| program(1.0) |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})] |
| { |
| func main<ios17>(tensor<fp32, [1, 128, ?]> mel) [FlexibleShapeInformation = tuple<tuple<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>, tuple<tensor<string, []>, dict<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>>>((("DefaultShapes", {{"mel", [1, 128, 100]}}), ("EnumeratedShapes", {{"mel_1_1_1_128_1000_", {{"mel", [1, 128, 1000]}}}, {"mel_1_1_1_128_100_", {{"mel", [1, 128, 100]}}}, {"mel_1_1_1_128_1500_", {{"mel", [1, 128, 1500]}}}, {"mel_1_1_1_128_2000_", {{"mel", [1, 128, 2000]}}}, {"mel_1_1_1_128_200_", {{"mel", [1, 128, 200]}}}, {"mel_1_1_1_128_3000_", {{"mel", [1, 128, 3000]}}}, {"mel_1_1_1_128_400_", {{"mel", [1, 128, 400]}}}, {"mel_1_1_1_128_600_", {{"mel", [1, 128, 600]}}}, {"mel_1_1_1_128_800_", {{"mel", [1, 128, 800]}}}})))] { |
| tensor<int32, [1]> input_1_axes_0 = const()[name = tensor<string, []>("input_1_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<string, []> mel_to_fp16_dtype_0 = const()[name = tensor<string, []>("mel_to_fp16_dtype_0"), val = tensor<string, []>("fp16")]; |
| tensor<fp16, [1, 128, ?]> mel_to_fp16 = cast(dtype = mel_to_fp16_dtype_0, x = mel)[name = tensor<string, []>("cast_2")]; |
| tensor<fp16, [1, 1, 128, ?]> input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = mel_to_fp16)[name = tensor<string, []>("input_1_cast_fp16")]; |
| tensor<string, []> var_71_pad_type_0 = const()[name = tensor<string, []>("op_71_pad_type_0"), val = tensor<string, []>("custom")]; |
| tensor<int32, [4]> var_71_pad_0 = const()[name = tensor<string, []>("op_71_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])]; |
| tensor<int32, [2]> var_71_strides_0 = const()[name = tensor<string, []>("op_71_strides_0"), val = tensor<int32, [2]>([2, 2])]; |
| tensor<int32, [2]> var_71_dilations_0 = const()[name = tensor<string, []>("op_71_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_71_groups_0 = const()[name = tensor<string, []>("op_71_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [480, 1, 3, 3]> conv2d1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2160]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2304))), name = tensor<string, []>("conv2d1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([480, 1, 3, 3])]; |
| tensor<fp16, [480]> conv2d1_bias_to_fp16 = const()[name = tensor<string, []>("conv2d1_bias_to_fp16"), val = tensor<fp16, [480]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2432)))]; |
| tensor<fp16, [1, 480, 64, ?]> var_71_cast_fp16 = conv(bias = conv2d1_bias_to_fp16, dilations = var_71_dilations_0, groups = var_71_groups_0, pad = var_71_pad_0, pad_type = var_71_pad_type_0, strides = var_71_strides_0, weight = conv2d1_weight_to_fp16_palettized, x = input_1_cast_fp16)[name = tensor<string, []>("op_71_cast_fp16")]; |
| tensor<string, []> input_3_mode_0 = const()[name = tensor<string, []>("input_3_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 480, 64, ?]> input_3_cast_fp16 = gelu(mode = input_3_mode_0, x = var_71_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")]; |
| tensor<string, []> var_85_pad_type_0 = const()[name = tensor<string, []>("op_85_pad_type_0"), val = tensor<string, []>("custom")]; |
| tensor<int32, [4]> var_85_pad_0 = const()[name = tensor<string, []>("op_85_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])]; |
| tensor<int32, [2]> var_85_strides_0 = const()[name = tensor<string, []>("op_85_strides_0"), val = tensor<int32, [2]>([2, 2])]; |
| tensor<int32, [2]> var_85_dilations_0 = const()[name = tensor<string, []>("op_85_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_85_groups_0 = const()[name = tensor<string, []>("op_85_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [480, 480, 3, 3]> conv2d2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1036800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3456))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1040320))), name = tensor<string, []>("conv2d2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([480, 480, 3, 3])]; |
| tensor<fp16, [480]> conv2d2_bias_to_fp16 = const()[name = tensor<string, []>("conv2d2_bias_to_fp16"), val = tensor<fp16, [480]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1040448)))]; |
| tensor<fp16, [1, 480, 32, ?]> var_85_cast_fp16 = conv(bias = conv2d2_bias_to_fp16, dilations = var_85_dilations_0, groups = var_85_groups_0, pad = var_85_pad_0, pad_type = var_85_pad_type_0, strides = var_85_strides_0, weight = conv2d2_weight_to_fp16_palettized, x = input_3_cast_fp16)[name = tensor<string, []>("op_85_cast_fp16")]; |
