| program(1.3) | |
| [buildInfo = dict<string, string>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.9.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})] | |
| { | |
| func main<ios18>(tensor<fp16, [1, 80, 3000]> logmel_data) { | |
| string var_28_pad_type_0 = const()[name = string("op_28_pad_type_0"), val = string("custom")]; | |
| tensor<int32, [2]> var_28_pad_0 = const()[name = string("op_28_pad_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [1]> var_28_strides_0 = const()[name = string("op_28_strides_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<int32, [1]> var_28_dilations_0 = const()[name = string("op_28_dilations_0"), val = tensor<int32, [1]>([1])]; | |
| int32 var_28_groups_0 = const()[name = string("op_28_groups_0"), val = int32(1)]; | |
| tensor<fp16, [384, 80, 3]> const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = tensor<fp16, [384, 80, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; | |
| tensor<fp16, [384]> const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184448)))]; | |
| tensor<fp16, [1, 384, 3000]> var_28_cast_fp16 = conv(bias = const_1_to_fp16, dilations = var_28_dilations_0, groups = var_28_groups_0, pad = var_28_pad_0, pad_type = var_28_pad_type_0, strides = var_28_strides_0, weight = const_0_to_fp16, x = logmel_data)[name = string("op_28_cast_fp16")]; | |
| string input_1_mode_0 = const()[name = string("input_1_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 384, 3000]> input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_28_cast_fp16)[name = string("input_1_cast_fp16")]; | |
| string var_46_pad_type_0 = const()[name = string("op_46_pad_type_0"), val = string("custom")]; | |
| tensor<int32, [2]> var_46_pad_0 = const()[name = string("op_46_pad_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [1]> var_46_strides_0 = const()[name = string("op_46_strides_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<int32, [1]> var_46_dilations_0 = const()[name = string("op_46_dilations_0"), val = tensor<int32, [1]>([1])]; | |
| int32 var_46_groups_0 = const()[name = string("op_46_groups_0"), val = int32(1)]; | |
| tensor<fp16, [384, 384, 3]> const_2_to_fp16 = const()[name = string("const_2_to_fp16"), val = tensor<fp16, [384, 384, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185280)))]; | |
| tensor<fp16, [384]> const_3_to_fp16 = const()[name = string("const_3_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1070080)))]; | |
| tensor<fp16, [1, 384, 1500]> var_46_cast_fp16 = conv(bias = const_3_to_fp16, dilations = var_46_dilations_0, groups = var_46_groups_0, pad = var_46_pad_0, pad_type = var_46_pad_type_0, strides = var_46_strides_0, weight = const_2_to_fp16, x = input_1_cast_fp16)[name = string("op_46_cast_fp16")]; | |
| string x_3_mode_0 = const()[name = string("x_3_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 384, 1500]> x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_46_cast_fp16)[name = string("x_3_cast_fp16")]; | |
| tensor<int32, [3]> var_52 = const()[name = string("op_52"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<fp16, [1500, 384]> positional_embedding_to_fp16 = const()[name = string("positional_embedding_to_fp16"), val = tensor<fp16, [1500, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1070912)))]; | |
| tensor<fp16, [1, 1500, 384]> x_5_cast_fp16 = transpose(perm = var_52, x = x_3_cast_fp16)[name = string("transpose_52")]; | |
| tensor<fp16, [1, 1500, 384]> var_55_cast_fp16 = add(x = x_5_cast_fp16, y = positional_embedding_to_fp16)[name = string("op_55_cast_fp16")]; | |
| tensor<int32, [1]> var_82_axes_0 = const()[name = string("op_82_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_0_attn_ln_weight_to_fp16 = const()[name = string("blocks_0_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2222976)))]; | |
| tensor<fp16, [384]> blocks_0_attn_ln_bias_to_fp16 = const()[name = string("blocks_0_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2223808)))]; | |
| fp16 var_72_to_fp16 = const()[name = string("op_72_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 384]> var_82_cast_fp16 = layer_norm(axes = var_82_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_72_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_55_cast_fp16)[name = string("op_82_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_4_to_fp16 = const()[name = string("const_4_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2224640)))]; | |
| tensor<fp16, [384]> const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2519616)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_0_cast_fp16 = linear(bias = const_5_to_fp16, weight = const_4_to_fp16, x = var_82_cast_fp16)[name = string("linear_0_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_6_to_fp16 = const()[name = string("const_6_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2520448)))]; | |
