| 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_32_pad_type_0 = const()[name = string("op_32_pad_type_0"), val = string("custom")]; |
| tensor<int32, [2]> var_32_pad_0 = const()[name = string("op_32_pad_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [1]> var_32_strides_0 = const()[name = string("op_32_strides_0"), val = tensor<int32, [1]>([1])]; |
| tensor<int32, [1]> var_32_dilations_0 = const()[name = string("op_32_dilations_0"), val = tensor<int32, [1]>([1])]; |
| int32 var_32_groups_0 = const()[name = string("op_32_groups_0"), val = int32(1)]; |
| tensor<fp16, [512, 80, 3]> const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = tensor<fp16, [512, 80, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; |
| tensor<fp16, [512]> const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245888)))]; |
| tensor<fp16, [1, 512, 3000]> var_32_cast_fp16 = conv(bias = const_1_to_fp16, dilations = var_32_dilations_0, groups = var_32_groups_0, pad = var_32_pad_0, pad_type = var_32_pad_type_0, strides = var_32_strides_0, weight = const_0_to_fp16, x = logmel_data)[name = string("op_32_cast_fp16")]; |
| string input_1_mode_0 = const()[name = string("input_1_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 512, 3000]> input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_32_cast_fp16)[name = string("input_1_cast_fp16")]; |
| string var_50_pad_type_0 = const()[name = string("op_50_pad_type_0"), val = string("custom")]; |
| tensor<int32, [2]> var_50_pad_0 = const()[name = string("op_50_pad_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [1]> var_50_strides_0 = const()[name = string("op_50_strides_0"), val = tensor<int32, [1]>([2])]; |
| tensor<int32, [1]> var_50_dilations_0 = const()[name = string("op_50_dilations_0"), val = tensor<int32, [1]>([1])]; |
| int32 var_50_groups_0 = const()[name = string("op_50_groups_0"), val = int32(1)]; |
| tensor<fp16, [512, 512, 3]> const_2_to_fp16 = const()[name = string("const_2_to_fp16"), val = tensor<fp16, [512, 512, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246976)))]; |
| tensor<fp16, [512]> const_3_to_fp16 = const()[name = string("const_3_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1819904)))]; |
| tensor<fp16, [1, 512, 1500]> var_50_cast_fp16 = conv(bias = const_3_to_fp16, dilations = var_50_dilations_0, groups = var_50_groups_0, pad = var_50_pad_0, pad_type = var_50_pad_type_0, strides = var_50_strides_0, weight = const_2_to_fp16, x = input_1_cast_fp16)[name = string("op_50_cast_fp16")]; |
| string x_3_mode_0 = const()[name = string("x_3_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 512, 1500]> x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_50_cast_fp16)[name = string("x_3_cast_fp16")]; |
| tensor<int32, [3]> var_56 = const()[name = string("op_56"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<fp16, [1500, 512]> positional_embedding_to_fp16 = const()[name = string("positional_embedding_to_fp16"), val = tensor<fp16, [1500, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1820992)))]; |
| tensor<fp16, [1, 1500, 512]> x_5_cast_fp16 = transpose(perm = var_56, x = x_3_cast_fp16)[name = string("transpose_78")]; |
| tensor<fp16, [1, 1500, 512]> var_59_cast_fp16 = add(x = x_5_cast_fp16, y = positional_embedding_to_fp16)[name = string("op_59_cast_fp16")]; |
| tensor<int32, [1]> var_87_axes_0 = const()[name = string("op_87_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [512]> blocks_0_attn_ln_weight_to_fp16 = const()[name = string("blocks_0_attn_ln_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3357056)))]; |
| tensor<fp16, [512]> blocks_0_attn_ln_bias_to_fp16 = const()[name = string("blocks_0_attn_ln_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3358144)))]; |
| fp16 var_77_to_fp16 = const()[name = string("op_77_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 512]> var_87_cast_fp16 = layer_norm(axes = var_87_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_77_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_59_cast_fp16)[name = string("op_87_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_4_to_fp16 = const()[name = string("const_4_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3359232)))]; |
| tensor<fp16, [512]> const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3883584)))]; |
| tensor<fp16, [1, 1500, 512]> linear_0_cast_fp16 = linear(bias = const_5_to_fp16, weight = const_4_to_fp16, x = var_87_cast_fp16)[name = string("linear_0_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_6_to_fp16 = const()[name = string("const_6_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3884672)))]; |
| tensor<fp16, [512]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4409024)))]; |
| tensor<fp16, [1, 1500, 512]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_6_to_fp16, x = var_87_cast_fp16)[name = string("linear_1_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4410112)))]; |
| tensor<fp16, [512]> const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4934464)))]; |
| tensor<fp16, [1, 1500, 512]> linear_2_cast_fp16 = linear(bias = const_8_to_fp16, weight = const_7_to_fp16, x = var_87_cast_fp16)[name = string("linear_2_cast_fp16")]; |
