var_151 = const()[name = string("op_151"), val = tensor([0, 2, 1, 3])]; tensor concat_0 = const()[name = string("concat_0"), val = tensor([1, 1500, 768])]; tensor var_152_cast_fp16 = transpose(perm = var_151, x = var_150_cast_fp16)[name = string("transpose_116")]; tensor x_11_cast_fp16 = reshape(shape = concat_0, x = var_152_cast_fp16)[name = string("x_11_cast_fp16")]; tensor var_156_to_fp16 = const()[name = string("op_156_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9762176)))]; tensor var_157_to_fp16 = const()[name = string("op_157_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10941888)))]; tensor linear_3_cast_fp16 = linear(bias = var_157_to_fp16, weight = var_156_to_fp16, x = x_11_cast_fp16)[name = string("linear_3_cast_fp16")]; tensor x_13_cast_fp16 = add(x = var_71_cast_fp16, y = linear_3_cast_fp16)[name = string("x_13_cast_fp16")]; tensor var_164_axes_0 = const()[name = string("op_164_axes_0"), val = tensor([-1])]; tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = string("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10943488)))]; tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = string("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10945088)))]; tensor var_164_cast_fp16 = layer_norm(axes = var_164_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_90_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = string("op_164_cast_fp16")]; tensor var_173_to_fp16 = const()[name = string("op_173_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10946688)))]; tensor var_174_to_fp16 = const()[name = string("op_174_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15665344)))]; tensor linear_4_cast_fp16 = linear(bias = var_174_to_fp16, weight = var_173_to_fp16, x = var_164_cast_fp16)[name = string("linear_4_cast_fp16")]; string x_17_mode_0 = const()[name = string("x_17_mode_0"), val = string("EXACT")]; tensor x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = string("x_17_cast_fp16")]; tensor var_179_to_fp16 = const()[name = string("op_179_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15671552)))]; tensor var_180_to_fp16 = const()[name = string("op_180_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20390208)))]; tensor linear_5_cast_fp16 = linear(bias = var_180_to_fp16, weight = var_179_to_fp16, x = x_17_cast_fp16)[name = string("linear_5_cast_fp16")]; tensor x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = string("x_19_cast_fp16")]; int32 var_190 = const()[name = string("op_190"), val = int32(-1)]; tensor var_206_axes_0 = const()[name = string("op_206_axes_0"), val = tensor([-1])]; tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = string("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20391808)))]; tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = string("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20393408)))]; fp16 var_196_to_fp16 = const()[name = string("op_196_to_fp16"), val = fp16(0x1.5p-17)]; tensor var_206_cast_fp16 = layer_norm(axes = var_206_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_196_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = string("op_206_cast_fp16")]; tensor var_217_to_fp16 = const()[name = string("op_217_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20395008)))]; tensor var_218_to_fp16 = const()[name = string("op_218_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21574720)))]; tensor linear_6_cast_fp16 = linear(bias = var_218_to_fp16, weight = var_217_to_fp16, x = var_206_cast_fp16)[name = string("linear_6_cast_fp16")]; tensor var_221_to_fp16 = const()[name = string("op_221_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21576320)))]; tensor linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_221_to_fp16, x = var_206_cast_fp16)[name = string("linear_7_cast_fp16")]; tensor var_225_to_fp16 = const()[name = string("op_225_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22756032)))]; tensor var_226_to_fp16 = const()[name = string("op_226_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23935744)))]; tensor linear_8_cast_fp16 = linear(bias = var_226_to_fp16, weight = var_225_to_fp16, x = var_206_cast_fp16)[name = string("linear_8_cast_fp16")]; tensor var_234 = const()[name = string("op_234"), val = tensor([1, 1500, 12, -1])]; tensor var_235_cast_fp16 = reshape(shape = var_234, x = linear_6_cast_fp16)[name = string("op_235_cast_fp16")]; tensor