| tensor<fp16, [384]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2815424)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_6_to_fp16, x = var_82_cast_fp16)[name = string("linear_1_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2816256)))]; | |
| tensor<fp16, [384]> const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3111232)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_2_cast_fp16 = linear(bias = const_8_to_fp16, weight = const_7_to_fp16, x = var_82_cast_fp16)[name = string("linear_2_cast_fp16")]; | |
| tensor<int32, [4]> var_106 = const()[name = string("op_106"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_107_cast_fp16 = reshape(shape = var_106, x = linear_0_cast_fp16)[name = string("op_107_cast_fp16")]; | |
| tensor<int32, [4]> var_112 = const()[name = string("op_112"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_113_cast_fp16 = reshape(shape = var_112, x = linear_1_cast_fp16)[name = string("op_113_cast_fp16")]; | |
| tensor<int32, [4]> var_118 = const()[name = string("op_118"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_119_cast_fp16 = reshape(shape = var_118, x = linear_2_cast_fp16)[name = string("op_119_cast_fp16")]; | |
| tensor<int32, [4]> transpose_24_perm_0 = const()[name = string("transpose_24_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_25_perm_0 = const()[name = string("transpose_25_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_26_perm_0 = const()[name = string("transpose_26_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_26 = transpose(perm = transpose_26_perm_0, x = var_119_cast_fp16)[name = string("transpose_49")]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_25 = transpose(perm = transpose_25_perm_0, x = var_113_cast_fp16)[name = string("transpose_50")]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_24 = transpose(perm = transpose_24_perm_0, x = var_107_cast_fp16)[name = string("transpose_51")]; | |
| tensor<fp16, [1, 6, 1500, 64]> a_1_cast_fp16 = scaled_dot_product_attention(key = transpose_25, query = transpose_24, value = transpose_26)[name = string("a_1_cast_fp16")]; | |
| tensor<int32, [4]> var_123 = const()[name = string("op_123"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_0 = const()[name = string("concat_0"), val = tensor<int32, [3]>([1, 1500, 384])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_124_cast_fp16 = transpose(perm = var_123, x = a_1_cast_fp16)[name = string("transpose_48")]; | |
| tensor<fp16, [1, 1500, 384]> x_11_cast_fp16 = reshape(shape = concat_0, x = var_124_cast_fp16)[name = string("x_11_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_15_to_fp16 = const()[name = string("const_15_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3112064)))]; | |
| tensor<fp16, [384]> const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3407040)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_3_cast_fp16 = linear(bias = const_16_to_fp16, weight = const_15_to_fp16, x = x_11_cast_fp16)[name = string("linear_3_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> x_13_cast_fp16 = add(x = var_55_cast_fp16, y = linear_3_cast_fp16)[name = string("x_13_cast_fp16")]; | |
| tensor<int32, [1]> var_136_axes_0 = const()[name = string("op_136_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_0_mlp_ln_weight_to_fp16 = const()[name = string("blocks_0_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3407872)))]; | |
| tensor<fp16, [384]> blocks_0_mlp_ln_bias_to_fp16 = const()[name = string("blocks_0_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3408704)))]; | |
| tensor<fp16, [1, 1500, 384]> var_136_cast_fp16 = layer_norm(axes = var_136_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_72_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = string("op_136_cast_fp16")]; | |
| tensor<fp16, [1536, 384]> const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3409536)))]; | |
| tensor<fp16, [1536]> const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4589248)))]; | |
| tensor<fp16, [1, 1500, 1536]> linear_4_cast_fp16 = linear(bias = const_18_to_fp16, weight = const_17_to_fp16, x = var_136_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, 1536]> x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = string("x_17_cast_fp16")]; | |
| tensor<fp16, [384, 1536]> const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4592384)))]; | |
| tensor<fp16, [384]> const_20_to_fp16 = const()[name = string("const_20_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5772096)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_5_cast_fp16 = linear(bias = const_20_to_fp16, weight = const_19_to_fp16, x = x_17_cast_fp16)[name = string("linear_5_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = string("x_19_cast_fp16")]; | |