| tensor<int32, [4]> var_111 = const()[name = string("op_111"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_112_cast_fp16 = reshape(shape = var_111, x = linear_0_cast_fp16)[name = string("op_112_cast_fp16")]; |
| tensor<int32, [4]> var_117 = const()[name = string("op_117"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_118_cast_fp16 = reshape(shape = var_117, x = linear_1_cast_fp16)[name = string("op_118_cast_fp16")]; |
| tensor<int32, [4]> var_123 = const()[name = string("op_123"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_124_cast_fp16 = reshape(shape = var_123, x = linear_2_cast_fp16)[name = string("op_124_cast_fp16")]; |
| tensor<int32, [4]> transpose_36_perm_0 = const()[name = string("transpose_36_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_37_perm_0 = const()[name = string("transpose_37_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_38_perm_0 = const()[name = string("transpose_38_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_38 = transpose(perm = transpose_38_perm_0, x = var_124_cast_fp16)[name = string("transpose_75")]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_37 = transpose(perm = transpose_37_perm_0, x = var_118_cast_fp16)[name = string("transpose_76")]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_36 = transpose(perm = transpose_36_perm_0, x = var_112_cast_fp16)[name = string("transpose_77")]; |
| tensor<fp16, [1, 8, 1500, 64]> a_1_cast_fp16 = scaled_dot_product_attention(key = transpose_37, query = transpose_36, value = transpose_38)[name = string("a_1_cast_fp16")]; |
| tensor<int32, [4]> var_128 = const()[name = string("op_128"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_0 = const()[name = string("concat_0"), val = tensor<int32, [3]>([1, 1500, 512])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_129_cast_fp16 = transpose(perm = var_128, x = a_1_cast_fp16)[name = string("transpose_74")]; |
| tensor<fp16, [1, 1500, 512]> x_11_cast_fp16 = reshape(shape = concat_0, x = var_129_cast_fp16)[name = string("x_11_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_15_to_fp16 = const()[name = string("const_15_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4935552)))]; |
| tensor<fp16, [512]> const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5459904)))]; |
| tensor<fp16, [1, 1500, 512]> 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, 512]> x_13_cast_fp16 = add(x = var_59_cast_fp16, y = linear_3_cast_fp16)[name = string("x_13_cast_fp16")]; |
| tensor<int32, [1]> var_141_axes_0 = const()[name = string("op_141_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [512]> blocks_0_mlp_ln_weight_to_fp16 = const()[name = string("blocks_0_mlp_ln_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5460992)))]; |
| tensor<fp16, [512]> blocks_0_mlp_ln_bias_to_fp16 = const()[name = string("blocks_0_mlp_ln_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5462080)))]; |
| tensor<fp16, [1, 1500, 512]> var_141_cast_fp16 = layer_norm(axes = var_141_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_77_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = string("op_141_cast_fp16")]; |
| tensor<fp16, [2048, 512]> const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = tensor<fp16, [2048, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5463168)))]; |
| tensor<fp16, [2048]> const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7560384)))]; |
| tensor<fp16, [1, 1500, 2048]> linear_4_cast_fp16 = linear(bias = const_18_to_fp16, weight = const_17_to_fp16, x = var_141_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, 2048]> x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = string("x_17_cast_fp16")]; |
| tensor<fp16, [512, 2048]> const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = tensor<fp16, [512, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7564544)))]; |
| tensor<fp16, [512]> const_20_to_fp16 = const()[name = string("const_20_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9661760)))]; |
| tensor<fp16, [1, 1500, 512]> 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, 512]> 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_182_axes_0 = const()[name = string("op_182_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [512]> blocks_1_attn_ln_weight_to_fp16 = const()[name = string("blocks_1_attn_ln_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9662848)))]; |
| tensor<fp16, [512]> blocks_1_attn_ln_bias_to_fp16 = const()[name = string("blocks_1_attn_ln_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9663936)))]; |
| fp16 var_172_to_fp16 = const()[name = string("op_172_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 512]> var_182_cast_fp16 = layer_norm(axes = var_182_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_172_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = string("op_182_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_21_to_fp16 = const()[name = string("const_21_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9665024)))]; |
| tensor<fp16, [512]> const_22_to_fp16 = const()[name = string("const_22_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10189376)))]; |
