| program(1.0) | |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3405.2.1"}, {"coremlc-version", "3404.23.1"}, {"coremltools-component-torch", "2.6.0+cu124"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.3.0"}})] | |
| { | |
| func main<ios16>(tensor<fp32, [?, 3, 224, 224]> colorImage) [FlexibleShapeInformation = tuple<tuple<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>, tuple<tensor<string, []>, dict<tensor<string, []>, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"colorImage", [1, 3, 224, 224]}}), ("RangeDims", {{"colorImage", [[1, 1], [3, 3], [224, 224], [224, 224]]}})))] { | |
| tensor<fp32, []> colorImage__scaled___y_0 = const()[name = tensor<string, []>("colorImage__scaled___y_0"), val = tensor<fp32, []>(0x1.de77c4p-7)]; | |
| tensor<fp32, [?, 3, 224, 224]> colorImage__scaled__ = mul(x = colorImage, y = colorImage__scaled___y_0)[name = tensor<string, []>("colorImage__scaled__")]; | |
| tensor<fp32, [1, 3, 1, 1]> colorImage__biased___y_0 = const()[name = tensor<string, []>("colorImage__biased___y_0"), val = tensor<fp32, [1, 3, 1, 1]>([[[[-0x1.cad1b8p+0]], [[-0x1.c0897p+0]], [[-0x1.7aefaep+0]]]])]; | |
| tensor<fp32, [?, 3, 224, 224]> colorImage__biased__ = add(x = colorImage__scaled__, y = colorImage__biased___y_0)[name = tensor<string, []>("colorImage__biased__")]; | |
| tensor<string, []> x_3_pad_type_0 = const()[name = tensor<string, []>("x_3_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> x_3_strides_0 = const()[name = tensor<string, []>("x_3_strides_0"), val = tensor<int32, [2]>([32, 32])]; | |
| tensor<int32, [4]> x_3_pad_0 = const()[name = tensor<string, []>("x_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> x_3_dilations_0 = const()[name = tensor<string, []>("x_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> x_3_groups_0 = const()[name = tensor<string, []>("x_3_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<string, []> colorImage_to_fp16_dtype_0 = const()[name = tensor<string, []>("colorImage_to_fp16_dtype_0"), val = tensor<string, []>("fp16")]; | |
| tensor<fp16, [768, 3, 32, 32]> conv1_weight_to_fp16 = const()[name = tensor<string, []>("conv1_weight_to_fp16"), val = tensor<fp16, [768, 3, 32, 32]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; | |
| tensor<fp16, [?, 3, 224, 224]> colorImage_to_fp16 = cast(dtype = colorImage_to_fp16_dtype_0, x = colorImage__biased__)[name = tensor<string, []>("cast_186")]; | |
| tensor<fp16, [?, 768, 7, 7]> x_3_cast_fp16 = conv(dilations = x_3_dilations_0, groups = x_3_groups_0, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = x_3_strides_0, weight = conv1_weight_to_fp16, x = colorImage_to_fp16)[name = tensor<string, []>("x_3_cast_fp16")]; | |
| tensor<int32, [4]> var_22_shape_cast_fp16 = shape(x = x_3_cast_fp16)[name = tensor<string, []>("op_22_shape_cast_fp16")]; | |
| tensor<int32, []> gather_0_indices_0 = const()[name = tensor<string, []>("gather_0_indices_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> gather_0_axis_0 = const()[name = tensor<string, []>("gather_0_axis_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> gather_0_batch_dims_0 = const()[name = tensor<string, []>("gather_0_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> gather_0 = gather(axis = gather_0_axis_0, batch_dims = gather_0_batch_dims_0, indices = gather_0_indices_0, x = var_22_shape_cast_fp16)[name = tensor<string, []>("gather_0")]; | |
| tensor<int32, []> var_28 = const()[name = tensor<string, []>("op_28"), val = tensor<int32, []>(768)]; | |
| tensor<int32, []> var_29 = const()[name = tensor<string, []>("op_29"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, []> concat_0_axis_0 = const()[name = tensor<string, []>("concat_0_axis_0"), val = tensor<int32, []>(0)]; | |
| tensor<bool, []> concat_0_interleave_0 = const()[name = tensor<string, []>("concat_0_interleave_0"), val = tensor<bool, []>(false)]; | |
| tensor<int32, [3]> concat_0 = concat(axis = concat_0_axis_0, interleave = concat_0_interleave_0, values = (gather_0, var_28, var_29))[name = tensor<string, []>("concat_0")]; | |
| tensor<fp16, [?, 768, 49]> x_5_cast_fp16 = reshape(shape = concat_0, x = x_3_cast_fp16)[name = tensor<string, []>("x_5_cast_fp16")]; | |
| tensor<int32, [3]> var_35 = const()[name = tensor<string, []>("op_35"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<fp16, [?, 49, 768]> x_7_cast_fp16 = transpose(perm = var_35, x = x_5_cast_fp16)[name = tensor<string, []>("transpose_63")]; | |
| tensor<int32, [3]> var_43_shape_cast_fp16 = shape(x = x_7_cast_fp16)[name = tensor<string, []>("op_43_shape_cast_fp16")]; | |
| tensor<int32, []> gather_2_indices_0 = const()[name = tensor<string, []>("gather_2_indices_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> gather_2_axis_0 = const()[name = tensor<string, []>("gather_2_axis_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> gather_2_batch_dims_0 = const()[name = tensor<string, []>("gather_2_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> gather_2 = gather(axis = gather_2_axis_0, batch_dims = gather_2_batch_dims_0, indices = gather_2_indices_0, x = var_43_shape_cast_fp16)[name = tensor<string, []>("gather_2")]; | |
| tensor<int32, []> var_49 = const()[name = tensor<string, []>("op_49"), val = tensor<int32, []>(768)]; | |
| tensor<int32, []> var_50 = const()[name = tensor<string, []>("op_50"), val = tensor<int32, []>(1)]; | |
| tensor<int32, []> concat_1_axis_0 = const()[name = tensor<string, []>("concat_1_axis_0"), val = tensor<int32, []>(0)]; | |
| tensor<bool, []> concat_1_interleave_0 = const()[name = tensor<string, []>("concat_1_interleave_0"), val = tensor<bool, []>(false)]; | |
| tensor<int32, [3]> concat_1 = concat(axis = concat_1_axis_0, interleave = concat_1_interleave_0, values = (gather_2, var_50, var_49))[name = tensor<string, []>("concat_1")]; | |
| tensor<fp16, []> var_56_value_0_to_fp16 = const()[name = tensor<string, []>("op_56_value_0_to_fp16"), val = tensor<fp16, []>(0x0p+0)]; | |
| tensor<fp16, [?, 1, 768]> var_56_cast_fp16 = fill(shape = concat_1, value = var_56_value_0_to_fp16)[name = tensor<string, []>("op_56_cast_fp16")]; | |
| tensor<fp16, [768]> const_0_to_fp16 = const()[name = tensor<string, []>("const_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4718720)))]; | |
| tensor<fp16, [?, 1, 768]> var_58_cast_fp16 = add(x = const_0_to_fp16, y = var_56_cast_fp16)[name = tensor<string, []>("op_58_cast_fp16")]; | |
| tensor<int32, []> var_60 = const()[name = tensor<string, []>("op_60"), val = tensor<int32, []>(1)]; | |
| tensor<bool, []> x_9_interleave_0 = const()[name = tensor<string, []>("x_9_interleave_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 50, 768]> x_9_cast_fp16 = concat(axis = var_60, interleave = x_9_interleave_0, values = (var_58_cast_fp16, x_7_cast_fp16))[name = tensor<string, []>("x_9_cast_fp16")]; | |
| tensor<fp16, [50, 768]> const_1_to_fp16 = const()[name = tensor<string, []>("const_1_to_fp16"), val = tensor<fp16, [50, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4720320)))]; | |
| tensor<fp16, [?, 50, 768]> x_11_cast_fp16 = add(x = x_9_cast_fp16, y = const_1_to_fp16)[name = tensor<string, []>("x_11_cast_fp16")]; | |
| tensor<int32, [1]> ret_1_axes_0 = const()[name = tensor<string, []>("ret_1_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> ln_pre_weight_to_fp16 = const()[name = tensor<string, []>("ln_pre_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4797184)))]; | |
| tensor<fp16, [768]> ln_pre_bias_to_fp16 = const()[name = tensor<string, []>("ln_pre_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4798784)))]; | |
| tensor<fp16, []> var_70_to_fp16 = const()[name = tensor<string, []>("op_70_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [?, 50, 768]> ret_1_cast_fp16 = layer_norm(axes = ret_1_axes_0, beta = ln_pre_bias_to_fp16, epsilon = var_70_to_fp16, gamma = ln_pre_weight_to_fp16, x = x_11_cast_fp16)[name = tensor<string, []>("ret_1_cast_fp16")]; | |
| tensor<int32, [3]> var_84 = const()[name = tensor<string, []>("op_84"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [1]> ret_3_axes_0 = const()[name = tensor<string, []>("ret_3_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_0_ln_1_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_0_ln_1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4800384)))]; | |
| tensor<fp16, [768]> transformer_resblocks_0_ln_1_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_0_ln_1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4801984)))]; | |
| tensor<fp16, []> var_100_to_fp16 = const()[name = tensor<string, []>("op_100_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [50, ?, 768]> x_15_cast_fp16 = transpose(perm = var_84, x = ret_1_cast_fp16)[name = tensor<string, []>("transpose_62")]; | |
| tensor<fp16, [50, ?, 768]> ret_3_cast_fp16 = layer_norm(axes = ret_3_axes_0, beta = transformer_resblocks_0_ln_1_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_0_ln_1_weight_to_fp16, x = x_15_cast_fp16)[name = tensor<string, []>("ret_3_cast_fp16")]; | |
| tensor<fp16, [2304, 768]> transformer_resblocks_0_attn_in_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_0_attn_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4803584)))]; | |
| tensor<fp16, [2304]> transformer_resblocks_0_attn_in_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_0_attn_in_proj_bias_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8342592)))]; | |
| tensor<fp16, [50, ?, 2304]> linear_0_cast_fp16 = linear(bias = transformer_resblocks_0_attn_in_proj_bias_to_fp16, weight = transformer_resblocks_0_attn_in_proj_weight_to_fp16, x = ret_3_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")]; | |
| tensor<int32, [4]> concat_2x = const()[name = tensor<string, []>("concat_2x"), val = tensor<int32, [4]>([50, -1, 3, 768])]; | |
| tensor<fp16, [50, ?, 3, 768]> var_150_cast_fp16 = reshape(shape = concat_2x, x = linear_0_cast_fp16)[name = tensor<string, []>("op_150_cast_fp16")]; | |
| tensor<int32, [1]> var_151_axes_0 = const()[name = tensor<string, []>("op_151_axes_0"), val = tensor<int32, [1]>([0])]; | |
| tensor<fp16, [1, 50, ?, 3, 768]> var_151_cast_fp16 = expand_dims(axes = var_151_axes_0, x = var_150_cast_fp16)[name = tensor<string, []>("op_151_cast_fp16")]; | |
| tensor<int32, [5]> var_152_perm_0 = const()[name = tensor<string, []>("op_152_perm_0"), val = tensor<int32, [5]>([-2, 1, 2, 0, 4])]; | |
| tensor<int32, [1]> var_153_axes_0 = const()[name = tensor<string, []>("op_153_axes_0"), val = tensor<int32, [1]>([-2])]; | |
| tensor<fp16, [3, 50, ?, 1, 768]> var_152_cast_fp16 = transpose(perm = var_152_perm_0, x = var_151_cast_fp16)[name = tensor<string, []>("transpose_61")]; | |
| tensor<fp16, [3, 50, ?, 768]> var_153_cast_fp16 = squeeze(axes = var_153_axes_0, x = var_152_cast_fp16)[name = tensor<string, []>("op_153_cast_fp16")]; | |
| tensor<int32, [4]> q_1_begin_0 = const()[name = tensor<string, []>("q_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> q_1_end_0 = const()[name = tensor<string, []>("q_1_end_0"), val = tensor<int32, [4]>([1, 50, 0, 768])]; | |
| tensor<bool, [4]> q_1_end_mask_0 = const()[name = tensor<string, []>("q_1_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> q_1_squeeze_mask_0 = const()[name = tensor<string, []>("q_1_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> q_1_cast_fp16 = slice_by_index(begin = q_1_begin_0, end = q_1_end_0, end_mask = q_1_end_mask_0, squeeze_mask = q_1_squeeze_mask_0, x = var_153_cast_fp16)[name = tensor<string, []>("q_1_cast_fp16")]; | |
| tensor<int32, [4]> k_1_begin_0 = const()[name = tensor<string, []>("k_1_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])]; | |
| tensor<int32, [4]> k_1_end_0 = const()[name = tensor<string, []>("k_1_end_0"), val = tensor<int32, [4]>([2, 50, 0, 768])]; | |
| tensor<bool, [4]> k_1_end_mask_0 = const()[name = tensor<string, []>("k_1_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> k_1_squeeze_mask_0 = const()[name = tensor<string, []>("k_1_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> k_1_cast_fp16 = slice_by_index(begin = k_1_begin_0, end = k_1_end_0, end_mask = k_1_end_mask_0, squeeze_mask = k_1_squeeze_mask_0, x = var_153_cast_fp16)[name = tensor<string, []>("k_1_cast_fp16")]; | |
| tensor<int32, [4]> v_1_begin_0 = const()[name = tensor<string, []>("v_1_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])]; | |
| tensor<int32, [4]> v_1_end_0 = const()[name = tensor<string, []>("v_1_end_0"), val = tensor<int32, [4]>([3, 50, 0, 768])]; | |
| tensor<bool, [4]> v_1_end_mask_0 = const()[name = tensor<string, []>("v_1_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> v_1_squeeze_mask_0 = const()[name = tensor<string, []>("v_1_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> v_1_cast_fp16 = slice_by_index(begin = v_1_begin_0, end = v_1_end_0, end_mask = v_1_end_mask_0, squeeze_mask = v_1_squeeze_mask_0, x = var_153_cast_fp16)[name = tensor<string, []>("v_1_cast_fp16")]; | |
| tensor<int32, [3]> concat_3x = const()[name = tensor<string, []>("concat_3x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_162_cast_fp16 = reshape(shape = concat_3x, x = q_1_cast_fp16)[name = tensor<string, []>("op_162_cast_fp16")]; | |
| tensor<int32, [3]> q_3_perm_0 = const()[name = tensor<string, []>("q_3_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_4x = const()[name = tensor<string, []>("concat_4x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_169_cast_fp16 = reshape(shape = concat_4x, x = k_1_cast_fp16)[name = tensor<string, []>("op_169_cast_fp16")]; | |
| tensor<int32, [3]> k_3_perm_0 = const()[name = tensor<string, []>("k_3_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_5x = const()[name = tensor<string, []>("concat_5x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_176_cast_fp16 = reshape(shape = concat_5x, x = v_1_cast_fp16)[name = tensor<string, []>("op_176_cast_fp16")]; | |
| tensor<int32, [3]> v_3_perm_0 = const()[name = tensor<string, []>("v_3_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [4]> concat_6x = const()[name = tensor<string, []>("concat_6x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> q_3_cast_fp16 = transpose(perm = q_3_perm_0, x = var_162_cast_fp16)[name = tensor<string, []>("transpose_60")]; | |
| tensor<fp16, [?, 12, 50, 64]> q_5_cast_fp16 = reshape(shape = concat_6x, x = q_3_cast_fp16)[name = tensor<string, []>("q_5_cast_fp16")]; | |
| tensor<int32, [4]> concat_7x = const()[name = tensor<string, []>("concat_7x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = var_169_cast_fp16)[name = tensor<string, []>("transpose_59")]; | |
| tensor<fp16, [?, 12, 50, 64]> k_5_cast_fp16 = reshape(shape = concat_7x, x = k_3_cast_fp16)[name = tensor<string, []>("k_5_cast_fp16")]; | |
| tensor<int32, [4]> concat_8x = const()[name = tensor<string, []>("concat_8x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> v_3_cast_fp16 = transpose(perm = v_3_perm_0, x = var_176_cast_fp16)[name = tensor<string, []>("transpose_58")]; | |
| tensor<fp16, [?, 12, 50, 64]> v_5_cast_fp16 = reshape(shape = concat_8x, x = v_3_cast_fp16)[name = tensor<string, []>("v_5_cast_fp16")]; | |
| tensor<fp16, []> mul_1_y_0_to_fp16 = const()[name = tensor<string, []>("mul_1_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [?, 12, 50, 64]> mul_1_cast_fp16 = mul(x = q_5_cast_fp16, y = mul_1_y_0_to_fp16)[name = tensor<string, []>("mul_1_cast_fp16")]; | |
| tensor<bool, []> matmul_0_transpose_y_0 = const()[name = tensor<string, []>("matmul_0_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> matmul_0_transpose_x_0 = const()[name = tensor<string, []>("matmul_0_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 50]> matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = mul_1_cast_fp16, y = k_5_cast_fp16)[name = tensor<string, []>("matmul_0_cast_fp16")]; | |
| tensor<int32, []> softmax_0_axis_0 = const()[name = tensor<string, []>("softmax_0_axis_0"), val = tensor<int32, []>(-1)]; | |
| tensor<fp16, [?, 12, 50, 50]> softmax_0_cast_fp16 = softmax(axis = softmax_0_axis_0, x = matmul_0_cast_fp16)[name = tensor<string, []>("softmax_0_cast_fp16")]; | |
| tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 64]> attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = softmax_0_cast_fp16, y = v_5_cast_fp16)[name = tensor<string, []>("attn_output_1_cast_fp16")]; | |
| tensor<int32, [4]> var_186 = const()[name = tensor<string, []>("op_186"), val = tensor<int32, [4]>([2, 0, 1, 3])]; | |
| tensor<int32, [2]> concat_9x = const()[name = tensor<string, []>("concat_9x"), val = tensor<int32, [2]>([-1, 768])]; | |
| tensor<fp16, [50, ?, 12, 64]> var_187_cast_fp16 = transpose(perm = var_186, x = attn_output_1_cast_fp16)[name = tensor<string, []>("transpose_57")]; | |
| tensor<fp16, [?, 768]> attn_output_3_cast_fp16 = reshape(shape = concat_9x, x = var_187_cast_fp16)[name = tensor<string, []>("attn_output_3_cast_fp16")]; | |
| tensor<fp16, [768, 768]> transformer_resblocks_0_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_0_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8347264)))]; | |
| tensor<fp16, [768]> transformer_resblocks_0_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_0_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9526976)))]; | |
| tensor<fp16, [?, 768]> linear_1_cast_fp16 = linear(bias = transformer_resblocks_0_attn_out_proj_bias_to_fp16, weight = transformer_resblocks_0_attn_out_proj_weight_to_fp16, x = attn_output_3_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")]; | |
| tensor<int32, [3]> concat_10x = const()[name = tensor<string, []>("concat_10x"), val = tensor<int32, [3]>([50, -1, 768])]; | |
| tensor<fp16, [50, ?, 768]> var_196_cast_fp16 = reshape(shape = concat_10x, x = linear_1_cast_fp16)[name = tensor<string, []>("op_196_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_17_cast_fp16 = add(x = x_15_cast_fp16, y = var_196_cast_fp16)[name = tensor<string, []>("x_17_cast_fp16")]; | |
| tensor<int32, [1]> ret_5_axes_0 = const()[name = tensor<string, []>("ret_5_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_0_ln_2_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_0_ln_2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9528576)))]; | |
| tensor<fp16, [768]> transformer_resblocks_0_ln_2_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_0_ln_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9530176)))]; | |
| tensor<fp16, [50, ?, 768]> ret_5_cast_fp16 = layer_norm(axes = ret_5_axes_0, beta = transformer_resblocks_0_ln_2_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_0_ln_2_weight_to_fp16, x = x_17_cast_fp16)[name = tensor<string, []>("ret_5_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> transformer_resblocks_0_mlp_c_fc_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_0_mlp_c_fc_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9531776)))]; | |
| tensor<fp16, [3072]> transformer_resblocks_0_mlp_c_fc_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_0_mlp_c_fc_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14250432)))]; | |
| tensor<fp16, [50, ?, 3072]> linear_2_cast_fp16 = linear(bias = transformer_resblocks_0_mlp_c_fc_bias_to_fp16, weight = transformer_resblocks_0_mlp_c_fc_weight_to_fp16, x = ret_5_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")]; | |
| tensor<fp16, []> var_212_to_fp16 = const()[name = tensor<string, []>("op_212_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [50, ?, 3072]> var_213_cast_fp16 = mul(x = linear_2_cast_fp16, y = var_212_to_fp16)[name = tensor<string, []>("op_213_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> var_214_cast_fp16 = sigmoid(x = var_213_cast_fp16)[name = tensor<string, []>("op_214_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> input_9_cast_fp16 = mul(x = linear_2_cast_fp16, y = var_214_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> transformer_resblocks_0_mlp_c_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_0_mlp_c_proj_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14256640)))]; | |
| tensor<fp16, [768]> transformer_resblocks_0_mlp_c_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_0_mlp_c_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18975296)))]; | |
| tensor<fp16, [50, ?, 768]> linear_3_cast_fp16 = linear(bias = transformer_resblocks_0_mlp_c_proj_bias_to_fp16, weight = transformer_resblocks_0_mlp_c_proj_weight_to_fp16, x = input_9_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_21_cast_fp16 = add(x = x_17_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("x_21_cast_fp16")]; | |