| tensor<int32, [1]> var_176_axes_0 = const()[name = string("op_176_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_1_attn_ln_weight_to_fp16 = const()[name = string("blocks_1_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5772928)))]; | |
| tensor<fp16, [384]> blocks_1_attn_ln_bias_to_fp16 = const()[name = string("blocks_1_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5773760)))]; | |
| fp16 var_166_to_fp16 = const()[name = string("op_166_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 384]> var_176_cast_fp16 = layer_norm(axes = var_176_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_166_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = string("op_176_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_21_to_fp16 = const()[name = string("const_21_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5774592)))]; | |
| tensor<fp16, [384]> const_22_to_fp16 = const()[name = string("const_22_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6069568)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_6_cast_fp16 = linear(bias = const_22_to_fp16, weight = const_21_to_fp16, x = var_176_cast_fp16)[name = string("linear_6_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_23_to_fp16 = const()[name = string("const_23_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6070400)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_23_to_fp16, x = var_176_cast_fp16)[name = string("linear_7_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6365376)))]; | |
| tensor<fp16, [384]> const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6660352)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_8_cast_fp16 = linear(bias = const_25_to_fp16, weight = const_24_to_fp16, x = var_176_cast_fp16)[name = string("linear_8_cast_fp16")]; | |
| tensor<int32, [4]> var_200 = const()[name = string("op_200"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_201_cast_fp16 = reshape(shape = var_200, x = linear_6_cast_fp16)[name = string("op_201_cast_fp16")]; | |
| tensor<int32, [4]> var_206 = const()[name = string("op_206"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_207_cast_fp16 = reshape(shape = var_206, x = linear_7_cast_fp16)[name = string("op_207_cast_fp16")]; | |
| tensor<int32, [4]> var_212 = const()[name = string("op_212"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_213_cast_fp16 = reshape(shape = var_212, x = linear_8_cast_fp16)[name = string("op_213_cast_fp16")]; | |
| tensor<int32, [4]> transpose_27_perm_0 = const()[name = string("transpose_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_28_perm_0 = const()[name = string("transpose_28_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_29_perm_0 = const()[name = string("transpose_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_29 = transpose(perm = transpose_29_perm_0, x = var_213_cast_fp16)[name = string("transpose_45")]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_28 = transpose(perm = transpose_28_perm_0, x = var_207_cast_fp16)[name = string("transpose_46")]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_27 = transpose(perm = transpose_27_perm_0, x = var_201_cast_fp16)[name = string("transpose_47")]; | |
| tensor<fp16, [1, 6, 1500, 64]> a_3_cast_fp16 = scaled_dot_product_attention(key = transpose_28, query = transpose_27, value = transpose_29)[name = string("a_3_cast_fp16")]; | |
| tensor<int32, [4]> var_217 = const()[name = string("op_217"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_1 = const()[name = string("concat_1"), val = tensor<int32, [3]>([1, 1500, 384])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_218_cast_fp16 = transpose(perm = var_217, x = a_3_cast_fp16)[name = string("transpose_44")]; | |
| tensor<fp16, [1, 1500, 384]> x_23_cast_fp16 = reshape(shape = concat_1, x = var_218_cast_fp16)[name = string("x_23_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_32_to_fp16 = const()[name = string("const_32_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6661184)))]; | |
| tensor<fp16, [384]> const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6956160)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_9_cast_fp16 = linear(bias = const_33_to_fp16, weight = const_32_to_fp16, x = x_23_cast_fp16)[name = string("linear_9_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> 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_230_axes_0 = const()[name = string("op_230_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_1_mlp_ln_weight_to_fp16 = const()[name = string("blocks_1_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6956992)))]; | |
| tensor<fp16, [384]> blocks_1_mlp_ln_bias_to_fp16 = const()[name = string("blocks_1_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6957824)))]; | |