| tensor<fp16, [1, 1500, 512]> linear_6_cast_fp16 = linear(bias = const_22_to_fp16, weight = const_21_to_fp16, x = var_182_cast_fp16)[name = string("linear_6_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_23_to_fp16 = const()[name = string("const_23_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10190464)))]; |
| tensor<fp16, [1, 1500, 512]> linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_23_to_fp16, x = var_182_cast_fp16)[name = string("linear_7_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10714816)))]; |
| tensor<fp16, [512]> const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11239168)))]; |
| tensor<fp16, [1, 1500, 512]> linear_8_cast_fp16 = linear(bias = const_25_to_fp16, weight = const_24_to_fp16, x = var_182_cast_fp16)[name = string("linear_8_cast_fp16")]; |
| tensor<int32, [4]> var_206 = const()[name = string("op_206"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_207_cast_fp16 = reshape(shape = var_206, x = linear_6_cast_fp16)[name = string("op_207_cast_fp16")]; |
| tensor<int32, [4]> var_212 = const()[name = string("op_212"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_213_cast_fp16 = reshape(shape = var_212, x = linear_7_cast_fp16)[name = string("op_213_cast_fp16")]; |
| tensor<int32, [4]> var_218 = const()[name = string("op_218"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_219_cast_fp16 = reshape(shape = var_218, x = linear_8_cast_fp16)[name = string("op_219_cast_fp16")]; |
| tensor<int32, [4]> transpose_39_perm_0 = const()[name = string("transpose_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_40_perm_0 = const()[name = string("transpose_40_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_41_perm_0 = const()[name = string("transpose_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_41 = transpose(perm = transpose_41_perm_0, x = var_219_cast_fp16)[name = string("transpose_71")]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_40 = transpose(perm = transpose_40_perm_0, x = var_213_cast_fp16)[name = string("transpose_72")]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_39 = transpose(perm = transpose_39_perm_0, x = var_207_cast_fp16)[name = string("transpose_73")]; |
| tensor<fp16, [1, 8, 1500, 64]> a_3_cast_fp16 = scaled_dot_product_attention(key = transpose_40, query = transpose_39, value = transpose_41)[name = string("a_3_cast_fp16")]; |
| tensor<int32, [4]> var_223 = const()[name = string("op_223"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_1 = const()[name = string("concat_1"), val = tensor<int32, [3]>([1, 1500, 512])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_224_cast_fp16 = transpose(perm = var_223, x = a_3_cast_fp16)[name = string("transpose_70")]; |
| tensor<fp16, [1, 1500, 512]> x_23_cast_fp16 = reshape(shape = concat_1, x = var_224_cast_fp16)[name = string("x_23_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_32_to_fp16 = const()[name = string("const_32_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11240256)))]; |
| tensor<fp16, [512]> const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11764608)))]; |
| tensor<fp16, [1, 1500, 512]> 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, 512]> 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_236_axes_0 = const()[name = string("op_236_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [512]> blocks_1_mlp_ln_weight_to_fp16 = const()[name = string("blocks_1_mlp_ln_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11765696)))]; |
| tensor<fp16, [512]> blocks_1_mlp_ln_bias_to_fp16 = const()[name = string("blocks_1_mlp_ln_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11766784)))]; |
| tensor<fp16, [1, 1500, 512]> var_236_cast_fp16 = layer_norm(axes = var_236_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_172_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = string("op_236_cast_fp16")]; |
| tensor<fp16, [2048, 512]> const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = tensor<fp16, [2048, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11767872)))]; |
| tensor<fp16, [2048]> const_35_to_fp16 = const()[name = string("const_35_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13865088)))]; |
| tensor<fp16, [1, 1500, 2048]> linear_10_cast_fp16 = linear(bias = const_35_to_fp16, weight = const_34_to_fp16, x = var_236_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, 2048]> x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = string("x_29_cast_fp16")]; |
| tensor<fp16, [512, 2048]> const_36_to_fp16 = const()[name = string("const_36_to_fp16"), val = tensor<fp16, [512, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13869248)))]; |
| tensor<fp16, [512]> const_37_to_fp16 = const()[name = string("const_37_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15966464)))]; |
| tensor<fp16, [1, 1500, 512]> 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, 512]> 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_277_axes_0 = const()[name = string("op_277_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [512]> blocks_2_attn_ln_weight_to_fp16 = const()[name = string("blocks_2_attn_ln_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15967552)))]; |