| tensor<int32, [1]> ret_7_axes_0 = const()[name = tensor<string, []>("ret_7_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_1_ln_1_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_1_ln_1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18976896)))]; | |
| tensor<fp16, [768]> transformer_resblocks_1_ln_1_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_1_ln_1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18978496)))]; | |
| tensor<fp16, [50, ?, 768]> ret_7_cast_fp16 = layer_norm(axes = ret_7_axes_0, beta = transformer_resblocks_1_ln_1_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_1_ln_1_weight_to_fp16, x = x_21_cast_fp16)[name = tensor<string, []>("ret_7_cast_fp16")]; | |
| tensor<fp16, [2304, 768]> transformer_resblocks_1_attn_in_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_1_attn_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18980096)))]; | |
| tensor<fp16, [2304]> transformer_resblocks_1_attn_in_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_1_attn_in_proj_bias_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22519104)))]; | |
| tensor<fp16, [50, ?, 2304]> linear_4_cast_fp16 = linear(bias = transformer_resblocks_1_attn_in_proj_bias_to_fp16, weight = transformer_resblocks_1_attn_in_proj_weight_to_fp16, x = ret_7_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")]; | |
| tensor<int32, [4]> concat_11x = const()[name = tensor<string, []>("concat_11x"), val = tensor<int32, [4]>([50, -1, 3, 768])]; | |
| tensor<fp16, [50, ?, 3, 768]> var_255_cast_fp16 = reshape(shape = concat_11x, x = linear_4_cast_fp16)[name = tensor<string, []>("op_255_cast_fp16")]; | |
| tensor<int32, [1]> var_256_axes_0 = const()[name = tensor<string, []>("op_256_axes_0"), val = tensor<int32, [1]>([0])]; | |
| tensor<fp16, [1, 50, ?, 3, 768]> var_256_cast_fp16 = expand_dims(axes = var_256_axes_0, x = var_255_cast_fp16)[name = tensor<string, []>("op_256_cast_fp16")]; | |
| tensor<int32, [5]> var_257_perm_0 = const()[name = tensor<string, []>("op_257_perm_0"), val = tensor<int32, [5]>([-2, 1, 2, 0, 4])]; | |
| tensor<int32, [1]> var_258_axes_0 = const()[name = tensor<string, []>("op_258_axes_0"), val = tensor<int32, [1]>([-2])]; | |
| tensor<fp16, [3, 50, ?, 1, 768]> var_257_cast_fp16 = transpose(perm = var_257_perm_0, x = var_256_cast_fp16)[name = tensor<string, []>("transpose_56")]; | |
| tensor<fp16, [3, 50, ?, 768]> var_258_cast_fp16 = squeeze(axes = var_258_axes_0, x = var_257_cast_fp16)[name = tensor<string, []>("op_258_cast_fp16")]; | |
| tensor<int32, [4]> q_7_begin_0 = const()[name = tensor<string, []>("q_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> q_7_end_0 = const()[name = tensor<string, []>("q_7_end_0"), val = tensor<int32, [4]>([1, 50, 0, 768])]; | |
| tensor<bool, [4]> q_7_end_mask_0 = const()[name = tensor<string, []>("q_7_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> q_7_squeeze_mask_0 = const()[name = tensor<string, []>("q_7_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> q_7_cast_fp16 = slice_by_index(begin = q_7_begin_0, end = q_7_end_0, end_mask = q_7_end_mask_0, squeeze_mask = q_7_squeeze_mask_0, x = var_258_cast_fp16)[name = tensor<string, []>("q_7_cast_fp16")]; | |
| tensor<int32, [4]> k_7_begin_0 = const()[name = tensor<string, []>("k_7_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])]; | |
| tensor<int32, [4]> k_7_end_0 = const()[name = tensor<string, []>("k_7_end_0"), val = tensor<int32, [4]>([2, 50, 0, 768])]; | |
| tensor<bool, [4]> k_7_end_mask_0 = const()[name = tensor<string, []>("k_7_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> k_7_squeeze_mask_0 = const()[name = tensor<string, []>("k_7_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> k_7_cast_fp16 = slice_by_index(begin = k_7_begin_0, end = k_7_end_0, end_mask = k_7_end_mask_0, squeeze_mask = k_7_squeeze_mask_0, x = var_258_cast_fp16)[name = tensor<string, []>("k_7_cast_fp16")]; | |
| tensor<int32, [4]> v_7_begin_0 = const()[name = tensor<string, []>("v_7_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])]; | |
| tensor<int32, [4]> v_7_end_0 = const()[name = tensor<string, []>("v_7_end_0"), val = tensor<int32, [4]>([3, 50, 0, 768])]; | |
| tensor<bool, [4]> v_7_end_mask_0 = const()[name = tensor<string, []>("v_7_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> v_7_squeeze_mask_0 = const()[name = tensor<string, []>("v_7_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> v_7_cast_fp16 = slice_by_index(begin = v_7_begin_0, end = v_7_end_0, end_mask = v_7_end_mask_0, squeeze_mask = v_7_squeeze_mask_0, x = var_258_cast_fp16)[name = tensor<string, []>("v_7_cast_fp16")]; | |
| tensor<int32, [3]> concat_12x = const()[name = tensor<string, []>("concat_12x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_267_cast_fp16 = reshape(shape = concat_12x, x = q_7_cast_fp16)[name = tensor<string, []>("op_267_cast_fp16")]; | |
| tensor<int32, [3]> q_9_perm_0 = const()[name = tensor<string, []>("q_9_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_13x = const()[name = tensor<string, []>("concat_13x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_274_cast_fp16 = reshape(shape = concat_13x, x = k_7_cast_fp16)[name = tensor<string, []>("op_274_cast_fp16")]; | |
| tensor<int32, [3]> k_9_perm_0 = const()[name = tensor<string, []>("k_9_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_14x = const()[name = tensor<string, []>("concat_14x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_281_cast_fp16 = reshape(shape = concat_14x, x = v_7_cast_fp16)[name = tensor<string, []>("op_281_cast_fp16")]; | |
| tensor<int32, [3]> v_9_perm_0 = const()[name = tensor<string, []>("v_9_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [4]> concat_15x = const()[name = tensor<string, []>("concat_15x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> q_9_cast_fp16 = transpose(perm = q_9_perm_0, x = var_267_cast_fp16)[name = tensor<string, []>("transpose_55")]; | |
| tensor<fp16, [?, 12, 50, 64]> q_11_cast_fp16 = reshape(shape = concat_15x, x = q_9_cast_fp16)[name = tensor<string, []>("q_11_cast_fp16")]; | |
| tensor<int32, [4]> concat_16x = const()[name = tensor<string, []>("concat_16x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> k_9_cast_fp16 = transpose(perm = k_9_perm_0, x = var_274_cast_fp16)[name = tensor<string, []>("transpose_54")]; | |
| tensor<fp16, [?, 12, 50, 64]> k_11_cast_fp16 = reshape(shape = concat_16x, x = k_9_cast_fp16)[name = tensor<string, []>("k_11_cast_fp16")]; | |
| tensor<int32, [4]> concat_17x = const()[name = tensor<string, []>("concat_17x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> v_9_cast_fp16 = transpose(perm = v_9_perm_0, x = var_281_cast_fp16)[name = tensor<string, []>("transpose_53")]; | |
| tensor<fp16, [?, 12, 50, 64]> v_11_cast_fp16 = reshape(shape = concat_17x, x = v_9_cast_fp16)[name = tensor<string, []>("v_11_cast_fp16")]; | |
| tensor<fp16, []> mul_3_y_0_to_fp16 = const()[name = tensor<string, []>("mul_3_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [?, 12, 50, 64]> mul_3_cast_fp16 = mul(x = q_11_cast_fp16, y = mul_3_y_0_to_fp16)[name = tensor<string, []>("mul_3_cast_fp16")]; | |
| tensor<bool, []> matmul_1_transpose_y_0 = const()[name = tensor<string, []>("matmul_1_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> matmul_1_transpose_x_0 = const()[name = tensor<string, []>("matmul_1_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 50]> matmul_1_cast_fp16 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = mul_3_cast_fp16, y = k_11_cast_fp16)[name = tensor<string, []>("matmul_1_cast_fp16")]; | |
| tensor<int32, []> softmax_1_axis_0 = const()[name = tensor<string, []>("softmax_1_axis_0"), val = tensor<int32, []>(-1)]; | |
| tensor<fp16, [?, 12, 50, 50]> softmax_1_cast_fp16 = softmax(axis = softmax_1_axis_0, x = matmul_1_cast_fp16)[name = tensor<string, []>("softmax_1_cast_fp16")]; | |
| tensor<bool, []> attn_output_7_transpose_x_0 = const()[name = tensor<string, []>("attn_output_7_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_7_transpose_y_0 = const()[name = tensor<string, []>("attn_output_7_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 64]> attn_output_7_cast_fp16 = matmul(transpose_x = attn_output_7_transpose_x_0, transpose_y = attn_output_7_transpose_y_0, x = softmax_1_cast_fp16, y = v_11_cast_fp16)[name = tensor<string, []>("attn_output_7_cast_fp16")]; | |
| tensor<int32, [4]> var_291 = const()[name = tensor<string, []>("op_291"), val = tensor<int32, [4]>([2, 0, 1, 3])]; | |
| tensor<int32, [2]> concat_18x = const()[name = tensor<string, []>("concat_18x"), val = tensor<int32, [2]>([-1, 768])]; | |
| tensor<fp16, [50, ?, 12, 64]> var_292_cast_fp16 = transpose(perm = var_291, x = attn_output_7_cast_fp16)[name = tensor<string, []>("transpose_52")]; | |
| tensor<fp16, [?, 768]> attn_output_9_cast_fp16 = reshape(shape = concat_18x, x = var_292_cast_fp16)[name = tensor<string, []>("attn_output_9_cast_fp16")]; | |
| tensor<fp16, [768, 768]> transformer_resblocks_1_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_1_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22523776)))]; | |
| tensor<fp16, [768]> transformer_resblocks_1_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_1_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23703488)))]; | |
| tensor<fp16, [?, 768]> linear_5_cast_fp16 = linear(bias = transformer_resblocks_1_attn_out_proj_bias_to_fp16, weight = transformer_resblocks_1_attn_out_proj_weight_to_fp16, x = attn_output_9_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")]; | |
| tensor<int32, [3]> concat_19x = const()[name = tensor<string, []>("concat_19x"), val = tensor<int32, [3]>([50, -1, 768])]; | |
| tensor<fp16, [50, ?, 768]> var_301_cast_fp16 = reshape(shape = concat_19x, x = linear_5_cast_fp16)[name = tensor<string, []>("op_301_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_23_cast_fp16 = add(x = x_21_cast_fp16, y = var_301_cast_fp16)[name = tensor<string, []>("x_23_cast_fp16")]; | |
| tensor<int32, [1]> ret_9_axes_0 = const()[name = tensor<string, []>("ret_9_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_1_ln_2_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_1_ln_2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23705088)))]; | |
| tensor<fp16, [768]> transformer_resblocks_1_ln_2_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_1_ln_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23706688)))]; | |
| tensor<fp16, [50, ?, 768]> ret_9_cast_fp16 = layer_norm(axes = ret_9_axes_0, beta = transformer_resblocks_1_ln_2_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_1_ln_2_weight_to_fp16, x = x_23_cast_fp16)[name = tensor<string, []>("ret_9_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> transformer_resblocks_1_mlp_c_fc_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_1_mlp_c_fc_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23708288)))]; | |
| tensor<fp16, [3072]> transformer_resblocks_1_mlp_c_fc_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_1_mlp_c_fc_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28426944)))]; | |
| tensor<fp16, [50, ?, 3072]> linear_6_cast_fp16 = linear(bias = transformer_resblocks_1_mlp_c_fc_bias_to_fp16, weight = transformer_resblocks_1_mlp_c_fc_weight_to_fp16, x = ret_9_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")]; | |
| tensor<fp16, []> var_317_to_fp16 = const()[name = tensor<string, []>("op_317_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [50, ?, 3072]> var_318_cast_fp16 = mul(x = linear_6_cast_fp16, y = var_317_to_fp16)[name = tensor<string, []>("op_318_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> var_319_cast_fp16 = sigmoid(x = var_318_cast_fp16)[name = tensor<string, []>("op_319_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> input_17_cast_fp16 = mul(x = linear_6_cast_fp16, y = var_319_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> transformer_resblocks_1_mlp_c_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_1_mlp_c_proj_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28433152)))]; | |
| tensor<fp16, [768]> transformer_resblocks_1_mlp_c_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_1_mlp_c_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33151808)))]; | |
| tensor<fp16, [50, ?, 768]> linear_7_cast_fp16 = linear(bias = transformer_resblocks_1_mlp_c_proj_bias_to_fp16, weight = transformer_resblocks_1_mlp_c_proj_weight_to_fp16, x = input_17_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_27_cast_fp16 = add(x = x_23_cast_fp16, y = linear_7_cast_fp16)[name = tensor<string, []>("x_27_cast_fp16")]; | |
| tensor<int32, [1]> ret_11_axes_0 = const()[name = tensor<string, []>("ret_11_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_2_ln_1_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_2_ln_1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33153408)))]; | |
| tensor<fp16, [768]> transformer_resblocks_2_ln_1_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_2_ln_1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33155008)))]; | |
| tensor<fp16, [50, ?, 768]> ret_11_cast_fp16 = layer_norm(axes = ret_11_axes_0, beta = transformer_resblocks_2_ln_1_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_2_ln_1_weight_to_fp16, x = x_27_cast_fp16)[name = tensor<string, []>("ret_11_cast_fp16")]; | |
| tensor<fp16, [2304, 768]> transformer_resblocks_2_attn_in_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_2_attn_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33156608)))]; | |
| tensor<fp16, [2304]> transformer_resblocks_2_attn_in_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_2_attn_in_proj_bias_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36695616)))]; | |
| tensor<fp16, [50, ?, 2304]> linear_8_cast_fp16 = linear(bias = transformer_resblocks_2_attn_in_proj_bias_to_fp16, weight = transformer_resblocks_2_attn_in_proj_weight_to_fp16, x = ret_11_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")]; | |
| tensor<int32, [4]> concat_20x = const()[name = tensor<string, []>("concat_20x"), val = tensor<int32, [4]>([50, -1, 3, 768])]; | |
| tensor<fp16, [50, ?, 3, 768]> var_360_cast_fp16 = reshape(shape = concat_20x, x = linear_8_cast_fp16)[name = tensor<string, []>("op_360_cast_fp16")]; | |
| tensor<int32, [1]> var_361_axes_0 = const()[name = tensor<string, []>("op_361_axes_0"), val = tensor<int32, [1]>([0])]; | |
| tensor<fp16, [1, 50, ?, 3, 768]> var_361_cast_fp16 = expand_dims(axes = var_361_axes_0, x = var_360_cast_fp16)[name = tensor<string, []>("op_361_cast_fp16")]; | |
| tensor<int32, [5]> var_362_perm_0 = const()[name = tensor<string, []>("op_362_perm_0"), val = tensor<int32, [5]>([-2, 1, 2, 0, 4])]; | |
| tensor<int32, [1]> var_363_axes_0 = const()[name = tensor<string, []>("op_363_axes_0"), val = tensor<int32, [1]>([-2])]; | |
| tensor<fp16, [3, 50, ?, 1, 768]> var_362_cast_fp16 = transpose(perm = var_362_perm_0, x = var_361_cast_fp16)[name = tensor<string, []>("transpose_51")]; | |
| tensor<fp16, [3, 50, ?, 768]> var_363_cast_fp16 = squeeze(axes = var_363_axes_0, x = var_362_cast_fp16)[name = tensor<string, []>("op_363_cast_fp16")]; | |
| tensor<int32, [4]> q_13_begin_0 = const()[name = tensor<string, []>("q_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> q_13_end_0 = const()[name = tensor<string, []>("q_13_end_0"), val = tensor<int32, [4]>([1, 50, 0, 768])]; | |
| tensor<bool, [4]> q_13_end_mask_0 = const()[name = tensor<string, []>("q_13_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> q_13_squeeze_mask_0 = const()[name = tensor<string, []>("q_13_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> q_13_cast_fp16 = slice_by_index(begin = q_13_begin_0, end = q_13_end_0, end_mask = q_13_end_mask_0, squeeze_mask = q_13_squeeze_mask_0, x = var_363_cast_fp16)[name = tensor<string, []>("q_13_cast_fp16")]; | |
| tensor<int32, [4]> k_13_begin_0 = const()[name = tensor<string, []>("k_13_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])]; | |
| tensor<int32, [4]> k_13_end_0 = const()[name = tensor<string, []>("k_13_end_0"), val = tensor<int32, [4]>([2, 50, 0, 768])]; | |
| tensor<bool, [4]> k_13_end_mask_0 = const()[name = tensor<string, []>("k_13_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> k_13_squeeze_mask_0 = const()[name = tensor<string, []>("k_13_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> k_13_cast_fp16 = slice_by_index(begin = k_13_begin_0, end = k_13_end_0, end_mask = k_13_end_mask_0, squeeze_mask = k_13_squeeze_mask_0, x = var_363_cast_fp16)[name = tensor<string, []>("k_13_cast_fp16")]; | |
| tensor<int32, [4]> v_13_begin_0 = const()[name = tensor<string, []>("v_13_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])]; | |
| tensor<int32, [4]> v_13_end_0 = const()[name = tensor<string, []>("v_13_end_0"), val = tensor<int32, [4]>([3, 50, 0, 768])]; | |
| tensor<bool, [4]> v_13_end_mask_0 = const()[name = tensor<string, []>("v_13_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> v_13_squeeze_mask_0 = const()[name = tensor<string, []>("v_13_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> v_13_cast_fp16 = slice_by_index(begin = v_13_begin_0, end = v_13_end_0, end_mask = v_13_end_mask_0, squeeze_mask = v_13_squeeze_mask_0, x = var_363_cast_fp16)[name = tensor<string, []>("v_13_cast_fp16")]; | |
| tensor<int32, [3]> concat_21x = const()[name = tensor<string, []>("concat_21x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_372_cast_fp16 = reshape(shape = concat_21x, x = q_13_cast_fp16)[name = tensor<string, []>("op_372_cast_fp16")]; | |
| tensor<int32, [3]> q_15_perm_0 = const()[name = tensor<string, []>("q_15_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_22x = const()[name = tensor<string, []>("concat_22x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_379_cast_fp16 = reshape(shape = concat_22x, x = k_13_cast_fp16)[name = tensor<string, []>("op_379_cast_fp16")]; | |
| tensor<int32, [3]> k_15_perm_0 = const()[name = tensor<string, []>("k_15_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_23x = const()[name = tensor<string, []>("concat_23x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_386_cast_fp16 = reshape(shape = concat_23x, x = v_13_cast_fp16)[name = tensor<string, []>("op_386_cast_fp16")]; | |
| tensor<int32, [3]> v_15_perm_0 = const()[name = tensor<string, []>("v_15_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [4]> concat_24x = const()[name = tensor<string, []>("concat_24x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> q_15_cast_fp16 = transpose(perm = q_15_perm_0, x = var_372_cast_fp16)[name = tensor<string, []>("transpose_50")]; | |
| tensor<fp16, [?, 12, 50, 64]> q_17_cast_fp16 = reshape(shape = concat_24x, x = q_15_cast_fp16)[name = tensor<string, []>("q_17_cast_fp16")]; | |
| tensor<int32, [4]> concat_25x = const()[name = tensor<string, []>("concat_25x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> k_15_cast_fp16 = transpose(perm = k_15_perm_0, x = var_379_cast_fp16)[name = tensor<string, []>("transpose_49")]; | |
| tensor<fp16, [?, 12, 50, 64]> k_17_cast_fp16 = reshape(shape = concat_25x, x = k_15_cast_fp16)[name = tensor<string, []>("k_17_cast_fp16")]; | |
| tensor<int32, [4]> concat_26x = const()[name = tensor<string, []>("concat_26x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> v_15_cast_fp16 = transpose(perm = v_15_perm_0, x = var_386_cast_fp16)[name = tensor<string, []>("transpose_48")]; | |
| tensor<fp16, [?, 12, 50, 64]> v_17_cast_fp16 = reshape(shape = concat_26x, x = v_15_cast_fp16)[name = tensor<string, []>("v_17_cast_fp16")]; | |
| tensor<fp16, []> mul_5_y_0_to_fp16 = const()[name = tensor<string, []>("mul_5_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [?, 12, 50, 64]> mul_5_cast_fp16 = mul(x = q_17_cast_fp16, y = mul_5_y_0_to_fp16)[name = tensor<string, []>("mul_5_cast_fp16")]; | |
| tensor<bool, []> matmul_2_transpose_y_0 = const()[name = tensor<string, []>("matmul_2_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> matmul_2_transpose_x_0 = const()[name = tensor<string, []>("matmul_2_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 50]> matmul_2_cast_fp16 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = mul_5_cast_fp16, y = k_17_cast_fp16)[name = tensor<string, []>("matmul_2_cast_fp16")]; | |