| tensor<fp16, [1, 1500, 384]> var_230_cast_fp16 = layer_norm(axes = var_230_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_166_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = string("op_230_cast_fp16")]; | |
| tensor<fp16, [1536, 384]> const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6958656)))]; | |
| tensor<fp16, [1536]> const_35_to_fp16 = const()[name = string("const_35_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8138368)))]; | |
| tensor<fp16, [1, 1500, 1536]> linear_10_cast_fp16 = linear(bias = const_35_to_fp16, weight = const_34_to_fp16, x = var_230_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, 1536]> x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = string("x_29_cast_fp16")]; | |
| tensor<fp16, [384, 1536]> const_36_to_fp16 = const()[name = string("const_36_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8141504)))]; | |
| tensor<fp16, [384]> const_37_to_fp16 = const()[name = string("const_37_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9321216)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_11_cast_fp16 = linear(bias = const_37_to_fp16, weight = const_36_to_fp16, x = x_29_cast_fp16)[name = string("linear_11_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = string("x_31_cast_fp16")]; | |
| tensor<int32, [1]> var_270_axes_0 = const()[name = string("op_270_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_2_attn_ln_weight_to_fp16 = const()[name = string("blocks_2_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9322048)))]; | |
| tensor<fp16, [384]> blocks_2_attn_ln_bias_to_fp16 = const()[name = string("blocks_2_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9322880)))]; | |
| fp16 var_260_to_fp16 = const()[name = string("op_260_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 384]> var_270_cast_fp16 = layer_norm(axes = var_270_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_260_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = string("op_270_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_38_to_fp16 = const()[name = string("const_38_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9323712)))]; | |
| tensor<fp16, [384]> const_39_to_fp16 = const()[name = string("const_39_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9618688)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_12_cast_fp16 = linear(bias = const_39_to_fp16, weight = const_38_to_fp16, x = var_270_cast_fp16)[name = string("linear_12_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_40_to_fp16 = const()[name = string("const_40_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9619520)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_40_to_fp16, x = var_270_cast_fp16)[name = string("linear_13_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_41_to_fp16 = const()[name = string("const_41_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9914496)))]; | |
| tensor<fp16, [384]> const_42_to_fp16 = const()[name = string("const_42_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10209472)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_14_cast_fp16 = linear(bias = const_42_to_fp16, weight = const_41_to_fp16, x = var_270_cast_fp16)[name = string("linear_14_cast_fp16")]; | |
| tensor<int32, [4]> var_294 = const()[name = string("op_294"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_295_cast_fp16 = reshape(shape = var_294, x = linear_12_cast_fp16)[name = string("op_295_cast_fp16")]; | |
| tensor<int32, [4]> var_300 = const()[name = string("op_300"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_301_cast_fp16 = reshape(shape = var_300, x = linear_13_cast_fp16)[name = string("op_301_cast_fp16")]; | |
| tensor<int32, [4]> var_306 = const()[name = string("op_306"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_307_cast_fp16 = reshape(shape = var_306, x = linear_14_cast_fp16)[name = string("op_307_cast_fp16")]; | |
| tensor<int32, [4]> transpose_30_perm_0 = const()[name = string("transpose_30_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_31_perm_0 = const()[name = string("transpose_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_32_perm_0 = const()[name = string("transpose_32_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_32 = transpose(perm = transpose_32_perm_0, x = var_307_cast_fp16)[name = string("transpose_41")]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_31 = transpose(perm = transpose_31_perm_0, x = var_301_cast_fp16)[name = string("transpose_42")]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_30 = transpose(perm = transpose_30_perm_0, x = var_295_cast_fp16)[name = string("transpose_43")]; | |
| tensor<fp16, [1, 6, 1500, 64]> a_5_cast_fp16 = scaled_dot_product_attention(key = transpose_31, query = transpose_30, value = transpose_32)[name = string("a_5_cast_fp16")]; | |