| tensor<fp16, [512]> blocks_2_attn_ln_bias_to_fp16 = const()[name = string("blocks_2_attn_ln_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15968640)))]; |
| fp16 var_267_to_fp16 = const()[name = string("op_267_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 512]> var_277_cast_fp16 = layer_norm(axes = var_277_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_267_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = string("op_277_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_38_to_fp16 = const()[name = string("const_38_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15969728)))]; |
| tensor<fp16, [512]> const_39_to_fp16 = const()[name = string("const_39_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16494080)))]; |
| tensor<fp16, [1, 1500, 512]> linear_12_cast_fp16 = linear(bias = const_39_to_fp16, weight = const_38_to_fp16, x = var_277_cast_fp16)[name = string("linear_12_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_40_to_fp16 = const()[name = string("const_40_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16495168)))]; |
| tensor<fp16, [1, 1500, 512]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_40_to_fp16, x = var_277_cast_fp16)[name = string("linear_13_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_41_to_fp16 = const()[name = string("const_41_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17019520)))]; |
| tensor<fp16, [512]> const_42_to_fp16 = const()[name = string("const_42_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17543872)))]; |
| tensor<fp16, [1, 1500, 512]> linear_14_cast_fp16 = linear(bias = const_42_to_fp16, weight = const_41_to_fp16, x = var_277_cast_fp16)[name = string("linear_14_cast_fp16")]; |
| tensor<int32, [4]> var_301 = const()[name = string("op_301"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_302_cast_fp16 = reshape(shape = var_301, x = linear_12_cast_fp16)[name = string("op_302_cast_fp16")]; |
| tensor<int32, [4]> var_307 = const()[name = string("op_307"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_308_cast_fp16 = reshape(shape = var_307, x = linear_13_cast_fp16)[name = string("op_308_cast_fp16")]; |
| tensor<int32, [4]> var_313 = const()[name = string("op_313"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_314_cast_fp16 = reshape(shape = var_313, x = linear_14_cast_fp16)[name = string("op_314_cast_fp16")]; |
| tensor<int32, [4]> transpose_42_perm_0 = const()[name = string("transpose_42_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_43_perm_0 = const()[name = string("transpose_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_44_perm_0 = const()[name = string("transpose_44_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_44 = transpose(perm = transpose_44_perm_0, x = var_314_cast_fp16)[name = string("transpose_67")]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_43 = transpose(perm = transpose_43_perm_0, x = var_308_cast_fp16)[name = string("transpose_68")]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_42 = transpose(perm = transpose_42_perm_0, x = var_302_cast_fp16)[name = string("transpose_69")]; |
| tensor<fp16, [1, 8, 1500, 64]> a_5_cast_fp16 = scaled_dot_product_attention(key = transpose_43, query = transpose_42, value = transpose_44)[name = string("a_5_cast_fp16")]; |
| tensor<int32, [4]> var_318 = const()[name = string("op_318"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_2 = const()[name = string("concat_2"), val = tensor<int32, [3]>([1, 1500, 512])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_319_cast_fp16 = transpose(perm = var_318, x = a_5_cast_fp16)[name = string("transpose_66")]; |
| tensor<fp16, [1, 1500, 512]> x_35_cast_fp16 = reshape(shape = concat_2, x = var_319_cast_fp16)[name = string("x_35_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_49_to_fp16 = const()[name = string("const_49_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17544960)))]; |
| tensor<fp16, [512]> const_50_to_fp16 = const()[name = string("const_50_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18069312)))]; |
| tensor<fp16, [1, 1500, 512]> 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, 512]> 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_331_axes_0 = const()[name = string("op_331_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [512]> blocks_2_mlp_ln_weight_to_fp16 = const()[name = string("blocks_2_mlp_ln_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18070400)))]; |
| tensor<fp16, [512]> blocks_2_mlp_ln_bias_to_fp16 = const()[name = string("blocks_2_mlp_ln_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18071488)))]; |
| tensor<fp16, [1, 1500, 512]> var_331_cast_fp16 = layer_norm(axes = var_331_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_267_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = string("op_331_cast_fp16")]; |
| tensor<fp16, [2048, 512]> const_51_to_fp16 = const()[name = string("const_51_to_fp16"), val = tensor<fp16, [2048, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18072576)))]; |
| tensor<fp16, [2048]> const_52_to_fp16 = const()[name = string("const_52_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20169792)))]; |