| tensor<int32, []> softmax_2_axis_0 = const()[name = tensor<string, []>("softmax_2_axis_0"), val = tensor<int32, []>(-1)]; | |
| tensor<fp16, [?, 12, 50, 50]> softmax_2_cast_fp16 = softmax(axis = softmax_2_axis_0, x = matmul_2_cast_fp16)[name = tensor<string, []>("softmax_2_cast_fp16")]; | |
| tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 64]> attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = softmax_2_cast_fp16, y = v_17_cast_fp16)[name = tensor<string, []>("attn_output_13_cast_fp16")]; | |
| tensor<int32, [4]> var_396 = const()[name = tensor<string, []>("op_396"), val = tensor<int32, [4]>([2, 0, 1, 3])]; | |
| tensor<int32, [2]> concat_27x = const()[name = tensor<string, []>("concat_27x"), val = tensor<int32, [2]>([-1, 768])]; | |
| tensor<fp16, [50, ?, 12, 64]> var_397_cast_fp16 = transpose(perm = var_396, x = attn_output_13_cast_fp16)[name = tensor<string, []>("transpose_47")]; | |
| tensor<fp16, [?, 768]> attn_output_15_cast_fp16 = reshape(shape = concat_27x, x = var_397_cast_fp16)[name = tensor<string, []>("attn_output_15_cast_fp16")]; | |
| tensor<fp16, [768, 768]> transformer_resblocks_2_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_2_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36700288)))]; | |
| tensor<fp16, [768]> transformer_resblocks_2_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_2_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37880000)))]; | |
| tensor<fp16, [?, 768]> linear_9_cast_fp16 = linear(bias = transformer_resblocks_2_attn_out_proj_bias_to_fp16, weight = transformer_resblocks_2_attn_out_proj_weight_to_fp16, x = attn_output_15_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")]; | |
| tensor<int32, [3]> concat_28x = const()[name = tensor<string, []>("concat_28x"), val = tensor<int32, [3]>([50, -1, 768])]; | |
| tensor<fp16, [50, ?, 768]> var_406_cast_fp16 = reshape(shape = concat_28x, x = linear_9_cast_fp16)[name = tensor<string, []>("op_406_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_29_cast_fp16 = add(x = x_27_cast_fp16, y = var_406_cast_fp16)[name = tensor<string, []>("x_29_cast_fp16")]; | |
| tensor<int32, [1]> ret_13_axes_0 = const()[name = tensor<string, []>("ret_13_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_2_ln_2_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_2_ln_2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37881600)))]; | |
| tensor<fp16, [768]> transformer_resblocks_2_ln_2_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_2_ln_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37883200)))]; | |
| tensor<fp16, [50, ?, 768]> ret_13_cast_fp16 = layer_norm(axes = ret_13_axes_0, beta = transformer_resblocks_2_ln_2_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_2_ln_2_weight_to_fp16, x = x_29_cast_fp16)[name = tensor<string, []>("ret_13_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> transformer_resblocks_2_mlp_c_fc_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_2_mlp_c_fc_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37884800)))]; | |
| tensor<fp16, [3072]> transformer_resblocks_2_mlp_c_fc_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_2_mlp_c_fc_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42603456)))]; | |
| tensor<fp16, [50, ?, 3072]> linear_10_cast_fp16 = linear(bias = transformer_resblocks_2_mlp_c_fc_bias_to_fp16, weight = transformer_resblocks_2_mlp_c_fc_weight_to_fp16, x = ret_13_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")]; | |
| tensor<fp16, []> var_422_to_fp16 = const()[name = tensor<string, []>("op_422_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [50, ?, 3072]> var_423_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_422_to_fp16)[name = tensor<string, []>("op_423_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> var_424_cast_fp16 = sigmoid(x = var_423_cast_fp16)[name = tensor<string, []>("op_424_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> input_25_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_424_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> transformer_resblocks_2_mlp_c_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_2_mlp_c_proj_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42609664)))]; | |
| tensor<fp16, [768]> transformer_resblocks_2_mlp_c_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_2_mlp_c_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47328320)))]; | |
| tensor<fp16, [50, ?, 768]> linear_11_cast_fp16 = linear(bias = transformer_resblocks_2_mlp_c_proj_bias_to_fp16, weight = transformer_resblocks_2_mlp_c_proj_weight_to_fp16, x = input_25_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_33_cast_fp16 = add(x = x_29_cast_fp16, y = linear_11_cast_fp16)[name = tensor<string, []>("x_33_cast_fp16")]; | |
| tensor<int32, [1]> ret_15_axes_0 = const()[name = tensor<string, []>("ret_15_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_3_ln_1_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_3_ln_1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47329920)))]; | |
| tensor<fp16, [768]> transformer_resblocks_3_ln_1_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_3_ln_1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47331520)))]; | |
| tensor<fp16, [50, ?, 768]> ret_15_cast_fp16 = layer_norm(axes = ret_15_axes_0, beta = transformer_resblocks_3_ln_1_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_3_ln_1_weight_to_fp16, x = x_33_cast_fp16)[name = tensor<string, []>("ret_15_cast_fp16")]; | |
| tensor<fp16, [2304, 768]> transformer_resblocks_3_attn_in_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_3_attn_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47333120)))]; | |
| tensor<fp16, [2304]> transformer_resblocks_3_attn_in_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_3_attn_in_proj_bias_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50872128)))]; | |
| tensor<fp16, [50, ?, 2304]> linear_12_cast_fp16 = linear(bias = transformer_resblocks_3_attn_in_proj_bias_to_fp16, weight = transformer_resblocks_3_attn_in_proj_weight_to_fp16, x = ret_15_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")]; | |
| tensor<int32, [4]> concat_29x = const()[name = tensor<string, []>("concat_29x"), val = tensor<int32, [4]>([50, -1, 3, 768])]; | |
| tensor<fp16, [50, ?, 3, 768]> var_465_cast_fp16 = reshape(shape = concat_29x, x = linear_12_cast_fp16)[name = tensor<string, []>("op_465_cast_fp16")]; | |
| tensor<int32, [1]> var_466_axes_0 = const()[name = tensor<string, []>("op_466_axes_0"), val = tensor<int32, [1]>([0])]; | |
| tensor<fp16, [1, 50, ?, 3, 768]> var_466_cast_fp16 = expand_dims(axes = var_466_axes_0, x = var_465_cast_fp16)[name = tensor<string, []>("op_466_cast_fp16")]; | |
| tensor<int32, [5]> var_467_perm_0 = const()[name = tensor<string, []>("op_467_perm_0"), val = tensor<int32, [5]>([-2, 1, 2, 0, 4])]; | |
| tensor<int32, [1]> var_468_axes_0 = const()[name = tensor<string, []>("op_468_axes_0"), val = tensor<int32, [1]>([-2])]; | |
| tensor<fp16, [3, 50, ?, 1, 768]> var_467_cast_fp16 = transpose(perm = var_467_perm_0, x = var_466_cast_fp16)[name = tensor<string, []>("transpose_46")]; | |
| tensor<fp16, [3, 50, ?, 768]> var_468_cast_fp16 = squeeze(axes = var_468_axes_0, x = var_467_cast_fp16)[name = tensor<string, []>("op_468_cast_fp16")]; | |
| tensor<int32, [4]> q_19_begin_0 = const()[name = tensor<string, []>("q_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> q_19_end_0 = const()[name = tensor<string, []>("q_19_end_0"), val = tensor<int32, [4]>([1, 50, 0, 768])]; | |
| tensor<bool, [4]> q_19_end_mask_0 = const()[name = tensor<string, []>("q_19_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> q_19_squeeze_mask_0 = const()[name = tensor<string, []>("q_19_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> q_19_cast_fp16 = slice_by_index(begin = q_19_begin_0, end = q_19_end_0, end_mask = q_19_end_mask_0, squeeze_mask = q_19_squeeze_mask_0, x = var_468_cast_fp16)[name = tensor<string, []>("q_19_cast_fp16")]; | |
| tensor<int32, [4]> k_19_begin_0 = const()[name = tensor<string, []>("k_19_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])]; | |
| tensor<int32, [4]> k_19_end_0 = const()[name = tensor<string, []>("k_19_end_0"), val = tensor<int32, [4]>([2, 50, 0, 768])]; | |
| tensor<bool, [4]> k_19_end_mask_0 = const()[name = tensor<string, []>("k_19_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> k_19_squeeze_mask_0 = const()[name = tensor<string, []>("k_19_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> k_19_cast_fp16 = slice_by_index(begin = k_19_begin_0, end = k_19_end_0, end_mask = k_19_end_mask_0, squeeze_mask = k_19_squeeze_mask_0, x = var_468_cast_fp16)[name = tensor<string, []>("k_19_cast_fp16")]; | |
| tensor<int32, [4]> v_19_begin_0 = const()[name = tensor<string, []>("v_19_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])]; | |
| tensor<int32, [4]> v_19_end_0 = const()[name = tensor<string, []>("v_19_end_0"), val = tensor<int32, [4]>([3, 50, 0, 768])]; | |
| tensor<bool, [4]> v_19_end_mask_0 = const()[name = tensor<string, []>("v_19_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> v_19_squeeze_mask_0 = const()[name = tensor<string, []>("v_19_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> v_19_cast_fp16 = slice_by_index(begin = v_19_begin_0, end = v_19_end_0, end_mask = v_19_end_mask_0, squeeze_mask = v_19_squeeze_mask_0, x = var_468_cast_fp16)[name = tensor<string, []>("v_19_cast_fp16")]; | |
| tensor<int32, [3]> concat_30x = const()[name = tensor<string, []>("concat_30x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_477_cast_fp16 = reshape(shape = concat_30x, x = q_19_cast_fp16)[name = tensor<string, []>("op_477_cast_fp16")]; | |
| tensor<int32, [3]> q_21_perm_0 = const()[name = tensor<string, []>("q_21_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_31x = const()[name = tensor<string, []>("concat_31x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_484_cast_fp16 = reshape(shape = concat_31x, x = k_19_cast_fp16)[name = tensor<string, []>("op_484_cast_fp16")]; | |
| tensor<int32, [3]> k_21_perm_0 = const()[name = tensor<string, []>("k_21_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_32x = const()[name = tensor<string, []>("concat_32x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_491_cast_fp16 = reshape(shape = concat_32x, x = v_19_cast_fp16)[name = tensor<string, []>("op_491_cast_fp16")]; | |
| tensor<int32, [3]> v_21_perm_0 = const()[name = tensor<string, []>("v_21_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [4]> concat_33x = const()[name = tensor<string, []>("concat_33x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> q_21_cast_fp16 = transpose(perm = q_21_perm_0, x = var_477_cast_fp16)[name = tensor<string, []>("transpose_45")]; | |
| tensor<fp16, [?, 12, 50, 64]> q_23_cast_fp16 = reshape(shape = concat_33x, x = q_21_cast_fp16)[name = tensor<string, []>("q_23_cast_fp16")]; | |
| tensor<int32, [4]> concat_34x = const()[name = tensor<string, []>("concat_34x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> k_21_cast_fp16 = transpose(perm = k_21_perm_0, x = var_484_cast_fp16)[name = tensor<string, []>("transpose_44")]; | |
| tensor<fp16, [?, 12, 50, 64]> k_23_cast_fp16 = reshape(shape = concat_34x, x = k_21_cast_fp16)[name = tensor<string, []>("k_23_cast_fp16")]; | |
| tensor<int32, [4]> concat_35x = const()[name = tensor<string, []>("concat_35x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> v_21_cast_fp16 = transpose(perm = v_21_perm_0, x = var_491_cast_fp16)[name = tensor<string, []>("transpose_43")]; | |
| tensor<fp16, [?, 12, 50, 64]> v_23_cast_fp16 = reshape(shape = concat_35x, x = v_21_cast_fp16)[name = tensor<string, []>("v_23_cast_fp16")]; | |
| tensor<fp16, []> mul_7_y_0_to_fp16 = const()[name = tensor<string, []>("mul_7_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [?, 12, 50, 64]> mul_7_cast_fp16 = mul(x = q_23_cast_fp16, y = mul_7_y_0_to_fp16)[name = tensor<string, []>("mul_7_cast_fp16")]; | |
| tensor<bool, []> matmul_3_transpose_y_0 = const()[name = tensor<string, []>("matmul_3_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> matmul_3_transpose_x_0 = const()[name = tensor<string, []>("matmul_3_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 50]> matmul_3_cast_fp16 = matmul(transpose_x = matmul_3_transpose_x_0, transpose_y = matmul_3_transpose_y_0, x = mul_7_cast_fp16, y = k_23_cast_fp16)[name = tensor<string, []>("matmul_3_cast_fp16")]; | |
| tensor<int32, []> softmax_3_axis_0 = const()[name = tensor<string, []>("softmax_3_axis_0"), val = tensor<int32, []>(-1)]; | |
| tensor<fp16, [?, 12, 50, 50]> softmax_3_cast_fp16 = softmax(axis = softmax_3_axis_0, x = matmul_3_cast_fp16)[name = tensor<string, []>("softmax_3_cast_fp16")]; | |
| tensor<bool, []> attn_output_19_transpose_x_0 = const()[name = tensor<string, []>("attn_output_19_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_19_transpose_y_0 = const()[name = tensor<string, []>("attn_output_19_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 64]> attn_output_19_cast_fp16 = matmul(transpose_x = attn_output_19_transpose_x_0, transpose_y = attn_output_19_transpose_y_0, x = softmax_3_cast_fp16, y = v_23_cast_fp16)[name = tensor<string, []>("attn_output_19_cast_fp16")]; | |
| tensor<int32, [4]> var_501 = const()[name = tensor<string, []>("op_501"), val = tensor<int32, [4]>([2, 0, 1, 3])]; | |
| tensor<int32, [2]> concat_36x = const()[name = tensor<string, []>("concat_36x"), val = tensor<int32, [2]>([-1, 768])]; | |
| tensor<fp16, [50, ?, 12, 64]> var_502_cast_fp16 = transpose(perm = var_501, x = attn_output_19_cast_fp16)[name = tensor<string, []>("transpose_42")]; | |
| tensor<fp16, [?, 768]> attn_output_21_cast_fp16 = reshape(shape = concat_36x, x = var_502_cast_fp16)[name = tensor<string, []>("attn_output_21_cast_fp16")]; | |
| tensor<fp16, [768, 768]> transformer_resblocks_3_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_3_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50876800)))]; | |
| tensor<fp16, [768]> transformer_resblocks_3_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_3_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52056512)))]; | |
| tensor<fp16, [?, 768]> linear_13_cast_fp16 = linear(bias = transformer_resblocks_3_attn_out_proj_bias_to_fp16, weight = transformer_resblocks_3_attn_out_proj_weight_to_fp16, x = attn_output_21_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")]; | |
| tensor<int32, [3]> concat_37x = const()[name = tensor<string, []>("concat_37x"), val = tensor<int32, [3]>([50, -1, 768])]; | |
| tensor<fp16, [50, ?, 768]> var_511_cast_fp16 = reshape(shape = concat_37x, x = linear_13_cast_fp16)[name = tensor<string, []>("op_511_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_35_cast_fp16 = add(x = x_33_cast_fp16, y = var_511_cast_fp16)[name = tensor<string, []>("x_35_cast_fp16")]; | |
| tensor<int32, [1]> ret_17_axes_0 = const()[name = tensor<string, []>("ret_17_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_3_ln_2_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_3_ln_2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52058112)))]; | |
| tensor<fp16, [768]> transformer_resblocks_3_ln_2_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_3_ln_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52059712)))]; | |
| tensor<fp16, [50, ?, 768]> ret_17_cast_fp16 = layer_norm(axes = ret_17_axes_0, beta = transformer_resblocks_3_ln_2_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_3_ln_2_weight_to_fp16, x = x_35_cast_fp16)[name = tensor<string, []>("ret_17_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> transformer_resblocks_3_mlp_c_fc_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_3_mlp_c_fc_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52061312)))]; | |
| tensor<fp16, [3072]> transformer_resblocks_3_mlp_c_fc_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_3_mlp_c_fc_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56779968)))]; | |
| tensor<fp16, [50, ?, 3072]> linear_14_cast_fp16 = linear(bias = transformer_resblocks_3_mlp_c_fc_bias_to_fp16, weight = transformer_resblocks_3_mlp_c_fc_weight_to_fp16, x = ret_17_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")]; | |
| tensor<fp16, []> var_527_to_fp16 = const()[name = tensor<string, []>("op_527_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [50, ?, 3072]> var_528_cast_fp16 = mul(x = linear_14_cast_fp16, y = var_527_to_fp16)[name = tensor<string, []>("op_528_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> var_529_cast_fp16 = sigmoid(x = var_528_cast_fp16)[name = tensor<string, []>("op_529_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> input_33_cast_fp16 = mul(x = linear_14_cast_fp16, y = var_529_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> transformer_resblocks_3_mlp_c_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_3_mlp_c_proj_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56786176)))]; | |
| tensor<fp16, [768]> transformer_resblocks_3_mlp_c_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_3_mlp_c_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61504832)))]; | |
| tensor<fp16, [50, ?, 768]> linear_15_cast_fp16 = linear(bias = transformer_resblocks_3_mlp_c_proj_bias_to_fp16, weight = transformer_resblocks_3_mlp_c_proj_weight_to_fp16, x = input_33_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_39_cast_fp16 = add(x = x_35_cast_fp16, y = linear_15_cast_fp16)[name = tensor<string, []>("x_39_cast_fp16")]; | |
| tensor<int32, [1]> ret_19_axes_0 = const()[name = tensor<string, []>("ret_19_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_4_ln_1_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_4_ln_1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61506432)))]; | |
| tensor<fp16, [768]> transformer_resblocks_4_ln_1_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_4_ln_1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61508032)))]; | |
| tensor<fp16, [50, ?, 768]> ret_19_cast_fp16 = layer_norm(axes = ret_19_axes_0, beta = transformer_resblocks_4_ln_1_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_4_ln_1_weight_to_fp16, x = x_39_cast_fp16)[name = tensor<string, []>("ret_19_cast_fp16")]; | |
| tensor<fp16, [2304, 768]> transformer_resblocks_4_attn_in_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_4_attn_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61509632)))]; | |
| tensor<fp16, [2304]> transformer_resblocks_4_attn_in_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_4_attn_in_proj_bias_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65048640)))]; | |
| tensor<fp16, [50, ?, 2304]> linear_16_cast_fp16 = linear(bias = transformer_resblocks_4_attn_in_proj_bias_to_fp16, weight = transformer_resblocks_4_attn_in_proj_weight_to_fp16, x = ret_19_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")]; | |
| tensor<int32, [4]> concat_38x = const()[name = tensor<string, []>("concat_38x"), val = tensor<int32, [4]>([50, -1, 3, 768])]; | |
| tensor<fp16, [50, ?, 3, 768]> var_570_cast_fp16 = reshape(shape = concat_38x, x = linear_16_cast_fp16)[name = tensor<string, []>("op_570_cast_fp16")]; | |
| tensor<int32, [1]> var_571_axes_0 = const()[name = tensor<string, []>("op_571_axes_0"), val = tensor<int32, [1]>([0])]; | |
| tensor<fp16, [1, 50, ?, 3, 768]> var_571_cast_fp16 = expand_dims(axes = var_571_axes_0, x = var_570_cast_fp16)[name = tensor<string, []>("op_571_cast_fp16")]; | |
| tensor<int32, [5]> var_572_perm_0 = const()[name = tensor<string, []>("op_572_perm_0"), val = tensor<int32, [5]>([-2, 1, 2, 0, 4])]; | |
| tensor<int32, [1]> var_573_axes_0 = const()[name = tensor<string, []>("op_573_axes_0"), val = tensor<int32, [1]>([-2])]; | |
| tensor<fp16, [3, 50, ?, 1, 768]> var_572_cast_fp16 = transpose(perm = var_572_perm_0, x = var_571_cast_fp16)[name = tensor<string, []>("transpose_41")]; | |
| tensor<fp16, [3, 50, ?, 768]> var_573_cast_fp16 = squeeze(axes = var_573_axes_0, x = var_572_cast_fp16)[name = tensor<string, []>("op_573_cast_fp16")]; | |
| tensor<int32, [4]> q_25_begin_0 = const()[name = tensor<string, []>("q_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> q_25_end_0 = const()[name = tensor<string, []>("q_25_end_0"), val = tensor<int32, [4]>([1, 50, 0, 768])]; | |