| tensor<string, []> input_5_mode_0 = const()[name = tensor<string, []>("input_5_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 480, 32, ?]> input_5_cast_fp16 = gelu(mode = input_5_mode_0, x = var_85_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")]; |
| tensor<string, []> var_99_pad_type_0 = const()[name = tensor<string, []>("op_99_pad_type_0"), val = tensor<string, []>("custom")]; |
| tensor<int32, [4]> var_99_pad_0 = const()[name = tensor<string, []>("op_99_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])]; |
| tensor<int32, [2]> var_99_strides_0 = const()[name = tensor<string, []>("op_99_strides_0"), val = tensor<int32, [2]>([2, 2])]; |
| tensor<int32, [2]> var_99_dilations_0 = const()[name = tensor<string, []>("op_99_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_99_groups_0 = const()[name = tensor<string, []>("op_99_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [480, 480, 3, 3]> conv2d3_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1036800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1041472))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2078336))), name = tensor<string, []>("conv2d3_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([480, 480, 3, 3])]; |
| tensor<fp16, [480]> conv2d3_bias_to_fp16 = const()[name = tensor<string, []>("conv2d3_bias_to_fp16"), val = tensor<fp16, [480]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2078464)))]; |
| tensor<fp16, [1, 480, 16, ?]> var_99_cast_fp16 = conv(bias = conv2d3_bias_to_fp16, dilations = var_99_dilations_0, groups = var_99_groups_0, pad = var_99_pad_0, pad_type = var_99_pad_type_0, strides = var_99_strides_0, weight = conv2d3_weight_to_fp16_palettized, x = input_5_cast_fp16)[name = tensor<string, []>("op_99_cast_fp16")]; |
| tensor<string, []> x_1_mode_0 = const()[name = tensor<string, []>("x_1_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 480, 16, ?]> x_1_cast_fp16 = gelu(mode = x_1_mode_0, x = var_99_cast_fp16)[name = tensor<string, []>("x_1_cast_fp16")]; |
| tensor<int32, [4]> var_120 = const()[name = tensor<string, []>("op_120"), val = tensor<int32, [4]>([0, 3, 1, 2])]; |
| tensor<int32, [3]> concat_0x = const()[name = tensor<string, []>("concat_0x"), val = tensor<int32, [3]>([1, -1, 7680])]; |
| tensor<fp16, [1, ?, 480, 16]> var_121_cast_fp16 = transpose(perm = var_120, x = x_1_cast_fp16)[name = tensor<string, []>("transpose_240")]; |
| tensor<fp16, [1, ?, 7680]> input_7_cast_fp16 = reshape(shape = concat_0x, x = var_121_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")]; |
| tensor<fp16, [1024, 7680]> conv_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3932160]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2079488))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6011712))), name = tensor<string, []>("conv_out_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 7680])]; |
| tensor<fp16, [1024]> linear_0_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_0_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6011840)))]; |
| tensor<fp16, [1, ?, 1024]> linear_0_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = conv_out_weight_to_fp16_palettized, x = input_7_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")]; |
| tensor<int32, [3]> var_130_shape_cast_fp16 = shape(x = linear_0_cast_fp16)[name = tensor<string, []>("op_130_shape_cast_fp16")]; |
| tensor<int32, []> gather_4_axis_0 = const()[name = tensor<string, []>("gather_4_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_4_batch_dims_0 = const()[name = tensor<string, []>("gather_4_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> gather_4_validate_indices_0 = const()[name = tensor<string, []>("gather_4_validate_indices_0"), val = tensor<bool, []>(false)]; |
| tensor<string, []> var_130_shape_cast_fp16_to_uint16_dtype_0 = const()[name = tensor<string, []>("op_130_shape_cast_fp16_to_uint16_dtype_0"), val = tensor<string, []>("uint16")]; |
| tensor<uint16, []> select_4_to_uint16 = const()[name = tensor<string, []>("select_4_to_uint16"), val = tensor<uint16, []>(1)]; |
| tensor<uint16, [3]> var_130_shape_cast_fp16_to_uint16 = cast(dtype = var_130_shape_cast_fp16_to_uint16_dtype_0, x = var_130_shape_cast_fp16)[name = tensor<string, []>("cast_1")]; |
| tensor<uint16, []> gather_4_cast_uint16 = gather(axis = gather_4_axis_0, batch_dims = gather_4_batch_dims_0, indices = select_4_to_uint16, validate_indices = gather_4_validate_indices_0, x = var_130_shape_cast_fp16_to_uint16)[name = tensor<string, []>("gather_4_cast_uint16")]; |
| tensor<string, []> gather_4_cast_uint16_to_int32_dtype_0 = const()[name = tensor<string, []>("gather_4_cast_uint16_to_int32_dtype_0"), val = tensor<string, []>("int32")]; |