| tensor<fp16, [1, 1500, 2048]> linear_16_cast_fp16 = linear(bias = const_52_to_fp16, weight = const_51_to_fp16, x = var_331_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, 2048]> x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = string("x_41_cast_fp16")]; |
| tensor<fp16, [512, 2048]> const_53_to_fp16 = const()[name = string("const_53_to_fp16"), val = tensor<fp16, [512, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20173952)))]; |
| tensor<fp16, [512]> const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22271168)))]; |
| tensor<fp16, [1, 1500, 512]> 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, 512]> 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_372_axes_0 = const()[name = string("op_372_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [512]> blocks_3_attn_ln_weight_to_fp16 = const()[name = string("blocks_3_attn_ln_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22272256)))]; |
| tensor<fp16, [512]> blocks_3_attn_ln_bias_to_fp16 = const()[name = string("blocks_3_attn_ln_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22273344)))]; |
| fp16 var_362_to_fp16 = const()[name = string("op_362_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 512]> var_372_cast_fp16 = layer_norm(axes = var_372_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_362_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = string("op_372_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_55_to_fp16 = const()[name = string("const_55_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22274432)))]; |
| tensor<fp16, [512]> const_56_to_fp16 = const()[name = string("const_56_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22798784)))]; |
| tensor<fp16, [1, 1500, 512]> linear_18_cast_fp16 = linear(bias = const_56_to_fp16, weight = const_55_to_fp16, x = var_372_cast_fp16)[name = string("linear_18_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22799872)))]; |
| tensor<fp16, [1, 1500, 512]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_57_to_fp16, x = var_372_cast_fp16)[name = string("linear_19_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_58_to_fp16 = const()[name = string("const_58_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23324224)))]; |
| tensor<fp16, [512]> const_59_to_fp16 = const()[name = string("const_59_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23848576)))]; |
| tensor<fp16, [1, 1500, 512]> linear_20_cast_fp16 = linear(bias = const_59_to_fp16, weight = const_58_to_fp16, x = var_372_cast_fp16)[name = string("linear_20_cast_fp16")]; |
| tensor<int32, [4]> var_396 = const()[name = string("op_396"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_397_cast_fp16 = reshape(shape = var_396, x = linear_18_cast_fp16)[name = string("op_397_cast_fp16")]; |
| tensor<int32, [4]> var_402 = const()[name = string("op_402"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_403_cast_fp16 = reshape(shape = var_402, x = linear_19_cast_fp16)[name = string("op_403_cast_fp16")]; |
| tensor<int32, [4]> var_408 = const()[name = string("op_408"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_409_cast_fp16 = reshape(shape = var_408, x = linear_20_cast_fp16)[name = string("op_409_cast_fp16")]; |
| tensor<int32, [4]> transpose_45_perm_0 = const()[name = string("transpose_45_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_46_perm_0 = const()[name = string("transpose_46_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_47_perm_0 = const()[name = string("transpose_47_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_47 = transpose(perm = transpose_47_perm_0, x = var_409_cast_fp16)[name = string("transpose_63")]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_46 = transpose(perm = transpose_46_perm_0, x = var_403_cast_fp16)[name = string("transpose_64")]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_45 = transpose(perm = transpose_45_perm_0, x = var_397_cast_fp16)[name = string("transpose_65")]; |
| tensor<fp16, [1, 8, 1500, 64]> a_7_cast_fp16 = scaled_dot_product_attention(key = transpose_46, query = transpose_45, value = transpose_47)[name = string("a_7_cast_fp16")]; |
| tensor<int32, [4]> var_413 = const()[name = string("op_413"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_3 = const()[name = string("concat_3"), val = tensor<int32, [3]>([1, 1500, 512])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_414_cast_fp16 = transpose(perm = var_413, x = a_7_cast_fp16)[name = string("transpose_62")]; |
| tensor<fp16, [1, 1500, 512]> x_47_cast_fp16 = reshape(shape = concat_3, x = var_414_cast_fp16)[name = string("x_47_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_66_to_fp16 = const()[name = string("const_66_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23849664)))]; |
| tensor<fp16, [512]> const_67_to_fp16 = const()[name = string("const_67_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24374016)))]; |
| tensor<fp16, [1, 1500, 512]> 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, 512]> 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_426_axes_0 = const()[name = string("op_426_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [512]> blocks_3_mlp_ln_weight_to_fp16 = const()[name = string("blocks_3_mlp_ln_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24375104)))]; |