const_86_to_fp16 = const()[name = string("const_86_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor q_7_cast_fp16 = mul(x = var_235_cast_fp16, y = const_86_to_fp16)[name = string("q_7_cast_fp16")]; tensor var_241 = const()[name = string("op_241"), val = tensor([1, 1500, 12, -1])]; tensor var_242_cast_fp16 = reshape(shape = var_241, x = linear_7_cast_fp16)[name = string("op_242_cast_fp16")]; tensor const_87_to_fp16 = const()[name = string("const_87_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor k_7_cast_fp16 = mul(x = var_242_cast_fp16, y = const_87_to_fp16)[name = string("k_7_cast_fp16")]; tensor var_248 = const()[name = string("op_248"), val = tensor([1, 1500, 12, -1])]; tensor var_249_cast_fp16 = reshape(shape = var_248, x = linear_8_cast_fp16)[name = string("op_249_cast_fp16")]; tensor var_250 = const()[name = string("op_250"), val = tensor([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 transpose_50_perm_0 = const()[name = string("transpose_50_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_51_perm_0 = const()[name = string("transpose_51_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_51 = transpose(perm = transpose_51_perm_0, x = k_7_cast_fp16)[name = string("transpose_113")]; tensor transpose_50 = transpose(perm = transpose_50_perm_0, x = q_7_cast_fp16)[name = string("transpose_114")]; tensor qk_3_cast_fp16 = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_50, y = transpose_51)[name = string("qk_3_cast_fp16")]; tensor var_254_cast_fp16 = softmax(axis = var_190, x = qk_3_cast_fp16)[name = string("op_254_cast_fp16")]; bool var_256_transpose_x_0 = const()[name = string("op_256_transpose_x_0"), val = bool(false)]; bool var_256_transpose_y_0 = const()[name = string("op_256_transpose_y_0"), val = bool(false)]; tensor v_7_cast_fp16 = transpose(perm = var_250, x = var_249_cast_fp16)[name = string("transpose_115")]; tensor var_256_cast_fp16 = matmul(transpose_x = var_256_transpose_x_0, transpose_y = var_256_transpose_y_0, x = var_254_cast_fp16, y = v_7_cast_fp16)[name = string("op_256_cast_fp16")]; tensor var_257 = const()[name = string("op_257"), val = tensor([0, 2, 1, 3])]; tensor concat_1 = const()[name = string("concat_1"), val = tensor([1, 1500, 768])]; tensor var_258_cast_fp16 = transpose(perm = var_257, x = var_256_cast_fp16)[name = string("transpose_112")]; tensor x_23_cast_fp16 = reshape(shape = concat_1, x = var_258_cast_fp16)[name = string("x_23_cast_fp16")]; tensor var_262_to_fp16 = const()[name = string("op_262_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23937344)))]; tensor var_263_to_fp16 = const()[name = string("op_263_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25117056)))]; tensor linear_9_cast_fp16 = linear(bias = var_263_to_fp16, weight = var_262_to_fp16, x = x_23_cast_fp16)[name = string("linear_9_cast_fp16")]; tensor x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = string("x_25_cast_fp16")]; tensor var_270_axes_0 = const()[name = string("op_270_axes_0"), val = tensor([-1])]; tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = string("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25118656)))]; tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = string("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25120256)))]; tensor var_270_cast_fp16 = layer_norm(axes = var_270_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_196_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = string("op_270_cast_fp16")]; tensor var_279_to_fp16 = const()[name = string("op_279_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25121856)))]; tensor var_280_to_fp16 = const()[name = string("op_280_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29840512)))]; tensor linear_10_cast_fp16 = linear(bias = var_280_to_fp16, weight = var_279_to_fp16, x = var_270_cast_fp16)[name = string("linear_10_cast_fp16")]; string x_29_mode_0 = const()[name = string("x_29_mode_0"), val = string("EXACT")]; tensor x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = string("x_29_cast_fp16")]; tensor var_285_to_fp16 = const()[name = string("op_285_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29846720)))]; tensor var_286_to_fp16 = const()[name = string("op_286_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34565376)))]; tensor linear_11_cast_fp16 = linear(bias = var_286_to_fp16, weight = var_285_to_fp16, x = x_29_cast_fp16)[name = string("linear_11_cast_fp16")]; tensor x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = string("x_31_cast_fp16")]; int32 var_296 = const()[name = string("op_296"), val = int32(-1)]; tensor var_312_axes_0 = const()[name = string("op_312_axes_0"), val = tensor([-1])]; tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = string("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34566976)))]; tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = string("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34568576)))]; fp16 var_302_to_fp16 = const()[name = string("op_302_to_fp16"), val = fp16(0x1.5p-17)]; tensor var_312_cast_fp16 = layer_norm(axes = var_312_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_302_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = string("op_312_cast_fp16")]; tensor var_323_to_fp16 = const()[name = string("op_323_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34570176)))]; tensor var_324_to_fp16 = const()[name = string("op_324_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35749888)))]; tensor linear_12_cast_fp16 = linear(bias = var_324_to_fp16, weight = var_323_to_fp16, x = var_312_cast_fp16)[name = string("linear_12_cast_fp16")]; tensor var_327_to_fp16 = const()[name = string("op_327_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35751488)))]; tensor linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_327_to_fp16, x = var_312_cast_fp16)[name = string("linear_13_cast_fp16")]; tensor var_331_to_fp16 = const()[name = string("op_331_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36931200)))]; tensor var_332_to_fp16 = const()[name = string("op_332_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38110912)))]; tensor linear_14_cast_fp16 = linear(bias = var_332_to_fp16, weight = var_331_to_fp16, x = var_312_cast_fp16)[name = string("linear_14_cast_fp16")]; tensor var_340 = const()[name = string("op_340"), val = tensor([1, 1500, 12, -1])]; tensor var_341_cast_fp16 = reshape(shape = var_340, x = linear_12_cast_fp16)[name = string("op_341_cast_fp16")]; tensor const_88_to_fp16 = const()[name = string("const_88_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor q_11_cast_fp16 = mul(x = var_341_cast_fp16, y = const_88_to_fp16)[name = string("q_11_cast_fp16")]; tensor var_347 = const()[name = string("op_347"), val = tensor([1, 1500, 12, -1])]; tensor var_348_cast_fp16 = reshape(shape = var_347, x = linear_13_cast_fp16)[name = string("op_348_cast_fp16")]; tensor const_89_to_fp16 = const()[name = string("const_89_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor k_11_cast_fp16 = mul(x = var_348_cast_fp16, y = const_89_to_fp16)[name = string("k_11_cast_fp16")]; tensor var_354 = const()[name = string("op_354"), val = tensor([1, 1500, 12, -1])]; tensor var_355_cast_fp16 = reshape(shape = var_354, x = linear_14_cast_fp16)[name = string("op_355_cast_fp16")]; tensor var_356 = const()[name = string("op_356"), val = tensor([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 transpose_52_perm_0 = const()[name = string("transpose_52_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_53_perm_0 = const()[name = string("transpose_53_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_53 = transpose(perm = transpose_53_perm_0, x = k_11_cast_fp16)[name = string("transpose_109")]; tensor transpose_52 = transpose(perm = transpose_52_perm_0, x = q_11_cast_fp16)[name = string("transpose_110")]; tensor qk_5_cast_fp16 = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_52, y = transpose_53)[name = string("qk_5_cast_fp16")]; tensor var_360_cast_fp16 = softmax(axis = var_296, x = qk_5_cast_fp16)[name = string("op_360_cast_fp16")]; bool var_362_transpose_x_0 = const()[name = string("op_362_transpose_x_0"), val = bool(false)]; bool var_362_transpose_y_0 = const()[name = string("op_362_transpose_y_0"), val = bool(false)]; tensor v_11_cast_fp16 = transpose(perm = var_356, x = var_355_cast_fp16)[name = string("transpose_111")]; tensor var_362_cast_fp16 = matmul(transpose_x = var_362_transpose_x_0, transpose_y = var_362_transpose_y_0, x = var_360_cast_fp16, y = v_11_cast_fp16)[name = string("op_362_cast_fp16")]; tensor var_363 = const()[name = string("op_363"), val = tensor([0, 2, 1, 3])]; tensor concat_2 = const()[name = string("concat_2"), val = tensor([1, 1500, 768])]; tensor var_364_cast_fp16 = transpose(perm = var_363, x = var_362_cast_fp16)[name = string("transpose_108")]; tensor x_35_cast_fp16 = reshape(shape = concat_2, x = var_364_cast_fp16)[name = string("x_35_cast_fp16")]; tensor var_368_to_fp16 = const()[name = string("op_368_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38112512)))]; tensor var_369_to_fp16 = const()[name = string("op_369_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39292224)))]; tensor linear_15_cast_fp16 = linear(bias = var_369_to_fp16, weight = var_368_to_fp16, x = x_35_cast_fp16)[name = string("linear_15_cast_fp16")]; tensor x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = string("x_37_cast_fp16")]; tensor var_376_axes_0 = const()[name = string("op_376_axes_0"), val = tensor([-1])]; tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = string("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39293824)))]; tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = string("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39295424)))]; tensor var_376_cast_fp16 = layer_norm(axes = var_376_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_302_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = string("op_376_cast_fp16")]; tensor var_385_to_fp16 = const()[name = string("op_385_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39297024)))]; tensor var_386_to_fp16 = const()[name = string("op_386_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44015680)))]; tensor linear_16_cast_fp16 = linear(bias = var_386_to_fp16, weight = var_385_to_fp16, x = var_376_cast_fp16)[name = string("linear_16_cast_fp16")]; string x_41_mode_0 = const()[name = string("x_41_mode_0"), val = string("EXACT")]; tensor x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = string("x_41_cast_fp16")]; tensor var_391_to_fp16 = const()[name = string("op_391_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44021888)))]; tensor var_392_to_fp16 = const()[name = string("op_392_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48740544)))]; tensor linear_17_cast_fp16 = linear(bias = var_392_to_fp16, weight = var_391_to_fp16, x = x_41_cast_fp16)[name = string("linear_17_cast_fp16")]; tensor x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = string("x_43_cast_fp16")]; int32 var_402 = const()[name = string("op_402"), val = int32(-1)]; tensor var_418_axes_0 = const()[name = string("op_418_axes_0"), val = tensor([-1])]; tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = string("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48742144)))]; tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = string("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48743744)))]; fp16 var_408_to_fp16 = const()[name = string("op_408_to_fp16"), val = fp16(0x1.5p-17)]; tensor var_418_cast_fp16 = layer_norm(axes = var_418_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_408_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = string("op_418_cast_fp16")]; tensor var_429_to_fp16 = const()[name = string("op_429_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48745344)))]; tensor var_430_to_fp16 = const()[name = string("op_430_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49925056)))]; tensor linear_18_cast_fp16 = linear(bias = var_430_to_fp16, weight = var_429_to_fp16, x = var_418_cast_fp16)[name = string("linear_18_cast_fp16")]; tensor var_433_to_fp16 = const()[name = string("op_433_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49926656)))]; tensor linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_433_to_fp16, x = var_418_cast_fp16)[name = string("linear_19_cast_fp16")]; tensor var_437_to_fp16 = const()[name = string("op_437_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51106368)))]; tensor var_438_to_fp16 = const()[name = string("op_438_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52286080)))]; tensor linear_20_cast_fp16 = linear(bias = var_438_to_fp16, weight = var_437_to_fp16, x = var_418_cast_fp16)[name = string("linear_20_cast_fp16")]; tensor var_446 = const()[name = string("op_446"), val = tensor([1, 1500, 12, -1])]; tensor var_447_cast_fp16 = reshape(shape = var_446, x = linear_18_cast_fp16)[name = string("op_447_cast_fp16")]; tensor const_90_to_fp16 = const()[name = string("const_90_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor q_15_cast_fp16 = mul(x = var_447_cast_fp16, y = const_90_to_fp16)[name = string("q_15_cast_fp16")]; tensor