| tensor<bool, [4]> q_25_end_mask_0 = const()[name = tensor<string, []>("q_25_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> q_25_squeeze_mask_0 = const()[name = tensor<string, []>("q_25_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> q_25_cast_fp16 = slice_by_index(begin = q_25_begin_0, end = q_25_end_0, end_mask = q_25_end_mask_0, squeeze_mask = q_25_squeeze_mask_0, x = var_573_cast_fp16)[name = tensor<string, []>("q_25_cast_fp16")]; | |
| tensor<int32, [4]> k_25_begin_0 = const()[name = tensor<string, []>("k_25_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])]; | |
| tensor<int32, [4]> k_25_end_0 = const()[name = tensor<string, []>("k_25_end_0"), val = tensor<int32, [4]>([2, 50, 0, 768])]; | |
| tensor<bool, [4]> k_25_end_mask_0 = const()[name = tensor<string, []>("k_25_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> k_25_squeeze_mask_0 = const()[name = tensor<string, []>("k_25_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> k_25_cast_fp16 = slice_by_index(begin = k_25_begin_0, end = k_25_end_0, end_mask = k_25_end_mask_0, squeeze_mask = k_25_squeeze_mask_0, x = var_573_cast_fp16)[name = tensor<string, []>("k_25_cast_fp16")]; | |
| tensor<int32, [4]> v_25_begin_0 = const()[name = tensor<string, []>("v_25_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])]; | |
| tensor<int32, [4]> v_25_end_0 = const()[name = tensor<string, []>("v_25_end_0"), val = tensor<int32, [4]>([3, 50, 0, 768])]; | |
| tensor<bool, [4]> v_25_end_mask_0 = const()[name = tensor<string, []>("v_25_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> v_25_squeeze_mask_0 = const()[name = tensor<string, []>("v_25_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> v_25_cast_fp16 = slice_by_index(begin = v_25_begin_0, end = v_25_end_0, end_mask = v_25_end_mask_0, squeeze_mask = v_25_squeeze_mask_0, x = var_573_cast_fp16)[name = tensor<string, []>("v_25_cast_fp16")]; | |
| tensor<int32, [3]> concat_39x = const()[name = tensor<string, []>("concat_39x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_582_cast_fp16 = reshape(shape = concat_39x, x = q_25_cast_fp16)[name = tensor<string, []>("op_582_cast_fp16")]; | |
| tensor<int32, [3]> q_27_perm_0 = const()[name = tensor<string, []>("q_27_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_40x = const()[name = tensor<string, []>("concat_40x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_589_cast_fp16 = reshape(shape = concat_40x, x = k_25_cast_fp16)[name = tensor<string, []>("op_589_cast_fp16")]; | |
| tensor<int32, [3]> k_27_perm_0 = const()[name = tensor<string, []>("k_27_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_41x = const()[name = tensor<string, []>("concat_41x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_596_cast_fp16 = reshape(shape = concat_41x, x = v_25_cast_fp16)[name = tensor<string, []>("op_596_cast_fp16")]; | |
| tensor<int32, [3]> v_27_perm_0 = const()[name = tensor<string, []>("v_27_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [4]> concat_42x = const()[name = tensor<string, []>("concat_42x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> q_27_cast_fp16 = transpose(perm = q_27_perm_0, x = var_582_cast_fp16)[name = tensor<string, []>("transpose_40")]; | |
| tensor<fp16, [?, 12, 50, 64]> q_29_cast_fp16 = reshape(shape = concat_42x, x = q_27_cast_fp16)[name = tensor<string, []>("q_29_cast_fp16")]; | |
| tensor<int32, [4]> concat_43x = const()[name = tensor<string, []>("concat_43x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> k_27_cast_fp16 = transpose(perm = k_27_perm_0, x = var_589_cast_fp16)[name = tensor<string, []>("transpose_39")]; | |
| tensor<fp16, [?, 12, 50, 64]> k_29_cast_fp16 = reshape(shape = concat_43x, x = k_27_cast_fp16)[name = tensor<string, []>("k_29_cast_fp16")]; | |
| tensor<int32, [4]> concat_44x = const()[name = tensor<string, []>("concat_44x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> v_27_cast_fp16 = transpose(perm = v_27_perm_0, x = var_596_cast_fp16)[name = tensor<string, []>("transpose_38")]; | |
| tensor<fp16, [?, 12, 50, 64]> v_29_cast_fp16 = reshape(shape = concat_44x, x = v_27_cast_fp16)[name = tensor<string, []>("v_29_cast_fp16")]; | |
| tensor<fp16, []> mul_9_y_0_to_fp16 = const()[name = tensor<string, []>("mul_9_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [?, 12, 50, 64]> mul_9_cast_fp16 = mul(x = q_29_cast_fp16, y = mul_9_y_0_to_fp16)[name = tensor<string, []>("mul_9_cast_fp16")]; | |
| tensor<bool, []> matmul_4_transpose_y_0 = const()[name = tensor<string, []>("matmul_4_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> matmul_4_transpose_x_0 = const()[name = tensor<string, []>("matmul_4_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 50]> matmul_4_cast_fp16 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = mul_9_cast_fp16, y = k_29_cast_fp16)[name = tensor<string, []>("matmul_4_cast_fp16")]; | |
| tensor<int32, []> softmax_4_axis_0 = const()[name = tensor<string, []>("softmax_4_axis_0"), val = tensor<int32, []>(-1)]; | |
| tensor<fp16, [?, 12, 50, 50]> softmax_4_cast_fp16 = softmax(axis = softmax_4_axis_0, x = matmul_4_cast_fp16)[name = tensor<string, []>("softmax_4_cast_fp16")]; | |
| tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 64]> attn_output_25_cast_fp16 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = softmax_4_cast_fp16, y = v_29_cast_fp16)[name = tensor<string, []>("attn_output_25_cast_fp16")]; | |
| tensor<int32, [4]> var_606 = const()[name = tensor<string, []>("op_606"), val = tensor<int32, [4]>([2, 0, 1, 3])]; | |
| tensor<int32, [2]> concat_45x = const()[name = tensor<string, []>("concat_45x"), val = tensor<int32, [2]>([-1, 768])]; | |
| tensor<fp16, [50, ?, 12, 64]> var_607_cast_fp16 = transpose(perm = var_606, x = attn_output_25_cast_fp16)[name = tensor<string, []>("transpose_37")]; | |
| tensor<fp16, [?, 768]> attn_output_27_cast_fp16 = reshape(shape = concat_45x, x = var_607_cast_fp16)[name = tensor<string, []>("attn_output_27_cast_fp16")]; | |
| tensor<fp16, [768, 768]> transformer_resblocks_4_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_4_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65053312)))]; | |
| tensor<fp16, [768]> transformer_resblocks_4_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_4_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66233024)))]; | |
| tensor<fp16, [?, 768]> linear_17_cast_fp16 = linear(bias = transformer_resblocks_4_attn_out_proj_bias_to_fp16, weight = transformer_resblocks_4_attn_out_proj_weight_to_fp16, x = attn_output_27_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")]; | |
| tensor<int32, [3]> concat_46x = const()[name = tensor<string, []>("concat_46x"), val = tensor<int32, [3]>([50, -1, 768])]; | |
| tensor<fp16, [50, ?, 768]> var_616_cast_fp16 = reshape(shape = concat_46x, x = linear_17_cast_fp16)[name = tensor<string, []>("op_616_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_41_cast_fp16 = add(x = x_39_cast_fp16, y = var_616_cast_fp16)[name = tensor<string, []>("x_41_cast_fp16")]; | |
| tensor<int32, [1]> ret_21_axes_0 = const()[name = tensor<string, []>("ret_21_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_4_ln_2_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_4_ln_2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66234624)))]; | |
| tensor<fp16, [768]> transformer_resblocks_4_ln_2_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_4_ln_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66236224)))]; | |
| tensor<fp16, [50, ?, 768]> ret_21_cast_fp16 = layer_norm(axes = ret_21_axes_0, beta = transformer_resblocks_4_ln_2_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_4_ln_2_weight_to_fp16, x = x_41_cast_fp16)[name = tensor<string, []>("ret_21_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> transformer_resblocks_4_mlp_c_fc_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_4_mlp_c_fc_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66237824)))]; | |
| tensor<fp16, [3072]> transformer_resblocks_4_mlp_c_fc_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_4_mlp_c_fc_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70956480)))]; | |
| tensor<fp16, [50, ?, 3072]> linear_18_cast_fp16 = linear(bias = transformer_resblocks_4_mlp_c_fc_bias_to_fp16, weight = transformer_resblocks_4_mlp_c_fc_weight_to_fp16, x = ret_21_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")]; | |
| tensor<fp16, []> var_632_to_fp16 = const()[name = tensor<string, []>("op_632_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [50, ?, 3072]> var_633_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_632_to_fp16)[name = tensor<string, []>("op_633_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> var_634_cast_fp16 = sigmoid(x = var_633_cast_fp16)[name = tensor<string, []>("op_634_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> input_41_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_634_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> transformer_resblocks_4_mlp_c_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_4_mlp_c_proj_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70962688)))]; | |
| tensor<fp16, [768]> transformer_resblocks_4_mlp_c_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_4_mlp_c_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75681344)))]; | |
| tensor<fp16, [50, ?, 768]> linear_19_cast_fp16 = linear(bias = transformer_resblocks_4_mlp_c_proj_bias_to_fp16, weight = transformer_resblocks_4_mlp_c_proj_weight_to_fp16, x = input_41_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_45_cast_fp16 = add(x = x_41_cast_fp16, y = linear_19_cast_fp16)[name = tensor<string, []>("x_45_cast_fp16")]; | |
| tensor<int32, [1]> ret_23_axes_0 = const()[name = tensor<string, []>("ret_23_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_5_ln_1_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_5_ln_1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75682944)))]; | |
| tensor<fp16, [768]> transformer_resblocks_5_ln_1_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_5_ln_1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75684544)))]; | |
| tensor<fp16, [50, ?, 768]> ret_23_cast_fp16 = layer_norm(axes = ret_23_axes_0, beta = transformer_resblocks_5_ln_1_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_5_ln_1_weight_to_fp16, x = x_45_cast_fp16)[name = tensor<string, []>("ret_23_cast_fp16")]; | |
| tensor<fp16, [2304, 768]> transformer_resblocks_5_attn_in_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_5_attn_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75686144)))]; | |
| tensor<fp16, [2304]> transformer_resblocks_5_attn_in_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_5_attn_in_proj_bias_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79225152)))]; | |
| tensor<fp16, [50, ?, 2304]> linear_20_cast_fp16 = linear(bias = transformer_resblocks_5_attn_in_proj_bias_to_fp16, weight = transformer_resblocks_5_attn_in_proj_weight_to_fp16, x = ret_23_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")]; | |
| tensor<int32, [4]> concat_47x = const()[name = tensor<string, []>("concat_47x"), val = tensor<int32, [4]>([50, -1, 3, 768])]; | |
| tensor<fp16, [50, ?, 3, 768]> var_675_cast_fp16 = reshape(shape = concat_47x, x = linear_20_cast_fp16)[name = tensor<string, []>("op_675_cast_fp16")]; | |
| tensor<int32, [1]> var_676_axes_0 = const()[name = tensor<string, []>("op_676_axes_0"), val = tensor<int32, [1]>([0])]; | |
| tensor<fp16, [1, 50, ?, 3, 768]> var_676_cast_fp16 = expand_dims(axes = var_676_axes_0, x = var_675_cast_fp16)[name = tensor<string, []>("op_676_cast_fp16")]; | |
| tensor<int32, [5]> var_677_perm_0 = const()[name = tensor<string, []>("op_677_perm_0"), val = tensor<int32, [5]>([-2, 1, 2, 0, 4])]; | |
| tensor<int32, [1]> var_678_axes_0 = const()[name = tensor<string, []>("op_678_axes_0"), val = tensor<int32, [1]>([-2])]; | |
| tensor<fp16, [3, 50, ?, 1, 768]> var_677_cast_fp16 = transpose(perm = var_677_perm_0, x = var_676_cast_fp16)[name = tensor<string, []>("transpose_36")]; | |
| tensor<fp16, [3, 50, ?, 768]> var_678_cast_fp16 = squeeze(axes = var_678_axes_0, x = var_677_cast_fp16)[name = tensor<string, []>("op_678_cast_fp16")]; | |
| tensor<int32, [4]> q_31_begin_0 = const()[name = tensor<string, []>("q_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> q_31_end_0 = const()[name = tensor<string, []>("q_31_end_0"), val = tensor<int32, [4]>([1, 50, 0, 768])]; | |
| tensor<bool, [4]> q_31_end_mask_0 = const()[name = tensor<string, []>("q_31_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> q_31_squeeze_mask_0 = const()[name = tensor<string, []>("q_31_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> q_31_cast_fp16 = slice_by_index(begin = q_31_begin_0, end = q_31_end_0, end_mask = q_31_end_mask_0, squeeze_mask = q_31_squeeze_mask_0, x = var_678_cast_fp16)[name = tensor<string, []>("q_31_cast_fp16")]; | |
| tensor<int32, [4]> k_31_begin_0 = const()[name = tensor<string, []>("k_31_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])]; | |
| tensor<int32, [4]> k_31_end_0 = const()[name = tensor<string, []>("k_31_end_0"), val = tensor<int32, [4]>([2, 50, 0, 768])]; | |
| tensor<bool, [4]> k_31_end_mask_0 = const()[name = tensor<string, []>("k_31_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> k_31_squeeze_mask_0 = const()[name = tensor<string, []>("k_31_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> k_31_cast_fp16 = slice_by_index(begin = k_31_begin_0, end = k_31_end_0, end_mask = k_31_end_mask_0, squeeze_mask = k_31_squeeze_mask_0, x = var_678_cast_fp16)[name = tensor<string, []>("k_31_cast_fp16")]; | |
| tensor<int32, [4]> v_31_begin_0 = const()[name = tensor<string, []>("v_31_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])]; | |
| tensor<int32, [4]> v_31_end_0 = const()[name = tensor<string, []>("v_31_end_0"), val = tensor<int32, [4]>([3, 50, 0, 768])]; | |
| tensor<bool, [4]> v_31_end_mask_0 = const()[name = tensor<string, []>("v_31_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> v_31_squeeze_mask_0 = const()[name = tensor<string, []>("v_31_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> v_31_cast_fp16 = slice_by_index(begin = v_31_begin_0, end = v_31_end_0, end_mask = v_31_end_mask_0, squeeze_mask = v_31_squeeze_mask_0, x = var_678_cast_fp16)[name = tensor<string, []>("v_31_cast_fp16")]; | |
| tensor<int32, [3]> concat_48x = const()[name = tensor<string, []>("concat_48x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_687_cast_fp16 = reshape(shape = concat_48x, x = q_31_cast_fp16)[name = tensor<string, []>("op_687_cast_fp16")]; | |
| tensor<int32, [3]> q_33_perm_0 = const()[name = tensor<string, []>("q_33_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_49x = const()[name = tensor<string, []>("concat_49x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_694_cast_fp16 = reshape(shape = concat_49x, x = k_31_cast_fp16)[name = tensor<string, []>("op_694_cast_fp16")]; | |
| tensor<int32, [3]> k_33_perm_0 = const()[name = tensor<string, []>("k_33_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_50x = const()[name = tensor<string, []>("concat_50x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_701_cast_fp16 = reshape(shape = concat_50x, x = v_31_cast_fp16)[name = tensor<string, []>("op_701_cast_fp16")]; | |
| tensor<int32, [3]> v_33_perm_0 = const()[name = tensor<string, []>("v_33_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [4]> concat_51x = const()[name = tensor<string, []>("concat_51x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> q_33_cast_fp16 = transpose(perm = q_33_perm_0, x = var_687_cast_fp16)[name = tensor<string, []>("transpose_35")]; | |
| tensor<fp16, [?, 12, 50, 64]> q_35_cast_fp16 = reshape(shape = concat_51x, x = q_33_cast_fp16)[name = tensor<string, []>("q_35_cast_fp16")]; | |
| tensor<int32, [4]> concat_52x = const()[name = tensor<string, []>("concat_52x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> k_33_cast_fp16 = transpose(perm = k_33_perm_0, x = var_694_cast_fp16)[name = tensor<string, []>("transpose_34")]; | |
| tensor<fp16, [?, 12, 50, 64]> k_35_cast_fp16 = reshape(shape = concat_52x, x = k_33_cast_fp16)[name = tensor<string, []>("k_35_cast_fp16")]; | |
| tensor<int32, [4]> concat_53x = const()[name = tensor<string, []>("concat_53x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> v_33_cast_fp16 = transpose(perm = v_33_perm_0, x = var_701_cast_fp16)[name = tensor<string, []>("transpose_33")]; | |
| tensor<fp16, [?, 12, 50, 64]> v_35_cast_fp16 = reshape(shape = concat_53x, x = v_33_cast_fp16)[name = tensor<string, []>("v_35_cast_fp16")]; | |
| tensor<fp16, []> mul_11_y_0_to_fp16 = const()[name = tensor<string, []>("mul_11_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [?, 12, 50, 64]> mul_11_cast_fp16 = mul(x = q_35_cast_fp16, y = mul_11_y_0_to_fp16)[name = tensor<string, []>("mul_11_cast_fp16")]; | |
| tensor<bool, []> matmul_5_transpose_y_0 = const()[name = tensor<string, []>("matmul_5_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> matmul_5_transpose_x_0 = const()[name = tensor<string, []>("matmul_5_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 50]> matmul_5_cast_fp16 = matmul(transpose_x = matmul_5_transpose_x_0, transpose_y = matmul_5_transpose_y_0, x = mul_11_cast_fp16, y = k_35_cast_fp16)[name = tensor<string, []>("matmul_5_cast_fp16")]; | |
| tensor<int32, []> softmax_5_axis_0 = const()[name = tensor<string, []>("softmax_5_axis_0"), val = tensor<int32, []>(-1)]; | |
| tensor<fp16, [?, 12, 50, 50]> softmax_5_cast_fp16 = softmax(axis = softmax_5_axis_0, x = matmul_5_cast_fp16)[name = tensor<string, []>("softmax_5_cast_fp16")]; | |
| tensor<bool, []> attn_output_31_transpose_x_0 = const()[name = tensor<string, []>("attn_output_31_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_31_transpose_y_0 = const()[name = tensor<string, []>("attn_output_31_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 64]> attn_output_31_cast_fp16 = matmul(transpose_x = attn_output_31_transpose_x_0, transpose_y = attn_output_31_transpose_y_0, x = softmax_5_cast_fp16, y = v_35_cast_fp16)[name = tensor<string, []>("attn_output_31_cast_fp16")]; | |
| tensor<int32, [4]> var_711 = const()[name = tensor<string, []>("op_711"), val = tensor<int32, [4]>([2, 0, 1, 3])]; | |
| tensor<int32, [2]> concat_54x = const()[name = tensor<string, []>("concat_54x"), val = tensor<int32, [2]>([-1, 768])]; | |
| tensor<fp16, [50, ?, 12, 64]> var_712_cast_fp16 = transpose(perm = var_711, x = attn_output_31_cast_fp16)[name = tensor<string, []>("transpose_32")]; | |
| tensor<fp16, [?, 768]> attn_output_33_cast_fp16 = reshape(shape = concat_54x, x = var_712_cast_fp16)[name = tensor<string, []>("attn_output_33_cast_fp16")]; | |
| tensor<fp16, [768, 768]> transformer_resblocks_5_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_5_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79229824)))]; | |
| tensor<fp16, [768]> transformer_resblocks_5_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_5_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80409536)))]; | |
| tensor<fp16, [?, 768]> linear_21_cast_fp16 = linear(bias = transformer_resblocks_5_attn_out_proj_bias_to_fp16, weight = transformer_resblocks_5_attn_out_proj_weight_to_fp16, x = attn_output_33_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")]; | |
| tensor<int32, [3]> concat_55x = const()[name = tensor<string, []>("concat_55x"), val = tensor<int32, [3]>([50, -1, 768])]; | |
| tensor<fp16, [50, ?, 768]> var_721_cast_fp16 = reshape(shape = concat_55x, x = linear_21_cast_fp16)[name = tensor<string, []>("op_721_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_47_cast_fp16 = add(x = x_45_cast_fp16, y = var_721_cast_fp16)[name = tensor<string, []>("x_47_cast_fp16")]; | |
| tensor<int32, [1]> ret_25_axes_0 = const()[name = tensor<string, []>("ret_25_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_5_ln_2_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_5_ln_2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80411136)))]; | |
| tensor<fp16, [768]> transformer_resblocks_5_ln_2_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_5_ln_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80412736)))]; | |