| tensor<fp16, [512]> blocks_3_mlp_ln_bias_to_fp16 = const()[name = string("blocks_3_mlp_ln_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24376192)))]; |
| tensor<fp16, [1, 1500, 512]> var_426_cast_fp16 = layer_norm(axes = var_426_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_362_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = string("op_426_cast_fp16")]; |
| tensor<fp16, [2048, 512]> const_68_to_fp16 = const()[name = string("const_68_to_fp16"), val = tensor<fp16, [2048, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24377280)))]; |
| tensor<fp16, [2048]> const_69_to_fp16 = const()[name = string("const_69_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26474496)))]; |
| tensor<fp16, [1, 1500, 2048]> linear_22_cast_fp16 = linear(bias = const_69_to_fp16, weight = const_68_to_fp16, x = var_426_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, 2048]> x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = string("x_53_cast_fp16")]; |
| tensor<fp16, [512, 2048]> const_70_to_fp16 = const()[name = string("const_70_to_fp16"), val = tensor<fp16, [512, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26478656)))]; |
| tensor<fp16, [512]> const_71_to_fp16 = const()[name = string("const_71_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28575872)))]; |
| tensor<fp16, [1, 1500, 512]> 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, 512]> x_55_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = string("x_55_cast_fp16")]; |
| tensor<int32, [1]> var_467_axes_0 = const()[name = string("op_467_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [512]> blocks_4_attn_ln_weight_to_fp16 = const()[name = string("blocks_4_attn_ln_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28576960)))]; |
| tensor<fp16, [512]> blocks_4_attn_ln_bias_to_fp16 = const()[name = string("blocks_4_attn_ln_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28578048)))]; |
| fp16 var_457_to_fp16 = const()[name = string("op_457_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 512]> var_467_cast_fp16 = layer_norm(axes = var_467_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_457_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast_fp16)[name = string("op_467_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_72_to_fp16 = const()[name = string("const_72_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28579136)))]; |
| tensor<fp16, [512]> const_73_to_fp16 = const()[name = string("const_73_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29103488)))]; |
| tensor<fp16, [1, 1500, 512]> linear_24_cast_fp16 = linear(bias = const_73_to_fp16, weight = const_72_to_fp16, x = var_467_cast_fp16)[name = string("linear_24_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_74_to_fp16 = const()[name = string("const_74_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29104576)))]; |
| tensor<fp16, [1, 1500, 512]> linear_25_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_74_to_fp16, x = var_467_cast_fp16)[name = string("linear_25_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_75_to_fp16 = const()[name = string("const_75_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29628928)))]; |
| tensor<fp16, [512]> const_76_to_fp16 = const()[name = string("const_76_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30153280)))]; |
| tensor<fp16, [1, 1500, 512]> linear_26_cast_fp16 = linear(bias = const_76_to_fp16, weight = const_75_to_fp16, x = var_467_cast_fp16)[name = string("linear_26_cast_fp16")]; |
| tensor<int32, [4]> var_491 = const()[name = string("op_491"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_492_cast_fp16 = reshape(shape = var_491, x = linear_24_cast_fp16)[name = string("op_492_cast_fp16")]; |
| tensor<int32, [4]> var_497 = const()[name = string("op_497"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_498_cast_fp16 = reshape(shape = var_497, x = linear_25_cast_fp16)[name = string("op_498_cast_fp16")]; |
| tensor<int32, [4]> var_503 = const()[name = string("op_503"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_504_cast_fp16 = reshape(shape = var_503, x = linear_26_cast_fp16)[name = string("op_504_cast_fp16")]; |
| tensor<int32, [4]> transpose_48_perm_0 = const()[name = string("transpose_48_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_49_perm_0 = const()[name = string("transpose_49_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_50_perm_0 = const()[name = string("transpose_50_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_50 = transpose(perm = transpose_50_perm_0, x = var_504_cast_fp16)[name = string("transpose_59")]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_49 = transpose(perm = transpose_49_perm_0, x = var_498_cast_fp16)[name = string("transpose_60")]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_48 = transpose(perm = transpose_48_perm_0, x = var_492_cast_fp16)[name = string("transpose_61")]; |
| tensor<fp16, [1, 8, 1500, 64]> a_9_cast_fp16 = scaled_dot_product_attention(key = transpose_49, query = transpose_48, value = transpose_50)[name = string("a_9_cast_fp16")]; |