| tensor<int32, [4]> var_311 = const()[name = string("op_311"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_2 = const()[name = string("concat_2"), val = tensor<int32, [3]>([1, 1500, 384])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_312_cast_fp16 = transpose(perm = var_311, x = a_5_cast_fp16)[name = string("transpose_40")]; | |
| tensor<fp16, [1, 1500, 384]> x_35_cast_fp16 = reshape(shape = concat_2, x = var_312_cast_fp16)[name = string("x_35_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_49_to_fp16 = const()[name = string("const_49_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10210304)))]; | |
| tensor<fp16, [384]> const_50_to_fp16 = const()[name = string("const_50_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10505280)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_15_cast_fp16 = linear(bias = const_50_to_fp16, weight = const_49_to_fp16, x = x_35_cast_fp16)[name = string("linear_15_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> 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_324_axes_0 = const()[name = string("op_324_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_2_mlp_ln_weight_to_fp16 = const()[name = string("blocks_2_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10506112)))]; | |
| tensor<fp16, [384]> blocks_2_mlp_ln_bias_to_fp16 = const()[name = string("blocks_2_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10506944)))]; | |
| tensor<fp16, [1, 1500, 384]> var_324_cast_fp16 = layer_norm(axes = var_324_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_260_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = string("op_324_cast_fp16")]; | |
| tensor<fp16, [1536, 384]> const_51_to_fp16 = const()[name = string("const_51_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10507776)))]; | |
| tensor<fp16, [1536]> const_52_to_fp16 = const()[name = string("const_52_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11687488)))]; | |
| tensor<fp16, [1, 1500, 1536]> linear_16_cast_fp16 = linear(bias = const_52_to_fp16, weight = const_51_to_fp16, x = var_324_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, 1536]> x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = string("x_41_cast_fp16")]; | |
| tensor<fp16, [384, 1536]> const_53_to_fp16 = const()[name = string("const_53_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11690624)))]; | |
| tensor<fp16, [384]> const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12870336)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_17_cast_fp16 = linear(bias = const_54_to_fp16, weight = const_53_to_fp16, x = x_41_cast_fp16)[name = string("linear_17_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = string("x_43_cast_fp16")]; | |
| tensor<int32, [1]> var_364_axes_0 = const()[name = string("op_364_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_3_attn_ln_weight_to_fp16 = const()[name = string("blocks_3_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12871168)))]; | |
| tensor<fp16, [384]> blocks_3_attn_ln_bias_to_fp16 = const()[name = string("blocks_3_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12872000)))]; | |
| fp16 var_354_to_fp16 = const()[name = string("op_354_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 384]> var_364_cast_fp16 = layer_norm(axes = var_364_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_354_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = string("op_364_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_55_to_fp16 = const()[name = string("const_55_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12872832)))]; | |
| tensor<fp16, [384]> const_56_to_fp16 = const()[name = string("const_56_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13167808)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_18_cast_fp16 = linear(bias = const_56_to_fp16, weight = const_55_to_fp16, x = var_364_cast_fp16)[name = string("linear_18_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13168640)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_57_to_fp16, x = var_364_cast_fp16)[name = string("linear_19_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_58_to_fp16 = const()[name = string("const_58_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13463616)))]; | |
| tensor<fp16, [384]> const_59_to_fp16 = const()[name = string("const_59_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13758592)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_20_cast_fp16 = linear(bias = const_59_to_fp16, weight = const_58_to_fp16, x = var_364_cast_fp16)[name = string("linear_20_cast_fp16")]; | |
| tensor<int32, [4]> var_388 = const()[name = string("op_388"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_389_cast_fp16 = reshape(shape = var_388, x = linear_18_cast_fp16)[name = string("op_389_cast_fp16")]; | |