| tensor<int32, []> concat_1_values0_0 = const()[name = tensor<string, []>("concat_1_values0_0"), val = tensor<int32, []>(1)]; |
| tensor<int32, []> concat_1_values2_0 = const()[name = tensor<string, []>("concat_1_values2_0"), val = tensor<int32, []>(1024)]; |
| 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, []> gather_4_cast_uint16_to_int32 = cast(dtype = gather_4_cast_uint16_to_int32_dtype_0, x = gather_4_cast_uint16)[name = tensor<string, []>("cast_0")]; |
| tensor<int32, [3]> concat_1 = concat(axis = concat_1_axis_0, interleave = concat_1_interleave_0, values = (concat_1_values0_0, gather_4_cast_uint16_to_int32, concat_1_values2_0))[name = tensor<string, []>("concat_1")]; |
| tensor<int32, [3]> var_141_begin_0 = const()[name = tensor<string, []>("op_141_begin_0"), val = tensor<int32, [3]>([0, 0, 0])]; |
| tensor<bool, [3]> var_141_end_mask_0 = const()[name = tensor<string, []>("op_141_end_mask_0"), val = tensor<bool, [3]>([true, false, true])]; |
| tensor<fp16, [1, 1500, 1024]> pos_embed_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [768000]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6013952))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6782016))), name = tensor<string, []>("pos_embed_to_fp16_palettized"), shape = tensor<uint32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, ?, 1024]> var_141_cast_fp16 = slice_by_index(begin = var_141_begin_0, end = concat_1, end_mask = var_141_end_mask_0, x = pos_embed_to_fp16_palettized)[name = tensor<string, []>("op_141_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_9_cast_fp16 = add(x = linear_0_cast_fp16, y = var_141_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")]; |
| tensor<int32, []> var_156 = const()[name = tensor<string, []>("op_156"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_5_axes_0 = const()[name = tensor<string, []>("x_5_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_0_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6782144)))]; |
| tensor<fp16, [1024]> layers_0_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6784256)))]; |
| tensor<fp16, []> var_159_to_fp16 = const()[name = tensor<string, []>("op_159_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_5_cast_fp16 = layer_norm(axes = x_5_axes_0, beta = layers_0_self_attn_layer_norm_bias_to_fp16, epsilon = var_159_to_fp16, gamma = layers_0_self_attn_layer_norm_weight_to_fp16, x = input_9_cast_fp16)[name = tensor<string, []>("x_5_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6786368))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7310720))), name = tensor<string, []>("layers_0_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7310848)))]; |
| tensor<fp16, [1, ?, 1024]> linear_1_cast_fp16 = linear(bias = layers_0_self_attn_q_proj_bias_to_fp16, weight = layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = x_5_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")]; |
| tensor<int32, [4]> concat_2x = const()[name = tensor<string, []>("concat_2x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_180_cast_fp16 = reshape(shape = concat_2x, x = linear_1_cast_fp16)[name = tensor<string, []>("op_180_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7312960))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7837312))), name = tensor<string, []>("layers_0_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7837440)))]; |
| tensor<fp16, [1, ?, 1024]> linear_2_cast_fp16 = linear(bias = layers_0_self_attn_k_proj_bias_to_fp16, weight = layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = x_5_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")]; |
| tensor<int32, [4]> concat_3x = const()[name = tensor<string, []>("concat_3x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_186_cast_fp16 = reshape(shape = concat_3x, x = linear_2_cast_fp16)[name = tensor<string, []>("op_186_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7839552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8363904))), name = tensor<string, []>("layers_0_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8364032)))]; |
| tensor<fp16, [1, ?, 1024]> linear_3_cast_fp16 = linear(bias = layers_0_self_attn_v_proj_bias_to_fp16, weight = layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = x_5_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")]; |
| tensor<int32, [4]> concat_4x = const()[name = tensor<string, []>("concat_4x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_192_cast_fp16 = reshape(shape = concat_4x, x = linear_3_cast_fp16)[name = tensor<string, []>("op_192_cast_fp16")]; |