| tensor<int32, [4]> var_508 = const()[name = string("op_508"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_4 = const()[name = string("concat_4"), val = tensor<int32, [3]>([1, 1500, 512])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_509_cast_fp16 = transpose(perm = var_508, x = a_9_cast_fp16)[name = string("transpose_58")]; |
| tensor<fp16, [1, 1500, 512]> x_59_cast_fp16 = reshape(shape = concat_4, x = var_509_cast_fp16)[name = string("x_59_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_83_to_fp16 = const()[name = string("const_83_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30154368)))]; |
| tensor<fp16, [512]> const_84_to_fp16 = const()[name = string("const_84_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30678720)))]; |
| tensor<fp16, [1, 1500, 512]> linear_27_cast_fp16 = linear(bias = const_84_to_fp16, weight = const_83_to_fp16, x = x_59_cast_fp16)[name = string("linear_27_cast_fp16")]; |
| tensor<fp16, [1, 1500, 512]> x_61_cast_fp16 = add(x = x_55_cast_fp16, y = linear_27_cast_fp16)[name = string("x_61_cast_fp16")]; |
| tensor<int32, [1]> var_521_axes_0 = const()[name = string("op_521_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [512]> blocks_4_mlp_ln_weight_to_fp16 = const()[name = string("blocks_4_mlp_ln_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30679808)))]; |
| tensor<fp16, [512]> blocks_4_mlp_ln_bias_to_fp16 = const()[name = string("blocks_4_mlp_ln_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30680896)))]; |
| tensor<fp16, [1, 1500, 512]> var_521_cast_fp16 = layer_norm(axes = var_521_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_457_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast_fp16)[name = string("op_521_cast_fp16")]; |
| tensor<fp16, [2048, 512]> const_85_to_fp16 = const()[name = string("const_85_to_fp16"), val = tensor<fp16, [2048, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30681984)))]; |
| tensor<fp16, [2048]> const_86_to_fp16 = const()[name = string("const_86_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32779200)))]; |
| tensor<fp16, [1, 1500, 2048]> linear_28_cast_fp16 = linear(bias = const_86_to_fp16, weight = const_85_to_fp16, x = var_521_cast_fp16)[name = string("linear_28_cast_fp16")]; |
| string x_65_mode_0 = const()[name = string("x_65_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 2048]> x_65_cast_fp16 = gelu(mode = x_65_mode_0, x = linear_28_cast_fp16)[name = string("x_65_cast_fp16")]; |
| tensor<fp16, [512, 2048]> const_87_to_fp16 = const()[name = string("const_87_to_fp16"), val = tensor<fp16, [512, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32783360)))]; |
| tensor<fp16, [512]> const_88_to_fp16 = const()[name = string("const_88_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34880576)))]; |
| tensor<fp16, [1, 1500, 512]> linear_29_cast_fp16 = linear(bias = const_88_to_fp16, weight = const_87_to_fp16, x = x_65_cast_fp16)[name = string("linear_29_cast_fp16")]; |
| tensor<fp16, [1, 1500, 512]> x_67_cast_fp16 = add(x = x_61_cast_fp16, y = linear_29_cast_fp16)[name = string("x_67_cast_fp16")]; |
| tensor<int32, [1]> var_562_axes_0 = const()[name = string("op_562_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [512]> blocks_5_attn_ln_weight_to_fp16 = const()[name = string("blocks_5_attn_ln_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34881664)))]; |
| tensor<fp16, [512]> blocks_5_attn_ln_bias_to_fp16 = const()[name = string("blocks_5_attn_ln_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34882752)))]; |
| fp16 var_552_to_fp16 = const()[name = string("op_552_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 512]> var_562_cast_fp16 = layer_norm(axes = var_562_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_552_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast_fp16)[name = string("op_562_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_89_to_fp16 = const()[name = string("const_89_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34883840)))]; |
| tensor<fp16, [512]> const_90_to_fp16 = const()[name = string("const_90_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35408192)))]; |
| tensor<fp16, [1, 1500, 512]> linear_30_cast_fp16 = linear(bias = const_90_to_fp16, weight = const_89_to_fp16, x = var_562_cast_fp16)[name = string("linear_30_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_91_to_fp16 = const()[name = string("const_91_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35409280)))]; |
| tensor<fp16, [1, 1500, 512]> linear_31_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_91_to_fp16, x = var_562_cast_fp16)[name = string("linear_31_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_92_to_fp16 = const()[name = string("const_92_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35933632)))]; |
| tensor<fp16, [512]> const_93_to_fp16 = const()[name = string("const_93_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36457984)))]; |
| tensor<fp16, [1, 1500, 512]> linear_32_cast_fp16 = linear(bias = const_93_to_fp16, weight = const_92_to_fp16, x = var_562_cast_fp16)[name = string("linear_32_cast_fp16")]; |
| tensor<int32, [4]> var_586 = const()[name = string("op_586"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_587_cast_fp16 = reshape(shape = var_586, x = linear_30_cast_fp16)[name = string("op_587_cast_fp16")]; |