| tensor<int32, [4]> var_394 = const()[name = string("op_394"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_395_cast_fp16 = reshape(shape = var_394, x = linear_19_cast_fp16)[name = string("op_395_cast_fp16")]; | |
| tensor<int32, [4]> var_400 = const()[name = string("op_400"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_401_cast_fp16 = reshape(shape = var_400, x = linear_20_cast_fp16)[name = string("op_401_cast_fp16")]; | |
| tensor<int32, [4]> transpose_33_perm_0 = const()[name = string("transpose_33_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_34_perm_0 = const()[name = string("transpose_34_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_35_perm_0 = const()[name = string("transpose_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_35 = transpose(perm = transpose_35_perm_0, x = var_401_cast_fp16)[name = string("transpose_37")]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_34 = transpose(perm = transpose_34_perm_0, x = var_395_cast_fp16)[name = string("transpose_38")]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_33 = transpose(perm = transpose_33_perm_0, x = var_389_cast_fp16)[name = string("transpose_39")]; | |
| tensor<fp16, [1, 6, 1500, 64]> a_cast_fp16 = scaled_dot_product_attention(key = transpose_34, query = transpose_33, value = transpose_35)[name = string("a_cast_fp16")]; | |
| tensor<int32, [4]> var_405 = const()[name = string("op_405"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_3 = const()[name = string("concat_3"), val = tensor<int32, [3]>([1, 1500, 384])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_406_cast_fp16 = transpose(perm = var_405, x = a_cast_fp16)[name = string("transpose_36")]; | |
| tensor<fp16, [1, 1500, 384]> x_47_cast_fp16 = reshape(shape = concat_3, x = var_406_cast_fp16)[name = string("x_47_cast_fp16")]; | |
| tensor<fp16, [384, 384]> const_66_to_fp16 = const()[name = string("const_66_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13759424)))]; | |
| tensor<fp16, [384]> const_67_to_fp16 = const()[name = string("const_67_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14054400)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_21_cast_fp16 = linear(bias = const_67_to_fp16, weight = const_66_to_fp16, x = x_47_cast_fp16)[name = string("linear_21_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> 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_418_axes_0 = const()[name = string("op_418_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_3_mlp_ln_weight_to_fp16 = const()[name = string("blocks_3_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14055232)))]; | |
| tensor<fp16, [384]> blocks_3_mlp_ln_bias_to_fp16 = const()[name = string("blocks_3_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14056064)))]; | |
| tensor<fp16, [1, 1500, 384]> var_418_cast_fp16 = layer_norm(axes = var_418_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_354_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = string("op_418_cast_fp16")]; | |
| tensor<fp16, [1536, 384]> const_68_to_fp16 = const()[name = string("const_68_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14056896)))]; | |
| tensor<fp16, [1536]> const_69_to_fp16 = const()[name = string("const_69_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15236608)))]; | |
| tensor<fp16, [1, 1500, 1536]> linear_22_cast_fp16 = linear(bias = const_69_to_fp16, weight = const_68_to_fp16, x = var_418_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, 1536]> x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = string("x_53_cast_fp16")]; | |
| tensor<fp16, [384, 1536]> const_70_to_fp16 = const()[name = string("const_70_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15239744)))]; | |
| tensor<fp16, [384]> const_71_to_fp16 = const()[name = string("const_71_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16419456)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_23_cast_fp16 = linear(bias = const_71_to_fp16, weight = const_70_to_fp16, x = x_53_cast_fp16)[name = string("linear_23_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> x_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = string("x_cast_fp16")]; | |
| tensor<int32, [1]> var_447_axes_0 = const()[name = string("op_447_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> ln_post_weight_to_fp16 = const()[name = string("ln_post_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16420288)))]; | |
| tensor<fp16, [384]> ln_post_bias_to_fp16 = const()[name = string("ln_post_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16421120)))]; | |
| fp16 var_438_to_fp16 = const()[name = string("op_438_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 384]> output = layer_norm(axes = var_447_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_438_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = string("op_447_cast_fp16")]; | |
| } -> (output); | |
| } |