| tensor<fp16, [50, ?, 768]> ret_25_cast_fp16 = layer_norm(axes = ret_25_axes_0, beta = transformer_resblocks_5_ln_2_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_5_ln_2_weight_to_fp16, x = x_47_cast_fp16)[name = tensor<string, []>("ret_25_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> transformer_resblocks_5_mlp_c_fc_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_5_mlp_c_fc_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80414336)))]; | |
| tensor<fp16, [3072]> transformer_resblocks_5_mlp_c_fc_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_5_mlp_c_fc_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85132992)))]; | |
| tensor<fp16, [50, ?, 3072]> linear_22_cast_fp16 = linear(bias = transformer_resblocks_5_mlp_c_fc_bias_to_fp16, weight = transformer_resblocks_5_mlp_c_fc_weight_to_fp16, x = ret_25_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")]; | |
| tensor<fp16, []> var_737_to_fp16 = const()[name = tensor<string, []>("op_737_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [50, ?, 3072]> var_738_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_737_to_fp16)[name = tensor<string, []>("op_738_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> var_739_cast_fp16 = sigmoid(x = var_738_cast_fp16)[name = tensor<string, []>("op_739_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> input_49_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_739_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> transformer_resblocks_5_mlp_c_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_5_mlp_c_proj_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85139200)))]; | |
| tensor<fp16, [768]> transformer_resblocks_5_mlp_c_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_5_mlp_c_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89857856)))]; | |
| tensor<fp16, [50, ?, 768]> linear_23_cast_fp16 = linear(bias = transformer_resblocks_5_mlp_c_proj_bias_to_fp16, weight = transformer_resblocks_5_mlp_c_proj_weight_to_fp16, x = input_49_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_51_cast_fp16 = add(x = x_47_cast_fp16, y = linear_23_cast_fp16)[name = tensor<string, []>("x_51_cast_fp16")]; | |
| tensor<int32, [1]> ret_27_axes_0 = const()[name = tensor<string, []>("ret_27_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_6_ln_1_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_6_ln_1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89859456)))]; | |
| tensor<fp16, [768]> transformer_resblocks_6_ln_1_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_6_ln_1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89861056)))]; | |
| tensor<fp16, [50, ?, 768]> ret_27_cast_fp16 = layer_norm(axes = ret_27_axes_0, beta = transformer_resblocks_6_ln_1_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_6_ln_1_weight_to_fp16, x = x_51_cast_fp16)[name = tensor<string, []>("ret_27_cast_fp16")]; | |
| tensor<fp16, [2304, 768]> transformer_resblocks_6_attn_in_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_6_attn_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89862656)))]; | |
| tensor<fp16, [2304]> transformer_resblocks_6_attn_in_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_6_attn_in_proj_bias_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93401664)))]; | |
| tensor<fp16, [50, ?, 2304]> linear_24_cast_fp16 = linear(bias = transformer_resblocks_6_attn_in_proj_bias_to_fp16, weight = transformer_resblocks_6_attn_in_proj_weight_to_fp16, x = ret_27_cast_fp16)[name = tensor<string, []>("linear_24_cast_fp16")]; | |
| tensor<int32, [4]> concat_56x = const()[name = tensor<string, []>("concat_56x"), val = tensor<int32, [4]>([50, -1, 3, 768])]; | |
| tensor<fp16, [50, ?, 3, 768]> var_780_cast_fp16 = reshape(shape = concat_56x, x = linear_24_cast_fp16)[name = tensor<string, []>("op_780_cast_fp16")]; | |
| tensor<int32, [1]> var_781_axes_0 = const()[name = tensor<string, []>("op_781_axes_0"), val = tensor<int32, [1]>([0])]; | |
| tensor<fp16, [1, 50, ?, 3, 768]> var_781_cast_fp16 = expand_dims(axes = var_781_axes_0, x = var_780_cast_fp16)[name = tensor<string, []>("op_781_cast_fp16")]; | |
| tensor<int32, [5]> var_782_perm_0 = const()[name = tensor<string, []>("op_782_perm_0"), val = tensor<int32, [5]>([-2, 1, 2, 0, 4])]; | |
| tensor<int32, [1]> var_783_axes_0 = const()[name = tensor<string, []>("op_783_axes_0"), val = tensor<int32, [1]>([-2])]; | |
| tensor<fp16, [3, 50, ?, 1, 768]> var_782_cast_fp16 = transpose(perm = var_782_perm_0, x = var_781_cast_fp16)[name = tensor<string, []>("transpose_31")]; | |
| tensor<fp16, [3, 50, ?, 768]> var_783_cast_fp16 = squeeze(axes = var_783_axes_0, x = var_782_cast_fp16)[name = tensor<string, []>("op_783_cast_fp16")]; | |
| tensor<int32, [4]> q_37_begin_0 = const()[name = tensor<string, []>("q_37_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> q_37_end_0 = const()[name = tensor<string, []>("q_37_end_0"), val = tensor<int32, [4]>([1, 50, 0, 768])]; | |
| tensor<bool, [4]> q_37_end_mask_0 = const()[name = tensor<string, []>("q_37_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> q_37_squeeze_mask_0 = const()[name = tensor<string, []>("q_37_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> q_37_cast_fp16 = slice_by_index(begin = q_37_begin_0, end = q_37_end_0, end_mask = q_37_end_mask_0, squeeze_mask = q_37_squeeze_mask_0, x = var_783_cast_fp16)[name = tensor<string, []>("q_37_cast_fp16")]; | |
| tensor<int32, [4]> k_37_begin_0 = const()[name = tensor<string, []>("k_37_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])]; | |
| tensor<int32, [4]> k_37_end_0 = const()[name = tensor<string, []>("k_37_end_0"), val = tensor<int32, [4]>([2, 50, 0, 768])]; | |
| tensor<bool, [4]> k_37_end_mask_0 = const()[name = tensor<string, []>("k_37_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> k_37_squeeze_mask_0 = const()[name = tensor<string, []>("k_37_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> k_37_cast_fp16 = slice_by_index(begin = k_37_begin_0, end = k_37_end_0, end_mask = k_37_end_mask_0, squeeze_mask = k_37_squeeze_mask_0, x = var_783_cast_fp16)[name = tensor<string, []>("k_37_cast_fp16")]; | |
| tensor<int32, [4]> v_37_begin_0 = const()[name = tensor<string, []>("v_37_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])]; | |
| tensor<int32, [4]> v_37_end_0 = const()[name = tensor<string, []>("v_37_end_0"), val = tensor<int32, [4]>([3, 50, 0, 768])]; | |
| tensor<bool, [4]> v_37_end_mask_0 = const()[name = tensor<string, []>("v_37_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> v_37_squeeze_mask_0 = const()[name = tensor<string, []>("v_37_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> v_37_cast_fp16 = slice_by_index(begin = v_37_begin_0, end = v_37_end_0, end_mask = v_37_end_mask_0, squeeze_mask = v_37_squeeze_mask_0, x = var_783_cast_fp16)[name = tensor<string, []>("v_37_cast_fp16")]; | |
| tensor<int32, [3]> concat_57x = const()[name = tensor<string, []>("concat_57x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_792_cast_fp16 = reshape(shape = concat_57x, x = q_37_cast_fp16)[name = tensor<string, []>("op_792_cast_fp16")]; | |
| tensor<int32, [3]> q_39_perm_0 = const()[name = tensor<string, []>("q_39_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_58x = const()[name = tensor<string, []>("concat_58x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_799_cast_fp16 = reshape(shape = concat_58x, x = k_37_cast_fp16)[name = tensor<string, []>("op_799_cast_fp16")]; | |
| tensor<int32, [3]> k_39_perm_0 = const()[name = tensor<string, []>("k_39_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_59x = const()[name = tensor<string, []>("concat_59x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_806_cast_fp16 = reshape(shape = concat_59x, x = v_37_cast_fp16)[name = tensor<string, []>("op_806_cast_fp16")]; | |
| tensor<int32, [3]> v_39_perm_0 = const()[name = tensor<string, []>("v_39_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [4]> concat_60x = const()[name = tensor<string, []>("concat_60x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> q_39_cast_fp16 = transpose(perm = q_39_perm_0, x = var_792_cast_fp16)[name = tensor<string, []>("transpose_30")]; | |
| tensor<fp16, [?, 12, 50, 64]> q_41_cast_fp16 = reshape(shape = concat_60x, x = q_39_cast_fp16)[name = tensor<string, []>("q_41_cast_fp16")]; | |
| tensor<int32, [4]> concat_61x = const()[name = tensor<string, []>("concat_61x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> k_39_cast_fp16 = transpose(perm = k_39_perm_0, x = var_799_cast_fp16)[name = tensor<string, []>("transpose_29")]; | |
| tensor<fp16, [?, 12, 50, 64]> k_41_cast_fp16 = reshape(shape = concat_61x, x = k_39_cast_fp16)[name = tensor<string, []>("k_41_cast_fp16")]; | |
| tensor<int32, [4]> concat_62x = const()[name = tensor<string, []>("concat_62x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> v_39_cast_fp16 = transpose(perm = v_39_perm_0, x = var_806_cast_fp16)[name = tensor<string, []>("transpose_28")]; | |
| tensor<fp16, [?, 12, 50, 64]> v_41_cast_fp16 = reshape(shape = concat_62x, x = v_39_cast_fp16)[name = tensor<string, []>("v_41_cast_fp16")]; | |
| tensor<fp16, []> mul_13_y_0_to_fp16 = const()[name = tensor<string, []>("mul_13_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [?, 12, 50, 64]> mul_13_cast_fp16 = mul(x = q_41_cast_fp16, y = mul_13_y_0_to_fp16)[name = tensor<string, []>("mul_13_cast_fp16")]; | |
| tensor<bool, []> matmul_6_transpose_y_0 = const()[name = tensor<string, []>("matmul_6_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> matmul_6_transpose_x_0 = const()[name = tensor<string, []>("matmul_6_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 50]> matmul_6_cast_fp16 = matmul(transpose_x = matmul_6_transpose_x_0, transpose_y = matmul_6_transpose_y_0, x = mul_13_cast_fp16, y = k_41_cast_fp16)[name = tensor<string, []>("matmul_6_cast_fp16")]; | |
| tensor<int32, []> softmax_6_axis_0 = const()[name = tensor<string, []>("softmax_6_axis_0"), val = tensor<int32, []>(-1)]; | |
| tensor<fp16, [?, 12, 50, 50]> softmax_6_cast_fp16 = softmax(axis = softmax_6_axis_0, x = matmul_6_cast_fp16)[name = tensor<string, []>("softmax_6_cast_fp16")]; | |
| tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 64]> attn_output_37_cast_fp16 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = softmax_6_cast_fp16, y = v_41_cast_fp16)[name = tensor<string, []>("attn_output_37_cast_fp16")]; | |
| tensor<int32, [4]> var_816 = const()[name = tensor<string, []>("op_816"), val = tensor<int32, [4]>([2, 0, 1, 3])]; | |
| tensor<int32, [2]> concat_63x = const()[name = tensor<string, []>("concat_63x"), val = tensor<int32, [2]>([-1, 768])]; | |
| tensor<fp16, [50, ?, 12, 64]> var_817_cast_fp16 = transpose(perm = var_816, x = attn_output_37_cast_fp16)[name = tensor<string, []>("transpose_27")]; | |
| tensor<fp16, [?, 768]> attn_output_39_cast_fp16 = reshape(shape = concat_63x, x = var_817_cast_fp16)[name = tensor<string, []>("attn_output_39_cast_fp16")]; | |
| tensor<fp16, [768, 768]> transformer_resblocks_6_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_6_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93406336)))]; | |
| tensor<fp16, [768]> transformer_resblocks_6_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_6_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94586048)))]; | |
| tensor<fp16, [?, 768]> linear_25_cast_fp16 = linear(bias = transformer_resblocks_6_attn_out_proj_bias_to_fp16, weight = transformer_resblocks_6_attn_out_proj_weight_to_fp16, x = attn_output_39_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")]; | |
| tensor<int32, [3]> concat_64x = const()[name = tensor<string, []>("concat_64x"), val = tensor<int32, [3]>([50, -1, 768])]; | |
| tensor<fp16, [50, ?, 768]> var_826_cast_fp16 = reshape(shape = concat_64x, x = linear_25_cast_fp16)[name = tensor<string, []>("op_826_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_53_cast_fp16 = add(x = x_51_cast_fp16, y = var_826_cast_fp16)[name = tensor<string, []>("x_53_cast_fp16")]; | |
| tensor<int32, [1]> ret_29_axes_0 = const()[name = tensor<string, []>("ret_29_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_6_ln_2_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_6_ln_2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94587648)))]; | |
| tensor<fp16, [768]> transformer_resblocks_6_ln_2_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_6_ln_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94589248)))]; | |
| tensor<fp16, [50, ?, 768]> ret_29_cast_fp16 = layer_norm(axes = ret_29_axes_0, beta = transformer_resblocks_6_ln_2_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_6_ln_2_weight_to_fp16, x = x_53_cast_fp16)[name = tensor<string, []>("ret_29_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> transformer_resblocks_6_mlp_c_fc_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_6_mlp_c_fc_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94590848)))]; | |
| tensor<fp16, [3072]> transformer_resblocks_6_mlp_c_fc_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_6_mlp_c_fc_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99309504)))]; | |
| tensor<fp16, [50, ?, 3072]> linear_26_cast_fp16 = linear(bias = transformer_resblocks_6_mlp_c_fc_bias_to_fp16, weight = transformer_resblocks_6_mlp_c_fc_weight_to_fp16, x = ret_29_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")]; | |
| tensor<fp16, []> var_842_to_fp16 = const()[name = tensor<string, []>("op_842_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [50, ?, 3072]> var_843_cast_fp16 = mul(x = linear_26_cast_fp16, y = var_842_to_fp16)[name = tensor<string, []>("op_843_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> var_844_cast_fp16 = sigmoid(x = var_843_cast_fp16)[name = tensor<string, []>("op_844_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> input_57_cast_fp16 = mul(x = linear_26_cast_fp16, y = var_844_cast_fp16)[name = tensor<string, []>("input_57_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> transformer_resblocks_6_mlp_c_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_6_mlp_c_proj_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99315712)))]; | |
| tensor<fp16, [768]> transformer_resblocks_6_mlp_c_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_6_mlp_c_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104034368)))]; | |
| tensor<fp16, [50, ?, 768]> linear_27_cast_fp16 = linear(bias = transformer_resblocks_6_mlp_c_proj_bias_to_fp16, weight = transformer_resblocks_6_mlp_c_proj_weight_to_fp16, x = input_57_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_57_cast_fp16 = add(x = x_53_cast_fp16, y = linear_27_cast_fp16)[name = tensor<string, []>("x_57_cast_fp16")]; | |
| tensor<int32, [1]> ret_31_axes_0 = const()[name = tensor<string, []>("ret_31_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_7_ln_1_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_7_ln_1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104035968)))]; | |
| tensor<fp16, [768]> transformer_resblocks_7_ln_1_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_7_ln_1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104037568)))]; | |
| tensor<fp16, [50, ?, 768]> ret_31_cast_fp16 = layer_norm(axes = ret_31_axes_0, beta = transformer_resblocks_7_ln_1_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_7_ln_1_weight_to_fp16, x = x_57_cast_fp16)[name = tensor<string, []>("ret_31_cast_fp16")]; | |
| tensor<fp16, [2304, 768]> transformer_resblocks_7_attn_in_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_7_attn_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104039168)))]; | |
| tensor<fp16, [2304]> transformer_resblocks_7_attn_in_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_7_attn_in_proj_bias_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107578176)))]; | |
| tensor<fp16, [50, ?, 2304]> linear_28_cast_fp16 = linear(bias = transformer_resblocks_7_attn_in_proj_bias_to_fp16, weight = transformer_resblocks_7_attn_in_proj_weight_to_fp16, x = ret_31_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")]; | |
| tensor<int32, [4]> concat_65x = const()[name = tensor<string, []>("concat_65x"), val = tensor<int32, [4]>([50, -1, 3, 768])]; | |
| tensor<fp16, [50, ?, 3, 768]> var_885_cast_fp16 = reshape(shape = concat_65x, x = linear_28_cast_fp16)[name = tensor<string, []>("op_885_cast_fp16")]; | |
| tensor<int32, [1]> var_886_axes_0 = const()[name = tensor<string, []>("op_886_axes_0"), val = tensor<int32, [1]>([0])]; | |
| tensor<fp16, [1, 50, ?, 3, 768]> var_886_cast_fp16 = expand_dims(axes = var_886_axes_0, x = var_885_cast_fp16)[name = tensor<string, []>("op_886_cast_fp16")]; | |
| tensor<int32, [5]> var_887_perm_0 = const()[name = tensor<string, []>("op_887_perm_0"), val = tensor<int32, [5]>([-2, 1, 2, 0, 4])]; | |
| tensor<int32, [1]> var_888_axes_0 = const()[name = tensor<string, []>("op_888_axes_0"), val = tensor<int32, [1]>([-2])]; | |
| tensor<fp16, [3, 50, ?, 1, 768]> var_887_cast_fp16 = transpose(perm = var_887_perm_0, x = var_886_cast_fp16)[name = tensor<string, []>("transpose_26")]; | |
| tensor<fp16, [3, 50, ?, 768]> var_888_cast_fp16 = squeeze(axes = var_888_axes_0, x = var_887_cast_fp16)[name = tensor<string, []>("op_888_cast_fp16")]; | |
| tensor<int32, [4]> q_43_begin_0 = const()[name = tensor<string, []>("q_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> q_43_end_0 = const()[name = tensor<string, []>("q_43_end_0"), val = tensor<int32, [4]>([1, 50, 0, 768])]; | |
| tensor<bool, [4]> q_43_end_mask_0 = const()[name = tensor<string, []>("q_43_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> q_43_squeeze_mask_0 = const()[name = tensor<string, []>("q_43_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> q_43_cast_fp16 = slice_by_index(begin = q_43_begin_0, end = q_43_end_0, end_mask = q_43_end_mask_0, squeeze_mask = q_43_squeeze_mask_0, x = var_888_cast_fp16)[name = tensor<string, []>("q_43_cast_fp16")]; | |
| tensor<int32, [4]> k_43_begin_0 = const()[name = tensor<string, []>("k_43_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])]; | |
| tensor<int32, [4]> k_43_end_0 = const()[name = tensor<string, []>("k_43_end_0"), val = tensor<int32, [4]>([2, 50, 0, 768])]; | |
| tensor<bool, [4]> k_43_end_mask_0 = const()[name = tensor<string, []>("k_43_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> k_43_squeeze_mask_0 = const()[name = tensor<string, []>("k_43_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> k_43_cast_fp16 = slice_by_index(begin = k_43_begin_0, end = k_43_end_0, end_mask = k_43_end_mask_0, squeeze_mask = k_43_squeeze_mask_0, x = var_888_cast_fp16)[name = tensor<string, []>("k_43_cast_fp16")]; | |
| tensor<int32, [4]> v_43_begin_0 = const()[name = tensor<string, []>("v_43_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])]; | |
| tensor<int32, [4]> v_43_end_0 = const()[name = tensor<string, []>("v_43_end_0"), val = tensor<int32, [4]>([3, 50, 0, 768])]; | |
| tensor<bool, [4]> v_43_end_mask_0 = const()[name = tensor<string, []>("v_43_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> v_43_squeeze_mask_0 = const()[name = tensor<string, []>("v_43_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> v_43_cast_fp16 = slice_by_index(begin = v_43_begin_0, end = v_43_end_0, end_mask = v_43_end_mask_0, squeeze_mask = v_43_squeeze_mask_0, x = var_888_cast_fp16)[name = tensor<string, []>("v_43_cast_fp16")]; | |
| tensor<int32, [3]> concat_66x = const()[name = tensor<string, []>("concat_66x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_897_cast_fp16 = reshape(shape = concat_66x, x = q_43_cast_fp16)[name = tensor<string, []>("op_897_cast_fp16")]; | |
| tensor<int32, [3]> q_45_perm_0 = const()[name = tensor<string, []>("q_45_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_67x = const()[name = tensor<string, []>("concat_67x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_904_cast_fp16 = reshape(shape = concat_67x, x = k_43_cast_fp16)[name = tensor<string, []>("op_904_cast_fp16")]; | |
| tensor<int32, [3]> k_45_perm_0 = const()[name = tensor<string, []>("k_45_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_68x = const()[name = tensor<string, []>("concat_68x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_911_cast_fp16 = reshape(shape = concat_68x, x = v_43_cast_fp16)[name = tensor<string, []>("op_911_cast_fp16")]; | |
| tensor<int32, [3]> v_45_perm_0 = const()[name = tensor<string, []>("v_45_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [4]> concat_69x = const()[name = tensor<string, []>("concat_69x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> q_45_cast_fp16 = transpose(perm = q_45_perm_0, x = var_897_cast_fp16)[name = tensor<string, []>("transpose_25")]; | |