| tensor<int32, [4]> var_592 = const()[name = string("op_592"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_593_cast_fp16 = reshape(shape = var_592, x = linear_31_cast_fp16)[name = string("op_593_cast_fp16")]; |
| tensor<int32, [4]> var_598 = const()[name = string("op_598"), val = tensor<int32, [4]>([1, 1500, 8, -1])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_599_cast_fp16 = reshape(shape = var_598, x = linear_32_cast_fp16)[name = string("op_599_cast_fp16")]; |
| tensor<int32, [4]> transpose_51_perm_0 = const()[name = string("transpose_51_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_52_perm_0 = const()[name = string("transpose_52_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_53_perm_0 = const()[name = string("transpose_53_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_53 = transpose(perm = transpose_53_perm_0, x = var_599_cast_fp16)[name = string("transpose_55")]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_52 = transpose(perm = transpose_52_perm_0, x = var_593_cast_fp16)[name = string("transpose_56")]; |
| tensor<fp16, [1, 8, 1500, 64]> transpose_51 = transpose(perm = transpose_51_perm_0, x = var_587_cast_fp16)[name = string("transpose_57")]; |
| tensor<fp16, [1, 8, 1500, 64]> a_cast_fp16 = scaled_dot_product_attention(key = transpose_52, query = transpose_51, value = transpose_53)[name = string("a_cast_fp16")]; |
| tensor<int32, [4]> var_603 = const()[name = string("op_603"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_5 = const()[name = string("concat_5"), val = tensor<int32, [3]>([1, 1500, 512])]; |
| tensor<fp16, [1, 1500, 8, 64]> var_604_cast_fp16 = transpose(perm = var_603, x = a_cast_fp16)[name = string("transpose_54")]; |
| tensor<fp16, [1, 1500, 512]> x_71_cast_fp16 = reshape(shape = concat_5, x = var_604_cast_fp16)[name = string("x_71_cast_fp16")]; |
| tensor<fp16, [512, 512]> const_100_to_fp16 = const()[name = string("const_100_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36459072)))]; |
| tensor<fp16, [512]> const_101_to_fp16 = const()[name = string("const_101_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36983424)))]; |
| tensor<fp16, [1, 1500, 512]> linear_33_cast_fp16 = linear(bias = const_101_to_fp16, weight = const_100_to_fp16, x = x_71_cast_fp16)[name = string("linear_33_cast_fp16")]; |
| tensor<fp16, [1, 1500, 512]> x_73_cast_fp16 = add(x = x_67_cast_fp16, y = linear_33_cast_fp16)[name = string("x_73_cast_fp16")]; |
| tensor<int32, [1]> var_616_axes_0 = const()[name = string("op_616_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [512]> blocks_5_mlp_ln_weight_to_fp16 = const()[name = string("blocks_5_mlp_ln_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36984512)))]; |
| tensor<fp16, [512]> blocks_5_mlp_ln_bias_to_fp16 = const()[name = string("blocks_5_mlp_ln_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36985600)))]; |
| tensor<fp16, [1, 1500, 512]> var_616_cast_fp16 = layer_norm(axes = var_616_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_552_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast_fp16)[name = string("op_616_cast_fp16")]; |
| tensor<fp16, [2048, 512]> const_102_to_fp16 = const()[name = string("const_102_to_fp16"), val = tensor<fp16, [2048, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36986688)))]; |
| tensor<fp16, [2048]> const_103_to_fp16 = const()[name = string("const_103_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39083904)))]; |
| tensor<fp16, [1, 1500, 2048]> linear_34_cast_fp16 = linear(bias = const_103_to_fp16, weight = const_102_to_fp16, x = var_616_cast_fp16)[name = string("linear_34_cast_fp16")]; |
| string x_77_mode_0 = const()[name = string("x_77_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 2048]> x_77_cast_fp16 = gelu(mode = x_77_mode_0, x = linear_34_cast_fp16)[name = string("x_77_cast_fp16")]; |
| tensor<fp16, [512, 2048]> const_104_to_fp16 = const()[name = string("const_104_to_fp16"), val = tensor<fp16, [512, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39088064)))]; |
| tensor<fp16, [512]> const_105_to_fp16 = const()[name = string("const_105_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41185280)))]; |
| tensor<fp16, [1, 1500, 512]> linear_35_cast_fp16 = linear(bias = const_105_to_fp16, weight = const_104_to_fp16, x = x_77_cast_fp16)[name = string("linear_35_cast_fp16")]; |
| tensor<fp16, [1, 1500, 512]> x_cast_fp16 = add(x = x_73_cast_fp16, y = linear_35_cast_fp16)[name = string("x_cast_fp16")]; |
| tensor<int32, [1]> var_645_axes_0 = const()[name = string("op_645_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [512]> ln_post_weight_to_fp16 = const()[name = string("ln_post_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41186368)))]; |
| tensor<fp16, [512]> ln_post_bias_to_fp16 = const()[name = string("ln_post_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41187456)))]; |
| fp16 var_636_to_fp16 = const()[name = string("op_636_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 512]> output = layer_norm(axes = var_645_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_636_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = string("op_645_cast_fp16")]; |
| } -> (output); |
| } |