| tensor<int32, [4]> v_1_perm_0 = const()[name = tensor<string, []>("v_1_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_195_transpose_x_0 = const()[name = tensor<string, []>("op_195_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_195_transpose_y_0 = const()[name = tensor<string, []>("op_195_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_96_perm_0 = const()[name = tensor<string, []>("transpose_96_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_97_perm_0 = const()[name = tensor<string, []>("transpose_97_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_97 = transpose(perm = transpose_97_perm_0, x = var_186_cast_fp16)[name = tensor<string, []>("transpose_238")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_96 = transpose(perm = transpose_96_perm_0, x = var_180_cast_fp16)[name = tensor<string, []>("transpose_239")]; |
| tensor<fp16, [1, 16, ?, ?]> var_195_cast_fp16 = matmul(transpose_x = var_195_transpose_x_0, transpose_y = var_195_transpose_y_0, x = transpose_96, y = transpose_97)[name = tensor<string, []>("op_195_cast_fp16")]; |
| tensor<fp16, []> var_196_to_fp16 = const()[name = tensor<string, []>("op_196_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_11_cast_fp16 = mul(x = var_195_cast_fp16, y = var_196_to_fp16)[name = tensor<string, []>("input_11_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_1_cast_fp16 = softmax(axis = var_156, x = input_11_cast_fp16)[name = tensor<string, []>("attn_1_cast_fp16")]; |
| tensor<bool, []> out_1_transpose_x_0 = const()[name = tensor<string, []>("out_1_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_1_transpose_y_0 = const()[name = tensor<string, []>("out_1_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_1_cast_fp16 = transpose(perm = v_1_perm_0, x = var_192_cast_fp16)[name = tensor<string, []>("transpose_237")]; |
| tensor<fp16, [1, 16, ?, 64]> out_1_cast_fp16 = matmul(transpose_x = out_1_transpose_x_0, transpose_y = out_1_transpose_y_0, x = attn_1_cast_fp16, y = v_1_cast_fp16)[name = tensor<string, []>("out_1_cast_fp16")]; |
| tensor<int32, [4]> var_200_perm_0 = const()[name = tensor<string, []>("op_200_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_5x = const()[name = tensor<string, []>("concat_5x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_200_cast_fp16 = transpose(perm = var_200_perm_0, x = out_1_cast_fp16)[name = tensor<string, []>("transpose_236")]; |
| tensor<fp16, [1, ?, 1024]> input_13_cast_fp16 = reshape(shape = concat_5x, x = var_200_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_0_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8366144))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8890496))), name = tensor<string, []>("layers_0_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8890624)))]; |
| tensor<fp16, [1, ?, 1024]> linear_4_cast_fp16 = linear(bias = layers_0_self_attn_out_proj_bias_to_fp16, weight = layers_0_self_attn_out_proj_weight_to_fp16_palettized, x = input_13_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_15_cast_fp16 = add(x = input_9_cast_fp16, y = linear_4_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")]; |
| tensor<int32, [1]> input_17_axes_0 = const()[name = tensor<string, []>("input_17_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_0_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8892736)))]; |
| tensor<fp16, [1024]> layers_0_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8894848)))]; |
| tensor<fp16, [1, ?, 1024]> input_17_cast_fp16 = layer_norm(axes = input_17_axes_0, beta = layers_0_final_layer_norm_bias_to_fp16, epsilon = var_159_to_fp16, gamma = layers_0_final_layer_norm_weight_to_fp16, x = input_15_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_0_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8896960))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10994176))), name = tensor<string, []>("layers_0_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_0_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10994304)))]; |
| tensor<fp16, [1, ?, 4096]> linear_5_cast_fp16 = linear(bias = layers_0_fc1_bias_to_fp16, weight = layers_0_fc1_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")]; |
| tensor<string, []> input_19_mode_0 = const()[name = tensor<string, []>("input_19_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = linear_5_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_0_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11002560))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13099776))), name = tensor<string, []>("layers_0_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_0_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13099904)))]; |