var_453 = const()[name = string("op_453"), val = tensor([1, 1500, 12, -1])]; tensor var_454_cast_fp16 = reshape(shape = var_453, x = linear_19_cast_fp16)[name = string("op_454_cast_fp16")]; tensor const_91_to_fp16 = const()[name = string("const_91_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor k_15_cast_fp16 = mul(x = var_454_cast_fp16, y = const_91_to_fp16)[name = string("k_15_cast_fp16")]; tensor var_460 = const()[name = string("op_460"), val = tensor([1, 1500, 12, -1])]; tensor var_461_cast_fp16 = reshape(shape = var_460, x = linear_20_cast_fp16)[name = string("op_461_cast_fp16")]; tensor var_462 = const()[name = string("op_462"), val = tensor([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 transpose_54_perm_0 = const()[name = string("transpose_54_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_55_perm_0 = const()[name = string("transpose_55_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_55 = transpose(perm = transpose_55_perm_0, x = k_15_cast_fp16)[name = string("transpose_105")]; tensor transpose_54 = transpose(perm = transpose_54_perm_0, x = q_15_cast_fp16)[name = string("transpose_106")]; tensor qk_7_cast_fp16 = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_54, y = transpose_55)[name = string("qk_7_cast_fp16")]; tensor var_466_cast_fp16 = softmax(axis = var_402, x = qk_7_cast_fp16)[name = string("op_466_cast_fp16")]; bool var_468_transpose_x_0 = const()[name = string("op_468_transpose_x_0"), val = bool(false)]; bool var_468_transpose_y_0 = const()[name = string("op_468_transpose_y_0"), val = bool(false)]; tensor v_15_cast_fp16 = transpose(perm = var_462, x = var_461_cast_fp16)[name = string("transpose_107")]; tensor var_468_cast_fp16 = matmul(transpose_x = var_468_transpose_x_0, transpose_y = var_468_transpose_y_0, x = var_466_cast_fp16, y = v_15_cast_fp16)[name = string("op_468_cast_fp16")]; tensor var_469 = const()[name = string("op_469"), val = tensor([0, 2, 1, 3])]; tensor concat_3 = const()[name = string("concat_3"), val = tensor([1, 1500, 768])]; tensor var_470_cast_fp16 = transpose(perm = var_469, x = var_468_cast_fp16)[name = string("transpose_104")]; tensor x_47_cast_fp16 = reshape(shape = concat_3, x = var_470_cast_fp16)[name = string("x_47_cast_fp16")]; tensor var_474_to_fp16 = const()[name = string("op_474_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52287680)))]; tensor var_475_to_fp16 = const()[name = string("op_475_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53467392)))]; tensor linear_21_cast_fp16 = linear(bias = var_475_to_fp16, weight = var_474_to_fp16, x = x_47_cast_fp16)[name = string("linear_21_cast_fp16")]; tensor x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = string("x_49_cast_fp16")]; tensor var_482_axes_0 = const()[name = string("op_482_axes_0"), val = tensor([-1])]; tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = string("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53468992)))]; tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = string("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53470592)))]; tensor var_482_cast_fp16 = layer_norm(axes = var_482_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_408_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = string("op_482_cast_fp16")]; tensor var_491_to_fp16 = const()[name = string("op_491_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53472192)))]; tensor var_492_to_fp16 = const()[name = string("op_492_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58190848)))]; tensor linear_22_cast_fp16 = linear(bias = var_492_to_fp16, weight = var_491_to_fp16, x = var_482_cast_fp16)[name = string("linear_22_cast_fp16")]; string x_53_mode_0 = const()[name = string("x_53_mode_0"), val = string("EXACT")]; tensor x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = string("x_53_cast_fp16")]; tensor var_497_to_fp16 = const()[name = string("op_497_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58197056)))]; tensor var_498_to_fp16 = const()[name = string("op_498_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62915712)))]; tensor linear_23_cast_fp16 = linear(bias = var_498_to_fp16, weight = var_497_to_fp16, x = x_53_cast_fp16)[name = string("linear_23_cast_fp16")]; tensor x_55_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = string("x_55_cast_fp16")]; int32 var_508 = const()[name = string("op_508"), val = int32(-1)]; tensor var_524_axes_0 = const()[name = string("op_524_axes_0"), val = tensor([-1])]; tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = string("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62917312)))]; tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = string("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62918912)))]; fp16 var_514_to_fp16 = const()[name = string("op_514_to_fp16"), val = fp16(0x1.5p-17)]; tensor var_524_cast_fp16 = layer_norm(axes = var_524_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_514_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast_fp16)[name = string("op_524_cast_fp16")]; tensor var_535_to_fp16 = const()[name = string("op_535_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62920512)))]; tensor var_536_to_fp16 = const()[name = string("op_536_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64100224)))]; tensor linear_24_cast_fp16 = linear(bias = var_536_to_fp16, weight = var_535_to_fp16, x = var_524_cast_fp16)[name = string("linear_24_cast_fp16")]; tensor var_539_to_fp16 = const()[name = string("op_539_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64101824)))]; tensor linear_25_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_539_to_fp16, x = var_524_cast_fp16)[name = string("linear_25_cast_fp16")]; tensor var_543_to_fp16 = const()[name = string("op_543_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65281536)))]; tensor var_544_to_fp16 = const()[name = string("op_544_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66461248)))]; tensor linear_26_cast_fp16 = linear(bias = var_544_to_fp16, weight = var_543_to_fp16, x = var_524_cast_fp16)[name = string("linear_26_cast_fp16")]; tensor var_552 = const()[name = string("op_552"), val = tensor([1, 1500, 12, -1])]; tensor var_553_cast_fp16 = reshape(shape = var_552, x = linear_24_cast_fp16)[name = string("op_553_cast_fp16")]; tensor const_92_to_fp16 = const()[name = string("const_92_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor q_19_cast_fp16 = mul(x = var_553_cast_fp16, y = const_92_to_fp16)[name = string("q_19_cast_fp16")]; tensor var_559 = const()[name = string("op_559"), val = tensor([1, 1500, 12, -1])]; tensor var_560_cast_fp16 = reshape(shape = var_559, x = linear_25_cast_fp16)[name = string("op_560_cast_fp16")]; tensor const_93_to_fp16 = const()[name = string("const_93_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor k_19_cast_fp16 = mul(x = var_560_cast_fp16, y = const_93_to_fp16)[name = string("k_19_cast_fp16")]; tensor var_566 = const()[name = string("op_566"), val = tensor([1, 1500, 12, -1])]; tensor var_567_cast_fp16 = reshape(shape = var_566, x = linear_26_cast_fp16)[name = string("op_567_cast_fp16")]; tensor var_568 = const()[name = string("op_568"), val = tensor([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 transpose_56_perm_0 = const()[name = string("transpose_56_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_57_perm_0 = const()[name = string("transpose_57_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_57 = transpose(perm = transpose_57_perm_0, x = k_19_cast_fp16)[name = string("transpose_101")]; tensor transpose_56 = transpose(perm = transpose_56_perm_0, x = q_19_cast_fp16)[name = string("transpose_102")]; tensor qk_9_cast_fp16 = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_56, y = transpose_57)[name = string("qk_9_cast_fp16")]; tensor var_572_cast_fp16 = softmax(axis = var_508, x = qk_9_cast_fp16)[name = string("op_572_cast_fp16")]; bool var_574_transpose_x_0 = const()[name = string("op_574_transpose_x_0"), val = bool(false)]; bool var_574_transpose_y_0 = const()[name = string("op_574_transpose_y_0"), val = bool(false)]; tensor v_19_cast_fp16 = transpose(perm = var_568, x = var_567_cast_fp16)[name = string("transpose_103")]; tensor var_574_cast_fp16 = matmul(transpose_x = var_574_transpose_x_0, transpose_y = var_574_transpose_y_0, x = var_572_cast_fp16, y = v_19_cast_fp16)[name = string("op_574_cast_fp16")]; tensor var_575 = const()[name = string("op_575"), val = tensor([0, 2, 1, 3])]; tensor concat_4 = const()[name = string("concat_4"), val = tensor([1, 1500, 768])]; tensor var_576_cast_fp16 = transpose(perm = var_575, x = var_574_cast_fp16)[name = string("transpose_100")]; tensor x_59_cast_fp16 = reshape(shape = concat_4, x = var_576_cast_fp16)[name = string("x_59_cast_fp16")]; tensor