| tensor<fp16, [?, 12, 50, 64]> q_47_cast_fp16 = reshape(shape = concat_69x, x = q_45_cast_fp16)[name = tensor<string, []>("q_47_cast_fp16")]; | |
| tensor<int32, [4]> concat_70x = const()[name = tensor<string, []>("concat_70x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> k_45_cast_fp16 = transpose(perm = k_45_perm_0, x = var_904_cast_fp16)[name = tensor<string, []>("transpose_24")]; | |
| tensor<fp16, [?, 12, 50, 64]> k_47_cast_fp16 = reshape(shape = concat_70x, x = k_45_cast_fp16)[name = tensor<string, []>("k_47_cast_fp16")]; | |
| tensor<int32, [4]> concat_71x = const()[name = tensor<string, []>("concat_71x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> v_45_cast_fp16 = transpose(perm = v_45_perm_0, x = var_911_cast_fp16)[name = tensor<string, []>("transpose_23")]; | |
| tensor<fp16, [?, 12, 50, 64]> v_47_cast_fp16 = reshape(shape = concat_71x, x = v_45_cast_fp16)[name = tensor<string, []>("v_47_cast_fp16")]; | |
| tensor<fp16, []> mul_15_y_0_to_fp16 = const()[name = tensor<string, []>("mul_15_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [?, 12, 50, 64]> mul_15_cast_fp16 = mul(x = q_47_cast_fp16, y = mul_15_y_0_to_fp16)[name = tensor<string, []>("mul_15_cast_fp16")]; | |
| tensor<bool, []> matmul_7_transpose_y_0 = const()[name = tensor<string, []>("matmul_7_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> matmul_7_transpose_x_0 = const()[name = tensor<string, []>("matmul_7_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 50]> matmul_7_cast_fp16 = matmul(transpose_x = matmul_7_transpose_x_0, transpose_y = matmul_7_transpose_y_0, x = mul_15_cast_fp16, y = k_47_cast_fp16)[name = tensor<string, []>("matmul_7_cast_fp16")]; | |
| tensor<int32, []> softmax_7_axis_0 = const()[name = tensor<string, []>("softmax_7_axis_0"), val = tensor<int32, []>(-1)]; | |
| tensor<fp16, [?, 12, 50, 50]> softmax_7_cast_fp16 = softmax(axis = softmax_7_axis_0, x = matmul_7_cast_fp16)[name = tensor<string, []>("softmax_7_cast_fp16")]; | |
| tensor<bool, []> attn_output_43_transpose_x_0 = const()[name = tensor<string, []>("attn_output_43_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_43_transpose_y_0 = const()[name = tensor<string, []>("attn_output_43_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 64]> attn_output_43_cast_fp16 = matmul(transpose_x = attn_output_43_transpose_x_0, transpose_y = attn_output_43_transpose_y_0, x = softmax_7_cast_fp16, y = v_47_cast_fp16)[name = tensor<string, []>("attn_output_43_cast_fp16")]; | |
| tensor<int32, [4]> var_921 = const()[name = tensor<string, []>("op_921"), val = tensor<int32, [4]>([2, 0, 1, 3])]; | |
| tensor<int32, [2]> concat_72x = const()[name = tensor<string, []>("concat_72x"), val = tensor<int32, [2]>([-1, 768])]; | |
| tensor<fp16, [50, ?, 12, 64]> var_922_cast_fp16 = transpose(perm = var_921, x = attn_output_43_cast_fp16)[name = tensor<string, []>("transpose_22")]; | |
| tensor<fp16, [?, 768]> attn_output_45_cast_fp16 = reshape(shape = concat_72x, x = var_922_cast_fp16)[name = tensor<string, []>("attn_output_45_cast_fp16")]; | |
| tensor<fp16, [768, 768]> transformer_resblocks_7_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_7_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107582848)))]; | |
| tensor<fp16, [768]> transformer_resblocks_7_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_7_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108762560)))]; | |
| tensor<fp16, [?, 768]> linear_29_cast_fp16 = linear(bias = transformer_resblocks_7_attn_out_proj_bias_to_fp16, weight = transformer_resblocks_7_attn_out_proj_weight_to_fp16, x = attn_output_45_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")]; | |
| tensor<int32, [3]> concat_73x = const()[name = tensor<string, []>("concat_73x"), val = tensor<int32, [3]>([50, -1, 768])]; | |
| tensor<fp16, [50, ?, 768]> var_931_cast_fp16 = reshape(shape = concat_73x, x = linear_29_cast_fp16)[name = tensor<string, []>("op_931_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_59_cast_fp16 = add(x = x_57_cast_fp16, y = var_931_cast_fp16)[name = tensor<string, []>("x_59_cast_fp16")]; | |
| tensor<int32, [1]> ret_33_axes_0 = const()[name = tensor<string, []>("ret_33_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_7_ln_2_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_7_ln_2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108764160)))]; | |
| tensor<fp16, [768]> transformer_resblocks_7_ln_2_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_7_ln_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108765760)))]; | |
| tensor<fp16, [50, ?, 768]> ret_33_cast_fp16 = layer_norm(axes = ret_33_axes_0, beta = transformer_resblocks_7_ln_2_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_7_ln_2_weight_to_fp16, x = x_59_cast_fp16)[name = tensor<string, []>("ret_33_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> transformer_resblocks_7_mlp_c_fc_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_7_mlp_c_fc_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108767360)))]; | |
| tensor<fp16, [3072]> transformer_resblocks_7_mlp_c_fc_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_7_mlp_c_fc_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113486016)))]; | |
| tensor<fp16, [50, ?, 3072]> linear_30_cast_fp16 = linear(bias = transformer_resblocks_7_mlp_c_fc_bias_to_fp16, weight = transformer_resblocks_7_mlp_c_fc_weight_to_fp16, x = ret_33_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")]; | |
| tensor<fp16, []> var_947_to_fp16 = const()[name = tensor<string, []>("op_947_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [50, ?, 3072]> var_948_cast_fp16 = mul(x = linear_30_cast_fp16, y = var_947_to_fp16)[name = tensor<string, []>("op_948_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> var_949_cast_fp16 = sigmoid(x = var_948_cast_fp16)[name = tensor<string, []>("op_949_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> input_65_cast_fp16 = mul(x = linear_30_cast_fp16, y = var_949_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> transformer_resblocks_7_mlp_c_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_7_mlp_c_proj_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113492224)))]; | |
| tensor<fp16, [768]> transformer_resblocks_7_mlp_c_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_7_mlp_c_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118210880)))]; | |
| tensor<fp16, [50, ?, 768]> linear_31_cast_fp16 = linear(bias = transformer_resblocks_7_mlp_c_proj_bias_to_fp16, weight = transformer_resblocks_7_mlp_c_proj_weight_to_fp16, x = input_65_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_63_cast_fp16 = add(x = x_59_cast_fp16, y = linear_31_cast_fp16)[name = tensor<string, []>("x_63_cast_fp16")]; | |
| tensor<int32, [1]> ret_35_axes_0 = const()[name = tensor<string, []>("ret_35_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_8_ln_1_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_8_ln_1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118212480)))]; | |
| tensor<fp16, [768]> transformer_resblocks_8_ln_1_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_8_ln_1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118214080)))]; | |
| tensor<fp16, [50, ?, 768]> ret_35_cast_fp16 = layer_norm(axes = ret_35_axes_0, beta = transformer_resblocks_8_ln_1_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_8_ln_1_weight_to_fp16, x = x_63_cast_fp16)[name = tensor<string, []>("ret_35_cast_fp16")]; | |
| tensor<fp16, [2304, 768]> transformer_resblocks_8_attn_in_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_8_attn_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118215680)))]; | |
| tensor<fp16, [2304]> transformer_resblocks_8_attn_in_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_8_attn_in_proj_bias_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121754688)))]; | |
| tensor<fp16, [50, ?, 2304]> linear_32_cast_fp16 = linear(bias = transformer_resblocks_8_attn_in_proj_bias_to_fp16, weight = transformer_resblocks_8_attn_in_proj_weight_to_fp16, x = ret_35_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")]; | |
| tensor<int32, [4]> concat_74x = const()[name = tensor<string, []>("concat_74x"), val = tensor<int32, [4]>([50, -1, 3, 768])]; | |
| tensor<fp16, [50, ?, 3, 768]> var_990_cast_fp16 = reshape(shape = concat_74x, x = linear_32_cast_fp16)[name = tensor<string, []>("op_990_cast_fp16")]; | |
| tensor<int32, [1]> var_991_axes_0 = const()[name = tensor<string, []>("op_991_axes_0"), val = tensor<int32, [1]>([0])]; | |
| tensor<fp16, [1, 50, ?, 3, 768]> var_991_cast_fp16 = expand_dims(axes = var_991_axes_0, x = var_990_cast_fp16)[name = tensor<string, []>("op_991_cast_fp16")]; | |
| tensor<int32, [5]> var_992_perm_0 = const()[name = tensor<string, []>("op_992_perm_0"), val = tensor<int32, [5]>([-2, 1, 2, 0, 4])]; | |
| tensor<int32, [1]> var_993_axes_0 = const()[name = tensor<string, []>("op_993_axes_0"), val = tensor<int32, [1]>([-2])]; | |
| tensor<fp16, [3, 50, ?, 1, 768]> var_992_cast_fp16 = transpose(perm = var_992_perm_0, x = var_991_cast_fp16)[name = tensor<string, []>("transpose_21")]; | |
| tensor<fp16, [3, 50, ?, 768]> var_993_cast_fp16 = squeeze(axes = var_993_axes_0, x = var_992_cast_fp16)[name = tensor<string, []>("op_993_cast_fp16")]; | |
| tensor<int32, [4]> q_49_begin_0 = const()[name = tensor<string, []>("q_49_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> q_49_end_0 = const()[name = tensor<string, []>("q_49_end_0"), val = tensor<int32, [4]>([1, 50, 0, 768])]; | |
| tensor<bool, [4]> q_49_end_mask_0 = const()[name = tensor<string, []>("q_49_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> q_49_squeeze_mask_0 = const()[name = tensor<string, []>("q_49_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> q_49_cast_fp16 = slice_by_index(begin = q_49_begin_0, end = q_49_end_0, end_mask = q_49_end_mask_0, squeeze_mask = q_49_squeeze_mask_0, x = var_993_cast_fp16)[name = tensor<string, []>("q_49_cast_fp16")]; | |
| tensor<int32, [4]> k_49_begin_0 = const()[name = tensor<string, []>("k_49_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])]; | |
| tensor<int32, [4]> k_49_end_0 = const()[name = tensor<string, []>("k_49_end_0"), val = tensor<int32, [4]>([2, 50, 0, 768])]; | |
| tensor<bool, [4]> k_49_end_mask_0 = const()[name = tensor<string, []>("k_49_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> k_49_squeeze_mask_0 = const()[name = tensor<string, []>("k_49_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> k_49_cast_fp16 = slice_by_index(begin = k_49_begin_0, end = k_49_end_0, end_mask = k_49_end_mask_0, squeeze_mask = k_49_squeeze_mask_0, x = var_993_cast_fp16)[name = tensor<string, []>("k_49_cast_fp16")]; | |
| tensor<int32, [4]> v_49_begin_0 = const()[name = tensor<string, []>("v_49_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])]; | |
| tensor<int32, [4]> v_49_end_0 = const()[name = tensor<string, []>("v_49_end_0"), val = tensor<int32, [4]>([3, 50, 0, 768])]; | |
| tensor<bool, [4]> v_49_end_mask_0 = const()[name = tensor<string, []>("v_49_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> v_49_squeeze_mask_0 = const()[name = tensor<string, []>("v_49_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> v_49_cast_fp16 = slice_by_index(begin = v_49_begin_0, end = v_49_end_0, end_mask = v_49_end_mask_0, squeeze_mask = v_49_squeeze_mask_0, x = var_993_cast_fp16)[name = tensor<string, []>("v_49_cast_fp16")]; | |
| tensor<int32, [3]> concat_75x = const()[name = tensor<string, []>("concat_75x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_1002_cast_fp16 = reshape(shape = concat_75x, x = q_49_cast_fp16)[name = tensor<string, []>("op_1002_cast_fp16")]; | |
| tensor<int32, [3]> q_51_perm_0 = const()[name = tensor<string, []>("q_51_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_76x = const()[name = tensor<string, []>("concat_76x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_1009_cast_fp16 = reshape(shape = concat_76x, x = k_49_cast_fp16)[name = tensor<string, []>("op_1009_cast_fp16")]; | |
| tensor<int32, [3]> k_51_perm_0 = const()[name = tensor<string, []>("k_51_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_77x = const()[name = tensor<string, []>("concat_77x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_1016_cast_fp16 = reshape(shape = concat_77x, x = v_49_cast_fp16)[name = tensor<string, []>("op_1016_cast_fp16")]; | |
| tensor<int32, [3]> v_51_perm_0 = const()[name = tensor<string, []>("v_51_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [4]> concat_78x = const()[name = tensor<string, []>("concat_78x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> q_51_cast_fp16 = transpose(perm = q_51_perm_0, x = var_1002_cast_fp16)[name = tensor<string, []>("transpose_20")]; | |
| tensor<fp16, [?, 12, 50, 64]> q_53_cast_fp16 = reshape(shape = concat_78x, x = q_51_cast_fp16)[name = tensor<string, []>("q_53_cast_fp16")]; | |
| tensor<int32, [4]> concat_79x = const()[name = tensor<string, []>("concat_79x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> k_51_cast_fp16 = transpose(perm = k_51_perm_0, x = var_1009_cast_fp16)[name = tensor<string, []>("transpose_19")]; | |
| tensor<fp16, [?, 12, 50, 64]> k_53_cast_fp16 = reshape(shape = concat_79x, x = k_51_cast_fp16)[name = tensor<string, []>("k_53_cast_fp16")]; | |
| tensor<int32, [4]> concat_80x = const()[name = tensor<string, []>("concat_80x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> v_51_cast_fp16 = transpose(perm = v_51_perm_0, x = var_1016_cast_fp16)[name = tensor<string, []>("transpose_18")]; | |
| tensor<fp16, [?, 12, 50, 64]> v_53_cast_fp16 = reshape(shape = concat_80x, x = v_51_cast_fp16)[name = tensor<string, []>("v_53_cast_fp16")]; | |
| tensor<fp16, []> mul_17_y_0_to_fp16 = const()[name = tensor<string, []>("mul_17_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [?, 12, 50, 64]> mul_17_cast_fp16 = mul(x = q_53_cast_fp16, y = mul_17_y_0_to_fp16)[name = tensor<string, []>("mul_17_cast_fp16")]; | |
| tensor<bool, []> matmul_8_transpose_y_0 = const()[name = tensor<string, []>("matmul_8_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> matmul_8_transpose_x_0 = const()[name = tensor<string, []>("matmul_8_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 50]> matmul_8_cast_fp16 = matmul(transpose_x = matmul_8_transpose_x_0, transpose_y = matmul_8_transpose_y_0, x = mul_17_cast_fp16, y = k_53_cast_fp16)[name = tensor<string, []>("matmul_8_cast_fp16")]; | |
| tensor<int32, []> softmax_8_axis_0 = const()[name = tensor<string, []>("softmax_8_axis_0"), val = tensor<int32, []>(-1)]; | |
| tensor<fp16, [?, 12, 50, 50]> softmax_8_cast_fp16 = softmax(axis = softmax_8_axis_0, x = matmul_8_cast_fp16)[name = tensor<string, []>("softmax_8_cast_fp16")]; | |
| tensor<bool, []> attn_output_49_transpose_x_0 = const()[name = tensor<string, []>("attn_output_49_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_49_transpose_y_0 = const()[name = tensor<string, []>("attn_output_49_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 64]> attn_output_49_cast_fp16 = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = softmax_8_cast_fp16, y = v_53_cast_fp16)[name = tensor<string, []>("attn_output_49_cast_fp16")]; | |
| tensor<int32, [4]> var_1026 = const()[name = tensor<string, []>("op_1026"), val = tensor<int32, [4]>([2, 0, 1, 3])]; | |
| tensor<int32, [2]> concat_81x = const()[name = tensor<string, []>("concat_81x"), val = tensor<int32, [2]>([-1, 768])]; | |
| tensor<fp16, [50, ?, 12, 64]> var_1027_cast_fp16 = transpose(perm = var_1026, x = attn_output_49_cast_fp16)[name = tensor<string, []>("transpose_17")]; | |
| tensor<fp16, [?, 768]> attn_output_51_cast_fp16 = reshape(shape = concat_81x, x = var_1027_cast_fp16)[name = tensor<string, []>("attn_output_51_cast_fp16")]; | |
| tensor<fp16, [768, 768]> transformer_resblocks_8_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_8_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121759360)))]; | |
| tensor<fp16, [768]> transformer_resblocks_8_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_8_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122939072)))]; | |
| tensor<fp16, [?, 768]> linear_33_cast_fp16 = linear(bias = transformer_resblocks_8_attn_out_proj_bias_to_fp16, weight = transformer_resblocks_8_attn_out_proj_weight_to_fp16, x = attn_output_51_cast_fp16)[name = tensor<string, []>("linear_33_cast_fp16")]; | |
| tensor<int32, [3]> concat_82x = const()[name = tensor<string, []>("concat_82x"), val = tensor<int32, [3]>([50, -1, 768])]; | |
| tensor<fp16, [50, ?, 768]> var_1036_cast_fp16 = reshape(shape = concat_82x, x = linear_33_cast_fp16)[name = tensor<string, []>("op_1036_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_65_cast_fp16 = add(x = x_63_cast_fp16, y = var_1036_cast_fp16)[name = tensor<string, []>("x_65_cast_fp16")]; | |
| tensor<int32, [1]> ret_37_axes_0 = const()[name = tensor<string, []>("ret_37_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_8_ln_2_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_8_ln_2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122940672)))]; | |
| tensor<fp16, [768]> transformer_resblocks_8_ln_2_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_8_ln_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122942272)))]; | |
| tensor<fp16, [50, ?, 768]> ret_37_cast_fp16 = layer_norm(axes = ret_37_axes_0, beta = transformer_resblocks_8_ln_2_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_8_ln_2_weight_to_fp16, x = x_65_cast_fp16)[name = tensor<string, []>("ret_37_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> transformer_resblocks_8_mlp_c_fc_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_8_mlp_c_fc_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122943872)))]; | |
| tensor<fp16, [3072]> transformer_resblocks_8_mlp_c_fc_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_8_mlp_c_fc_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127662528)))]; | |
| tensor<fp16, [50, ?, 3072]> linear_34_cast_fp16 = linear(bias = transformer_resblocks_8_mlp_c_fc_bias_to_fp16, weight = transformer_resblocks_8_mlp_c_fc_weight_to_fp16, x = ret_37_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")]; | |
| tensor<fp16, []> var_1052_to_fp16 = const()[name = tensor<string, []>("op_1052_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [50, ?, 3072]> var_1053_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_1052_to_fp16)[name = tensor<string, []>("op_1053_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> var_1054_cast_fp16 = sigmoid(x = var_1053_cast_fp16)[name = tensor<string, []>("op_1054_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> input_73_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_1054_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> transformer_resblocks_8_mlp_c_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_8_mlp_c_proj_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127668736)))]; | |
| tensor<fp16, [768]> transformer_resblocks_8_mlp_c_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_8_mlp_c_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132387392)))]; | |
| tensor<fp16, [50, ?, 768]> linear_35_cast_fp16 = linear(bias = transformer_resblocks_8_mlp_c_proj_bias_to_fp16, weight = transformer_resblocks_8_mlp_c_proj_weight_to_fp16, x = input_73_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_69_cast_fp16 = add(x = x_65_cast_fp16, y = linear_35_cast_fp16)[name = tensor<string, []>("x_69_cast_fp16")]; | |
| tensor<int32, [1]> ret_39_axes_0 = const()[name = tensor<string, []>("ret_39_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_9_ln_1_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_9_ln_1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132388992)))]; | |
| tensor<fp16, [768]> transformer_resblocks_9_ln_1_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_9_ln_1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132390592)))]; | |
| tensor<fp16, [50, ?, 768]> ret_39_cast_fp16 = layer_norm(axes = ret_39_axes_0, beta = transformer_resblocks_9_ln_1_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_9_ln_1_weight_to_fp16, x = x_69_cast_fp16)[name = tensor<string, []>("ret_39_cast_fp16")]; | |
| tensor<fp16, [2304, 768]> transformer_resblocks_9_attn_in_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_9_attn_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132392192)))]; | |
| tensor<fp16, [2304]> transformer_resblocks_9_attn_in_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_9_attn_in_proj_bias_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135931200)))]; | |
| tensor<fp16, [50, ?, 2304]> linear_36_cast_fp16 = linear(bias = transformer_resblocks_9_attn_in_proj_bias_to_fp16, weight = transformer_resblocks_9_attn_in_proj_weight_to_fp16, x = ret_39_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")]; | |
| tensor<int32, [4]> concat_83x = const()[name = tensor<string, []>("concat_83x"), val = tensor<int32, [4]>([50, -1, 3, 768])]; | |
| tensor<fp16, [50, ?, 3, 768]> var_1095_cast_fp16 = reshape(shape = concat_83x, x = linear_36_cast_fp16)[name = tensor<string, []>("op_1095_cast_fp16")]; | |
| tensor<int32, [1]> var_1096_axes_0 = const()[name = tensor<string, []>("op_1096_axes_0"), val = tensor<int32, [1]>([0])]; | |
| tensor<fp16, [1, 50, ?, 3, 768]> var_1096_cast_fp16 = expand_dims(axes = var_1096_axes_0, x = var_1095_cast_fp16)[name = tensor<string, []>("op_1096_cast_fp16")]; | |
| tensor<int32, [5]> var_1097_perm_0 = const()[name = tensor<string, []>("op_1097_perm_0"), val = tensor<int32, [5]>([-2, 1, 2, 0, 4])]; | |
| tensor<int32, [1]> var_1098_axes_0 = const()[name = tensor<string, []>("op_1098_axes_0"), val = tensor<int32, [1]>([-2])]; | |
| tensor<fp16, [3, 50, ?, 1, 768]> var_1097_cast_fp16 = transpose(perm = var_1097_perm_0, x = var_1096_cast_fp16)[name = tensor<string, []>("transpose_16")]; | |
| tensor<fp16, [3, 50, ?, 768]> var_1098_cast_fp16 = squeeze(axes = var_1098_axes_0, x = var_1097_cast_fp16)[name = tensor<string, []>("op_1098_cast_fp16")]; | |
| tensor<int32, [4]> q_55_begin_0 = const()[name = tensor<string, []>("q_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> q_55_end_0 = const()[name = tensor<string, []>("q_55_end_0"), val = tensor<int32, [4]>([1, 50, 0, 768])]; | |
| tensor<bool, [4]> q_55_end_mask_0 = const()[name = tensor<string, []>("q_55_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> q_55_squeeze_mask_0 = const()[name = tensor<string, []>("q_55_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> q_55_cast_fp16 = slice_by_index(begin = q_55_begin_0, end = q_55_end_0, end_mask = q_55_end_mask_0, squeeze_mask = q_55_squeeze_mask_0, x = var_1098_cast_fp16)[name = tensor<string, []>("q_55_cast_fp16")]; | |
| tensor<int32, [4]> k_55_begin_0 = const()[name = tensor<string, []>("k_55_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])]; | |
| tensor<int32, [4]> k_55_end_0 = const()[name = tensor<string, []>("k_55_end_0"), val = tensor<int32, [4]>([2, 50, 0, 768])]; | |
| tensor<bool, [4]> k_55_end_mask_0 = const()[name = tensor<string, []>("k_55_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> k_55_squeeze_mask_0 = const()[name = tensor<string, []>("k_55_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> k_55_cast_fp16 = slice_by_index(begin = k_55_begin_0, end = k_55_end_0, end_mask = k_55_end_mask_0, squeeze_mask = k_55_squeeze_mask_0, x = var_1098_cast_fp16)[name = tensor<string, []>("k_55_cast_fp16")]; | |
| tensor<int32, [4]> v_55_begin_0 = const()[name = tensor<string, []>("v_55_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])]; | |
| tensor<int32, [4]> v_55_end_0 = const()[name = tensor<string, []>("v_55_end_0"), val = tensor<int32, [4]>([3, 50, 0, 768])]; | |
| tensor<bool, [4]> v_55_end_mask_0 = const()[name = tensor<string, []>("v_55_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> v_55_squeeze_mask_0 = const()[name = tensor<string, []>("v_55_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> v_55_cast_fp16 = slice_by_index(begin = v_55_begin_0, end = v_55_end_0, end_mask = v_55_end_mask_0, squeeze_mask = v_55_squeeze_mask_0, x = var_1098_cast_fp16)[name = tensor<string, []>("v_55_cast_fp16")]; | |
| tensor<int32, [3]> concat_84x = const()[name = tensor<string, []>("concat_84x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_1107_cast_fp16 = reshape(shape = concat_84x, x = q_55_cast_fp16)[name = tensor<string, []>("op_1107_cast_fp16")]; | |
| tensor<int32, [3]> q_57_perm_0 = const()[name = tensor<string, []>("q_57_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_85x = const()[name = tensor<string, []>("concat_85x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_1114_cast_fp16 = reshape(shape = concat_85x, x = k_55_cast_fp16)[name = tensor<string, []>("op_1114_cast_fp16")]; | |
| tensor<int32, [3]> k_57_perm_0 = const()[name = tensor<string, []>("k_57_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_86x = const()[name = tensor<string, []>("concat_86x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_1121_cast_fp16 = reshape(shape = concat_86x, x = v_55_cast_fp16)[name = tensor<string, []>("op_1121_cast_fp16")]; | |
| tensor<int32, [3]> v_57_perm_0 = const()[name = tensor<string, []>("v_57_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [4]> concat_87x = const()[name = tensor<string, []>("concat_87x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> q_57_cast_fp16 = transpose(perm = q_57_perm_0, x = var_1107_cast_fp16)[name = tensor<string, []>("transpose_15")]; | |
| tensor<fp16, [?, 12, 50, 64]> q_59_cast_fp16 = reshape(shape = concat_87x, x = q_57_cast_fp16)[name = tensor<string, []>("q_59_cast_fp16")]; | |
| tensor<int32, [4]> concat_88x = const()[name = tensor<string, []>("concat_88x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> k_57_cast_fp16 = transpose(perm = k_57_perm_0, x = var_1114_cast_fp16)[name = tensor<string, []>("transpose_14")]; | |
| tensor<fp16, [?, 12, 50, 64]> k_59_cast_fp16 = reshape(shape = concat_88x, x = k_57_cast_fp16)[name = tensor<string, []>("k_59_cast_fp16")]; | |
| tensor<int32, [4]> concat_89x = const()[name = tensor<string, []>("concat_89x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> v_57_cast_fp16 = transpose(perm = v_57_perm_0, x = var_1121_cast_fp16)[name = tensor<string, []>("transpose_13")]; | |
| tensor<fp16, [?, 12, 50, 64]> v_59_cast_fp16 = reshape(shape = concat_89x, x = v_57_cast_fp16)[name = tensor<string, []>("v_59_cast_fp16")]; | |
| tensor<fp16, []> mul_19_y_0_to_fp16 = const()[name = tensor<string, []>("mul_19_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [?, 12, 50, 64]> mul_19_cast_fp16 = mul(x = q_59_cast_fp16, y = mul_19_y_0_to_fp16)[name = tensor<string, []>("mul_19_cast_fp16")]; | |
| tensor<bool, []> matmul_9_transpose_y_0 = const()[name = tensor<string, []>("matmul_9_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> matmul_9_transpose_x_0 = const()[name = tensor<string, []>("matmul_9_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 50]> matmul_9_cast_fp16 = matmul(transpose_x = matmul_9_transpose_x_0, transpose_y = matmul_9_transpose_y_0, x = mul_19_cast_fp16, y = k_59_cast_fp16)[name = tensor<string, []>("matmul_9_cast_fp16")]; | |
| tensor<int32, []> softmax_9_axis_0 = const()[name = tensor<string, []>("softmax_9_axis_0"), val = tensor<int32, []>(-1)]; | |
| tensor<fp16, [?, 12, 50, 50]> softmax_9_cast_fp16 = softmax(axis = softmax_9_axis_0, x = matmul_9_cast_fp16)[name = tensor<string, []>("softmax_9_cast_fp16")]; | |
| tensor<bool, []> attn_output_55_transpose_x_0 = const()[name = tensor<string, []>("attn_output_55_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_55_transpose_y_0 = const()[name = tensor<string, []>("attn_output_55_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 64]> attn_output_55_cast_fp16 = matmul(transpose_x = attn_output_55_transpose_x_0, transpose_y = attn_output_55_transpose_y_0, x = softmax_9_cast_fp16, y = v_59_cast_fp16)[name = tensor<string, []>("attn_output_55_cast_fp16")]; | |
| tensor<int32, [4]> var_1131 = const()[name = tensor<string, []>("op_1131"), val = tensor<int32, [4]>([2, 0, 1, 3])]; | |
| tensor<int32, [2]> concat_90x = const()[name = tensor<string, []>("concat_90x"), val = tensor<int32, [2]>([-1, 768])]; | |
| tensor<fp16, [50, ?, 12, 64]> var_1132_cast_fp16 = transpose(perm = var_1131, x = attn_output_55_cast_fp16)[name = tensor<string, []>("transpose_12")]; | |
| tensor<fp16, [?, 768]> attn_output_57_cast_fp16 = reshape(shape = concat_90x, x = var_1132_cast_fp16)[name = tensor<string, []>("attn_output_57_cast_fp16")]; | |
| tensor<fp16, [768, 768]> transformer_resblocks_9_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_9_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135935872)))]; | |
| tensor<fp16, [768]> transformer_resblocks_9_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_9_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137115584)))]; | |
| tensor<fp16, [?, 768]> linear_37_cast_fp16 = linear(bias = transformer_resblocks_9_attn_out_proj_bias_to_fp16, weight = transformer_resblocks_9_attn_out_proj_weight_to_fp16, x = attn_output_57_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")]; | |
| tensor<int32, [3]> concat_91x = const()[name = tensor<string, []>("concat_91x"), val = tensor<int32, [3]>([50, -1, 768])]; | |
| tensor<fp16, [50, ?, 768]> var_1141_cast_fp16 = reshape(shape = concat_91x, x = linear_37_cast_fp16)[name = tensor<string, []>("op_1141_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_71_cast_fp16 = add(x = x_69_cast_fp16, y = var_1141_cast_fp16)[name = tensor<string, []>("x_71_cast_fp16")]; | |
| tensor<int32, [1]> ret_41_axes_0 = const()[name = tensor<string, []>("ret_41_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_9_ln_2_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_9_ln_2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137117184)))]; | |
| tensor<fp16, [768]> transformer_resblocks_9_ln_2_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_9_ln_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137118784)))]; | |
| tensor<fp16, [50, ?, 768]> ret_41_cast_fp16 = layer_norm(axes = ret_41_axes_0, beta = transformer_resblocks_9_ln_2_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_9_ln_2_weight_to_fp16, x = x_71_cast_fp16)[name = tensor<string, []>("ret_41_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> transformer_resblocks_9_mlp_c_fc_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_9_mlp_c_fc_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137120384)))]; | |
| tensor<fp16, [3072]> transformer_resblocks_9_mlp_c_fc_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_9_mlp_c_fc_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141839040)))]; | |
| tensor<fp16, [50, ?, 3072]> linear_38_cast_fp16 = linear(bias = transformer_resblocks_9_mlp_c_fc_bias_to_fp16, weight = transformer_resblocks_9_mlp_c_fc_weight_to_fp16, x = ret_41_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")]; | |
| tensor<fp16, []> var_1157_to_fp16 = const()[name = tensor<string, []>("op_1157_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [50, ?, 3072]> var_1158_cast_fp16 = mul(x = linear_38_cast_fp16, y = var_1157_to_fp16)[name = tensor<string, []>("op_1158_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> var_1159_cast_fp16 = sigmoid(x = var_1158_cast_fp16)[name = tensor<string, []>("op_1159_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> input_81_cast_fp16 = mul(x = linear_38_cast_fp16, y = var_1159_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> transformer_resblocks_9_mlp_c_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_9_mlp_c_proj_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141845248)))]; | |
| tensor<fp16, [768]> transformer_resblocks_9_mlp_c_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_9_mlp_c_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146563904)))]; | |
| tensor<fp16, [50, ?, 768]> linear_39_cast_fp16 = linear(bias = transformer_resblocks_9_mlp_c_proj_bias_to_fp16, weight = transformer_resblocks_9_mlp_c_proj_weight_to_fp16, x = input_81_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_75_cast_fp16 = add(x = x_71_cast_fp16, y = linear_39_cast_fp16)[name = tensor<string, []>("x_75_cast_fp16")]; | |
| tensor<int32, [1]> ret_43_axes_0 = const()[name = tensor<string, []>("ret_43_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_10_ln_1_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_10_ln_1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146565504)))]; | |
| tensor<fp16, [768]> transformer_resblocks_10_ln_1_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_10_ln_1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146567104)))]; | |
| tensor<fp16, [50, ?, 768]> ret_43_cast_fp16 = layer_norm(axes = ret_43_axes_0, beta = transformer_resblocks_10_ln_1_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_10_ln_1_weight_to_fp16, x = x_75_cast_fp16)[name = tensor<string, []>("ret_43_cast_fp16")]; | |
| tensor<fp16, [2304, 768]> transformer_resblocks_10_attn_in_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_10_attn_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146568704)))]; | |
| tensor<fp16, [2304]> transformer_resblocks_10_attn_in_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_10_attn_in_proj_bias_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150107712)))]; | |
| tensor<fp16, [50, ?, 2304]> linear_40_cast_fp16 = linear(bias = transformer_resblocks_10_attn_in_proj_bias_to_fp16, weight = transformer_resblocks_10_attn_in_proj_weight_to_fp16, x = ret_43_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")]; | |
| tensor<int32, [4]> concat_92x = const()[name = tensor<string, []>("concat_92x"), val = tensor<int32, [4]>([50, -1, 3, 768])]; | |
| tensor<fp16, [50, ?, 3, 768]> var_1200_cast_fp16 = reshape(shape = concat_92x, x = linear_40_cast_fp16)[name = tensor<string, []>("op_1200_cast_fp16")]; | |
| tensor<int32, [1]> var_1201_axes_0 = const()[name = tensor<string, []>("op_1201_axes_0"), val = tensor<int32, [1]>([0])]; | |
| tensor<fp16, [1, 50, ?, 3, 768]> var_1201_cast_fp16 = expand_dims(axes = var_1201_axes_0, x = var_1200_cast_fp16)[name = tensor<string, []>("op_1201_cast_fp16")]; | |
| tensor<int32, [5]> var_1202_perm_0 = const()[name = tensor<string, []>("op_1202_perm_0"), val = tensor<int32, [5]>([-2, 1, 2, 0, 4])]; | |
| tensor<int32, [1]> var_1203_axes_0 = const()[name = tensor<string, []>("op_1203_axes_0"), val = tensor<int32, [1]>([-2])]; | |
| tensor<fp16, [3, 50, ?, 1, 768]> var_1202_cast_fp16 = transpose(perm = var_1202_perm_0, x = var_1201_cast_fp16)[name = tensor<string, []>("transpose_11")]; | |
| tensor<fp16, [3, 50, ?, 768]> var_1203_cast_fp16 = squeeze(axes = var_1203_axes_0, x = var_1202_cast_fp16)[name = tensor<string, []>("op_1203_cast_fp16")]; | |
| tensor<int32, [4]> q_61_begin_0 = const()[name = tensor<string, []>("q_61_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> q_61_end_0 = const()[name = tensor<string, []>("q_61_end_0"), val = tensor<int32, [4]>([1, 50, 0, 768])]; | |
| tensor<bool, [4]> q_61_end_mask_0 = const()[name = tensor<string, []>("q_61_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> q_61_squeeze_mask_0 = const()[name = tensor<string, []>("q_61_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> q_61_cast_fp16 = slice_by_index(begin = q_61_begin_0, end = q_61_end_0, end_mask = q_61_end_mask_0, squeeze_mask = q_61_squeeze_mask_0, x = var_1203_cast_fp16)[name = tensor<string, []>("q_61_cast_fp16")]; | |
| tensor<int32, [4]> k_61_begin_0 = const()[name = tensor<string, []>("k_61_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])]; | |
| tensor<int32, [4]> k_61_end_0 = const()[name = tensor<string, []>("k_61_end_0"), val = tensor<int32, [4]>([2, 50, 0, 768])]; | |
| tensor<bool, [4]> k_61_end_mask_0 = const()[name = tensor<string, []>("k_61_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> k_61_squeeze_mask_0 = const()[name = tensor<string, []>("k_61_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> k_61_cast_fp16 = slice_by_index(begin = k_61_begin_0, end = k_61_end_0, end_mask = k_61_end_mask_0, squeeze_mask = k_61_squeeze_mask_0, x = var_1203_cast_fp16)[name = tensor<string, []>("k_61_cast_fp16")]; | |
| tensor<int32, [4]> v_61_begin_0 = const()[name = tensor<string, []>("v_61_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])]; | |
| tensor<int32, [4]> v_61_end_0 = const()[name = tensor<string, []>("v_61_end_0"), val = tensor<int32, [4]>([3, 50, 0, 768])]; | |
| tensor<bool, [4]> v_61_end_mask_0 = const()[name = tensor<string, []>("v_61_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> v_61_squeeze_mask_0 = const()[name = tensor<string, []>("v_61_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> v_61_cast_fp16 = slice_by_index(begin = v_61_begin_0, end = v_61_end_0, end_mask = v_61_end_mask_0, squeeze_mask = v_61_squeeze_mask_0, x = var_1203_cast_fp16)[name = tensor<string, []>("v_61_cast_fp16")]; | |
| tensor<int32, [3]> concat_93x = const()[name = tensor<string, []>("concat_93x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_1212_cast_fp16 = reshape(shape = concat_93x, x = q_61_cast_fp16)[name = tensor<string, []>("op_1212_cast_fp16")]; | |
| tensor<int32, [3]> q_63_perm_0 = const()[name = tensor<string, []>("q_63_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_94x = const()[name = tensor<string, []>("concat_94x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_1219_cast_fp16 = reshape(shape = concat_94x, x = k_61_cast_fp16)[name = tensor<string, []>("op_1219_cast_fp16")]; | |
| tensor<int32, [3]> k_63_perm_0 = const()[name = tensor<string, []>("k_63_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_95x = const()[name = tensor<string, []>("concat_95x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_1226_cast_fp16 = reshape(shape = concat_95x, x = v_61_cast_fp16)[name = tensor<string, []>("op_1226_cast_fp16")]; | |
| tensor<int32, [3]> v_63_perm_0 = const()[name = tensor<string, []>("v_63_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [4]> concat_96x = const()[name = tensor<string, []>("concat_96x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> q_63_cast_fp16 = transpose(perm = q_63_perm_0, x = var_1212_cast_fp16)[name = tensor<string, []>("transpose_10")]; | |
| tensor<fp16, [?, 12, 50, 64]> q_65_cast_fp16 = reshape(shape = concat_96x, x = q_63_cast_fp16)[name = tensor<string, []>("q_65_cast_fp16")]; | |
| tensor<int32, [4]> concat_97x = const()[name = tensor<string, []>("concat_97x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> k_63_cast_fp16 = transpose(perm = k_63_perm_0, x = var_1219_cast_fp16)[name = tensor<string, []>("transpose_9")]; | |
| tensor<fp16, [?, 12, 50, 64]> k_65_cast_fp16 = reshape(shape = concat_97x, x = k_63_cast_fp16)[name = tensor<string, []>("k_65_cast_fp16")]; | |
| tensor<int32, [4]> concat_98x = const()[name = tensor<string, []>("concat_98x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> v_63_cast_fp16 = transpose(perm = v_63_perm_0, x = var_1226_cast_fp16)[name = tensor<string, []>("transpose_8")]; | |
| tensor<fp16, [?, 12, 50, 64]> v_65_cast_fp16 = reshape(shape = concat_98x, x = v_63_cast_fp16)[name = tensor<string, []>("v_65_cast_fp16")]; | |
| tensor<fp16, []> mul_21_y_0_to_fp16 = const()[name = tensor<string, []>("mul_21_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [?, 12, 50, 64]> mul_21_cast_fp16 = mul(x = q_65_cast_fp16, y = mul_21_y_0_to_fp16)[name = tensor<string, []>("mul_21_cast_fp16")]; | |
| tensor<bool, []> matmul_10_transpose_y_0 = const()[name = tensor<string, []>("matmul_10_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> matmul_10_transpose_x_0 = const()[name = tensor<string, []>("matmul_10_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 50]> matmul_10_cast_fp16 = matmul(transpose_x = matmul_10_transpose_x_0, transpose_y = matmul_10_transpose_y_0, x = mul_21_cast_fp16, y = k_65_cast_fp16)[name = tensor<string, []>("matmul_10_cast_fp16")]; | |
| tensor<int32, []> softmax_10_axis_0 = const()[name = tensor<string, []>("softmax_10_axis_0"), val = tensor<int32, []>(-1)]; | |
| tensor<fp16, [?, 12, 50, 50]> softmax_10_cast_fp16 = softmax(axis = softmax_10_axis_0, x = matmul_10_cast_fp16)[name = tensor<string, []>("softmax_10_cast_fp16")]; | |
| tensor<bool, []> attn_output_61_transpose_x_0 = const()[name = tensor<string, []>("attn_output_61_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_61_transpose_y_0 = const()[name = tensor<string, []>("attn_output_61_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 64]> attn_output_61_cast_fp16 = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = softmax_10_cast_fp16, y = v_65_cast_fp16)[name = tensor<string, []>("attn_output_61_cast_fp16")]; | |
| tensor<int32, [4]> var_1236 = const()[name = tensor<string, []>("op_1236"), val = tensor<int32, [4]>([2, 0, 1, 3])]; | |
| tensor<int32, [2]> concat_99x = const()[name = tensor<string, []>("concat_99x"), val = tensor<int32, [2]>([-1, 768])]; | |
| tensor<fp16, [50, ?, 12, 64]> var_1237_cast_fp16 = transpose(perm = var_1236, x = attn_output_61_cast_fp16)[name = tensor<string, []>("transpose_7")]; | |
| tensor<fp16, [?, 768]> attn_output_63_cast_fp16 = reshape(shape = concat_99x, x = var_1237_cast_fp16)[name = tensor<string, []>("attn_output_63_cast_fp16")]; | |
| tensor<fp16, [768, 768]> transformer_resblocks_10_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_10_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150112384)))]; | |
| tensor<fp16, [768]> transformer_resblocks_10_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_10_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151292096)))]; | |
| tensor<fp16, [?, 768]> linear_41_cast_fp16 = linear(bias = transformer_resblocks_10_attn_out_proj_bias_to_fp16, weight = transformer_resblocks_10_attn_out_proj_weight_to_fp16, x = attn_output_63_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")]; | |
| tensor<int32, [3]> concat_100x = const()[name = tensor<string, []>("concat_100x"), val = tensor<int32, [3]>([50, -1, 768])]; | |
| tensor<fp16, [50, ?, 768]> var_1246_cast_fp16 = reshape(shape = concat_100x, x = linear_41_cast_fp16)[name = tensor<string, []>("op_1246_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_77_cast_fp16 = add(x = x_75_cast_fp16, y = var_1246_cast_fp16)[name = tensor<string, []>("x_77_cast_fp16")]; | |
| tensor<int32, [1]> ret_45_axes_0 = const()[name = tensor<string, []>("ret_45_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_10_ln_2_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_10_ln_2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151293696)))]; | |
| tensor<fp16, [768]> transformer_resblocks_10_ln_2_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_10_ln_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151295296)))]; | |
| tensor<fp16, [50, ?, 768]> ret_45_cast_fp16 = layer_norm(axes = ret_45_axes_0, beta = transformer_resblocks_10_ln_2_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_10_ln_2_weight_to_fp16, x = x_77_cast_fp16)[name = tensor<string, []>("ret_45_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> transformer_resblocks_10_mlp_c_fc_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_10_mlp_c_fc_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151296896)))]; | |
| tensor<fp16, [3072]> transformer_resblocks_10_mlp_c_fc_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_10_mlp_c_fc_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156015552)))]; | |
| tensor<fp16, [50, ?, 3072]> linear_42_cast_fp16 = linear(bias = transformer_resblocks_10_mlp_c_fc_bias_to_fp16, weight = transformer_resblocks_10_mlp_c_fc_weight_to_fp16, x = ret_45_cast_fp16)[name = tensor<string, []>("linear_42_cast_fp16")]; | |
| tensor<fp16, []> var_1262_to_fp16 = const()[name = tensor<string, []>("op_1262_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [50, ?, 3072]> var_1263_cast_fp16 = mul(x = linear_42_cast_fp16, y = var_1262_to_fp16)[name = tensor<string, []>("op_1263_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> var_1264_cast_fp16 = sigmoid(x = var_1263_cast_fp16)[name = tensor<string, []>("op_1264_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> input_89_cast_fp16 = mul(x = linear_42_cast_fp16, y = var_1264_cast_fp16)[name = tensor<string, []>("input_89_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> transformer_resblocks_10_mlp_c_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_10_mlp_c_proj_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156021760)))]; | |
| tensor<fp16, [768]> transformer_resblocks_10_mlp_c_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_10_mlp_c_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160740416)))]; | |
| tensor<fp16, [50, ?, 768]> linear_43_cast_fp16 = linear(bias = transformer_resblocks_10_mlp_c_proj_bias_to_fp16, weight = transformer_resblocks_10_mlp_c_proj_weight_to_fp16, x = input_89_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_81_cast_fp16 = add(x = x_77_cast_fp16, y = linear_43_cast_fp16)[name = tensor<string, []>("x_81_cast_fp16")]; | |
| tensor<int32, [1]> ret_47_axes_0 = const()[name = tensor<string, []>("ret_47_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_11_ln_1_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_11_ln_1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160742016)))]; | |
| tensor<fp16, [768]> transformer_resblocks_11_ln_1_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_11_ln_1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160743616)))]; | |
| tensor<fp16, [50, ?, 768]> ret_47_cast_fp16 = layer_norm(axes = ret_47_axes_0, beta = transformer_resblocks_11_ln_1_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_11_ln_1_weight_to_fp16, x = x_81_cast_fp16)[name = tensor<string, []>("ret_47_cast_fp16")]; | |
| tensor<fp16, [2304, 768]> transformer_resblocks_11_attn_in_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_11_attn_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160745216)))]; | |
| tensor<fp16, [2304]> transformer_resblocks_11_attn_in_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_11_attn_in_proj_bias_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164284224)))]; | |
| tensor<fp16, [50, ?, 2304]> linear_44_cast_fp16 = linear(bias = transformer_resblocks_11_attn_in_proj_bias_to_fp16, weight = transformer_resblocks_11_attn_in_proj_weight_to_fp16, x = ret_47_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")]; | |
| tensor<int32, [4]> concat_101x = const()[name = tensor<string, []>("concat_101x"), val = tensor<int32, [4]>([50, -1, 3, 768])]; | |
| tensor<fp16, [50, ?, 3, 768]> var_1305_cast_fp16 = reshape(shape = concat_101x, x = linear_44_cast_fp16)[name = tensor<string, []>("op_1305_cast_fp16")]; | |
| tensor<int32, [1]> var_1306_axes_0 = const()[name = tensor<string, []>("op_1306_axes_0"), val = tensor<int32, [1]>([0])]; | |
| tensor<fp16, [1, 50, ?, 3, 768]> var_1306_cast_fp16 = expand_dims(axes = var_1306_axes_0, x = var_1305_cast_fp16)[name = tensor<string, []>("op_1306_cast_fp16")]; | |
| tensor<int32, [5]> var_1307_perm_0 = const()[name = tensor<string, []>("op_1307_perm_0"), val = tensor<int32, [5]>([-2, 1, 2, 0, 4])]; | |
| tensor<int32, [1]> var_1308_axes_0 = const()[name = tensor<string, []>("op_1308_axes_0"), val = tensor<int32, [1]>([-2])]; | |
| tensor<fp16, [3, 50, ?, 1, 768]> var_1307_cast_fp16 = transpose(perm = var_1307_perm_0, x = var_1306_cast_fp16)[name = tensor<string, []>("transpose_6")]; | |
| tensor<fp16, [3, 50, ?, 768]> var_1308_cast_fp16 = squeeze(axes = var_1308_axes_0, x = var_1307_cast_fp16)[name = tensor<string, []>("op_1308_cast_fp16")]; | |
| tensor<int32, [4]> q_67_begin_0 = const()[name = tensor<string, []>("q_67_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> q_67_end_0 = const()[name = tensor<string, []>("q_67_end_0"), val = tensor<int32, [4]>([1, 50, 0, 768])]; | |
| tensor<bool, [4]> q_67_end_mask_0 = const()[name = tensor<string, []>("q_67_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> q_67_squeeze_mask_0 = const()[name = tensor<string, []>("q_67_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> q_67_cast_fp16 = slice_by_index(begin = q_67_begin_0, end = q_67_end_0, end_mask = q_67_end_mask_0, squeeze_mask = q_67_squeeze_mask_0, x = var_1308_cast_fp16)[name = tensor<string, []>("q_67_cast_fp16")]; | |
| tensor<int32, [4]> k_67_begin_0 = const()[name = tensor<string, []>("k_67_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])]; | |
| tensor<int32, [4]> k_67_end_0 = const()[name = tensor<string, []>("k_67_end_0"), val = tensor<int32, [4]>([2, 50, 0, 768])]; | |
| tensor<bool, [4]> k_67_end_mask_0 = const()[name = tensor<string, []>("k_67_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> k_67_squeeze_mask_0 = const()[name = tensor<string, []>("k_67_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> k_67_cast_fp16 = slice_by_index(begin = k_67_begin_0, end = k_67_end_0, end_mask = k_67_end_mask_0, squeeze_mask = k_67_squeeze_mask_0, x = var_1308_cast_fp16)[name = tensor<string, []>("k_67_cast_fp16")]; | |
| tensor<int32, [4]> v_67_begin_0 = const()[name = tensor<string, []>("v_67_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])]; | |
| tensor<int32, [4]> v_67_end_0 = const()[name = tensor<string, []>("v_67_end_0"), val = tensor<int32, [4]>([3, 50, 0, 768])]; | |
| tensor<bool, [4]> v_67_end_mask_0 = const()[name = tensor<string, []>("v_67_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<bool, [4]> v_67_squeeze_mask_0 = const()[name = tensor<string, []>("v_67_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])]; | |
| tensor<fp16, [50, ?, 768]> v_67_cast_fp16 = slice_by_index(begin = v_67_begin_0, end = v_67_end_0, end_mask = v_67_end_mask_0, squeeze_mask = v_67_squeeze_mask_0, x = var_1308_cast_fp16)[name = tensor<string, []>("v_67_cast_fp16")]; | |
| tensor<int32, [3]> concat_102x = const()[name = tensor<string, []>("concat_102x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_1317_cast_fp16 = reshape(shape = concat_102x, x = q_67_cast_fp16)[name = tensor<string, []>("op_1317_cast_fp16")]; | |
| tensor<int32, [3]> q_69_perm_0 = const()[name = tensor<string, []>("q_69_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_103x = const()[name = tensor<string, []>("concat_103x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_1324_cast_fp16 = reshape(shape = concat_103x, x = k_67_cast_fp16)[name = tensor<string, []>("op_1324_cast_fp16")]; | |
| tensor<int32, [3]> k_69_perm_0 = const()[name = tensor<string, []>("k_69_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> concat_104x = const()[name = tensor<string, []>("concat_104x"), val = tensor<int32, [3]>([50, -1, 64])]; | |
| tensor<fp16, [50, ?, 64]> var_1331_cast_fp16 = reshape(shape = concat_104x, x = v_67_cast_fp16)[name = tensor<string, []>("op_1331_cast_fp16")]; | |
| tensor<int32, [3]> v_69_perm_0 = const()[name = tensor<string, []>("v_69_perm_0"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [4]> concat_105x = const()[name = tensor<string, []>("concat_105x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> q_69_cast_fp16 = transpose(perm = q_69_perm_0, x = var_1317_cast_fp16)[name = tensor<string, []>("transpose_5")]; | |
| tensor<fp16, [?, 12, 50, 64]> q_cast_fp16 = reshape(shape = concat_105x, x = q_69_cast_fp16)[name = tensor<string, []>("q_cast_fp16")]; | |
| tensor<int32, [4]> concat_106x = const()[name = tensor<string, []>("concat_106x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> k_69_cast_fp16 = transpose(perm = k_69_perm_0, x = var_1324_cast_fp16)[name = tensor<string, []>("transpose_4")]; | |
| tensor<fp16, [?, 12, 50, 64]> k_cast_fp16 = reshape(shape = concat_106x, x = k_69_cast_fp16)[name = tensor<string, []>("k_cast_fp16")]; | |
| tensor<int32, [4]> concat_107x = const()[name = tensor<string, []>("concat_107x"), val = tensor<int32, [4]>([-1, 12, 50, 64])]; | |
| tensor<fp16, [?, 50, 64]> v_69_cast_fp16 = transpose(perm = v_69_perm_0, x = var_1331_cast_fp16)[name = tensor<string, []>("transpose_3")]; | |
| tensor<fp16, [?, 12, 50, 64]> v_cast_fp16 = reshape(shape = concat_107x, x = v_69_cast_fp16)[name = tensor<string, []>("v_cast_fp16")]; | |
| tensor<fp16, []> mul_23_y_0_to_fp16 = const()[name = tensor<string, []>("mul_23_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [?, 12, 50, 64]> mul_23_cast_fp16 = mul(x = q_cast_fp16, y = mul_23_y_0_to_fp16)[name = tensor<string, []>("mul_23_cast_fp16")]; | |
| tensor<bool, []> matmul_11_transpose_y_0 = const()[name = tensor<string, []>("matmul_11_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> matmul_11_transpose_x_0 = const()[name = tensor<string, []>("matmul_11_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 50]> matmul_11_cast_fp16 = matmul(transpose_x = matmul_11_transpose_x_0, transpose_y = matmul_11_transpose_y_0, x = mul_23_cast_fp16, y = k_cast_fp16)[name = tensor<string, []>("matmul_11_cast_fp16")]; | |
| tensor<int32, []> softmax_11_axis_0 = const()[name = tensor<string, []>("softmax_11_axis_0"), val = tensor<int32, []>(-1)]; | |
| tensor<fp16, [?, 12, 50, 50]> softmax_11_cast_fp16 = softmax(axis = softmax_11_axis_0, x = matmul_11_cast_fp16)[name = tensor<string, []>("softmax_11_cast_fp16")]; | |
| tensor<bool, []> attn_output_67_transpose_x_0 = const()[name = tensor<string, []>("attn_output_67_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_67_transpose_y_0 = const()[name = tensor<string, []>("attn_output_67_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [?, 12, 50, 64]> attn_output_67_cast_fp16 = matmul(transpose_x = attn_output_67_transpose_x_0, transpose_y = attn_output_67_transpose_y_0, x = softmax_11_cast_fp16, y = v_cast_fp16)[name = tensor<string, []>("attn_output_67_cast_fp16")]; | |
| tensor<int32, [4]> var_1341 = const()[name = tensor<string, []>("op_1341"), val = tensor<int32, [4]>([2, 0, 1, 3])]; | |
| tensor<int32, [2]> concat_108x = const()[name = tensor<string, []>("concat_108x"), val = tensor<int32, [2]>([-1, 768])]; | |
| tensor<fp16, [50, ?, 12, 64]> var_1342_cast_fp16 = transpose(perm = var_1341, x = attn_output_67_cast_fp16)[name = tensor<string, []>("transpose_2")]; | |
| tensor<fp16, [?, 768]> attn_output_69_cast_fp16 = reshape(shape = concat_108x, x = var_1342_cast_fp16)[name = tensor<string, []>("attn_output_69_cast_fp16")]; | |
| tensor<fp16, [768, 768]> transformer_resblocks_11_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_11_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164288896)))]; | |
| tensor<fp16, [768]> transformer_resblocks_11_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_11_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165468608)))]; | |
| tensor<fp16, [?, 768]> linear_45_cast_fp16 = linear(bias = transformer_resblocks_11_attn_out_proj_bias_to_fp16, weight = transformer_resblocks_11_attn_out_proj_weight_to_fp16, x = attn_output_69_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")]; | |
| tensor<int32, [3]> concat_109x = const()[name = tensor<string, []>("concat_109x"), val = tensor<int32, [3]>([50, -1, 768])]; | |
| tensor<fp16, [50, ?, 768]> var_1351_cast_fp16 = reshape(shape = concat_109x, x = linear_45_cast_fp16)[name = tensor<string, []>("op_1351_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_83_cast_fp16 = add(x = x_81_cast_fp16, y = var_1351_cast_fp16)[name = tensor<string, []>("x_83_cast_fp16")]; | |
| tensor<int32, [1]> ret_49_axes_0 = const()[name = tensor<string, []>("ret_49_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> transformer_resblocks_11_ln_2_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_11_ln_2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165470208)))]; | |
| tensor<fp16, [768]> transformer_resblocks_11_ln_2_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_11_ln_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165471808)))]; | |
| tensor<fp16, [50, ?, 768]> ret_49_cast_fp16 = layer_norm(axes = ret_49_axes_0, beta = transformer_resblocks_11_ln_2_bias_to_fp16, epsilon = var_100_to_fp16, gamma = transformer_resblocks_11_ln_2_weight_to_fp16, x = x_83_cast_fp16)[name = tensor<string, []>("ret_49_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> transformer_resblocks_11_mlp_c_fc_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_11_mlp_c_fc_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165473408)))]; | |
| tensor<fp16, [3072]> transformer_resblocks_11_mlp_c_fc_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_11_mlp_c_fc_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170192064)))]; | |
| tensor<fp16, [50, ?, 3072]> linear_46_cast_fp16 = linear(bias = transformer_resblocks_11_mlp_c_fc_bias_to_fp16, weight = transformer_resblocks_11_mlp_c_fc_weight_to_fp16, x = ret_49_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")]; | |
| tensor<fp16, []> var_1367_to_fp16 = const()[name = tensor<string, []>("op_1367_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [50, ?, 3072]> var_1368_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_1367_to_fp16)[name = tensor<string, []>("op_1368_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> var_1369_cast_fp16 = sigmoid(x = var_1368_cast_fp16)[name = tensor<string, []>("op_1369_cast_fp16")]; | |
| tensor<fp16, [50, ?, 3072]> input_97_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_1369_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> transformer_resblocks_11_mlp_c_proj_weight_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_11_mlp_c_proj_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170198272)))]; | |
| tensor<fp16, [768]> transformer_resblocks_11_mlp_c_proj_bias_to_fp16 = const()[name = tensor<string, []>("transformer_resblocks_11_mlp_c_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174916928)))]; | |
| tensor<fp16, [50, ?, 768]> linear_47_cast_fp16 = linear(bias = transformer_resblocks_11_mlp_c_proj_bias_to_fp16, weight = transformer_resblocks_11_mlp_c_proj_weight_to_fp16, x = input_97_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")]; | |
| tensor<fp16, [50, ?, 768]> x_87_cast_fp16 = add(x = x_83_cast_fp16, y = linear_47_cast_fp16)[name = tensor<string, []>("x_87_cast_fp16")]; | |
| tensor<int32, [3]> var_1378 = const()[name = tensor<string, []>("op_1378"), val = tensor<int32, [3]>([1, 0, 2])]; | |
| tensor<int32, [3]> var_1387_begin_0 = const()[name = tensor<string, []>("op_1387_begin_0"), val = tensor<int32, [3]>([0, 0, 0])]; | |
| tensor<int32, [3]> var_1387_end_0 = const()[name = tensor<string, []>("op_1387_end_0"), val = tensor<int32, [3]>([0, 1, 768])]; | |
| tensor<bool, [3]> var_1387_end_mask_0 = const()[name = tensor<string, []>("op_1387_end_mask_0"), val = tensor<bool, [3]>([true, false, true])]; | |
| tensor<bool, [3]> var_1387_squeeze_mask_0 = const()[name = tensor<string, []>("op_1387_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])]; | |
| tensor<fp16, [?, 50, 768]> x_89_cast_fp16 = transpose(perm = var_1378, x = x_87_cast_fp16)[name = tensor<string, []>("transpose_1")]; | |
| tensor<fp16, [?, 768]> var_1387_cast_fp16 = slice_by_index(begin = var_1387_begin_0, end = var_1387_end_0, end_mask = var_1387_end_mask_0, squeeze_mask = var_1387_squeeze_mask_0, x = x_89_cast_fp16)[name = tensor<string, []>("op_1387_cast_fp16")]; | |
| tensor<int32, [1]> ret_axes_0 = const()[name = tensor<string, []>("ret_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> ln_post_weight_to_fp16 = const()[name = tensor<string, []>("ln_post_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174918528)))]; | |
| tensor<fp16, [768]> ln_post_bias_to_fp16 = const()[name = tensor<string, []>("ln_post_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174920128)))]; | |
| tensor<fp16, []> var_1394_to_fp16 = const()[name = tensor<string, []>("op_1394_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [?, 768]> ret_cast_fp16 = layer_norm(axes = ret_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_1394_to_fp16, gamma = ln_post_weight_to_fp16, x = var_1387_cast_fp16)[name = tensor<string, []>("ret_cast_fp16")]; | |
| tensor<fp16, [512, 768]> transpose_0_to_fp16 = const()[name = tensor<string, []>("transpose_0_to_fp16"), val = tensor<fp16, [512, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174921728)))]; | |
| tensor<fp16, [512]> var_1405_bias_0_to_fp16 = const()[name = tensor<string, []>("op_1405_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175708224)))]; | |
| tensor<fp16, [?, 512]> var_1405_cast_fp16 = linear(bias = var_1405_bias_0_to_fp16, weight = transpose_0_to_fp16, x = ret_cast_fp16)[name = tensor<string, []>("op_1405_cast_fp16")]; | |
| tensor<string, []> var_1405_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_1405_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; | |
| tensor<fp32, [?, 512]> embOutput = cast(dtype = var_1405_cast_fp16_to_fp32_dtype_0, x = var_1405_cast_fp16)[name = tensor<string, []>("cast_185")]; | |
| } -> (embOutput); | |
| } |