| tensor<fp16, [1, ?, 1024]> linear_6_cast_fp16 = linear(bias = layers_0_fc2_bias_to_fp16, weight = layers_0_fc2_weight_to_fp16_palettized, x = input_19_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_21_cast_fp16 = add(x = input_15_cast_fp16, y = linear_6_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")]; |
| tensor<int32, []> var_226 = const()[name = tensor<string, []>("op_226"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_11_axes_0 = const()[name = tensor<string, []>("x_11_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_1_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13102016)))]; |
| tensor<fp16, [1024]> layers_1_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13104128)))]; |
| tensor<fp16, []> var_229_to_fp16 = const()[name = tensor<string, []>("op_229_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_11_cast_fp16 = layer_norm(axes = x_11_axes_0, beta = layers_1_self_attn_layer_norm_bias_to_fp16, epsilon = var_229_to_fp16, gamma = layers_1_self_attn_layer_norm_weight_to_fp16, x = input_21_cast_fp16)[name = tensor<string, []>("x_11_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13106240))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13630592))), name = tensor<string, []>("layers_1_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13630720)))]; |
| tensor<fp16, [1, ?, 1024]> linear_7_cast_fp16 = linear(bias = layers_1_self_attn_q_proj_bias_to_fp16, weight = layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = x_11_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")]; |
| tensor<int32, [4]> concat_6x = const()[name = tensor<string, []>("concat_6x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_250_cast_fp16 = reshape(shape = concat_6x, x = linear_7_cast_fp16)[name = tensor<string, []>("op_250_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13632832))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14157184))), name = tensor<string, []>("layers_1_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_1_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14157312)))]; |
| tensor<fp16, [1, ?, 1024]> linear_8_cast_fp16 = linear(bias = layers_1_self_attn_k_proj_bias_to_fp16, weight = layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = x_11_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")]; |
| tensor<int32, [4]> concat_7x = const()[name = tensor<string, []>("concat_7x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_256_cast_fp16 = reshape(shape = concat_7x, x = linear_8_cast_fp16)[name = tensor<string, []>("op_256_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14159424))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14683776))), name = tensor<string, []>("layers_1_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14683904)))]; |
| tensor<fp16, [1, ?, 1024]> linear_9_cast_fp16 = linear(bias = layers_1_self_attn_v_proj_bias_to_fp16, weight = layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = x_11_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")]; |
| tensor<int32, [4]> concat_8x = const()[name = tensor<string, []>("concat_8x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_262_cast_fp16 = reshape(shape = concat_8x, x = linear_9_cast_fp16)[name = tensor<string, []>("op_262_cast_fp16")]; |
| tensor<int32, [4]> v_3_perm_0 = const()[name = tensor<string, []>("v_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_265_transpose_x_0 = const()[name = tensor<string, []>("op_265_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_265_transpose_y_0 = const()[name = tensor<string, []>("op_265_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_98_perm_0 = const()[name = tensor<string, []>("transpose_98_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_99_perm_0 = const()[name = tensor<string, []>("transpose_99_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_99 = transpose(perm = transpose_99_perm_0, x = var_256_cast_fp16)[name = tensor<string, []>("transpose_234")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_98 = transpose(perm = transpose_98_perm_0, x = var_250_cast_fp16)[name = tensor<string, []>("transpose_235")]; |
| tensor<fp16, [1, 16, ?, ?]> var_265_cast_fp16 = matmul(transpose_x = var_265_transpose_x_0, transpose_y = var_265_transpose_y_0, x = transpose_98, y = transpose_99)[name = tensor<string, []>("op_265_cast_fp16")]; |
| tensor<fp16, []> var_266_to_fp16 = const()[name = tensor<string, []>("op_266_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_23_cast_fp16 = mul(x = var_265_cast_fp16, y = var_266_to_fp16)[name = tensor<string, []>("input_23_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_3_cast_fp16 = softmax(axis = var_226, x = input_23_cast_fp16)[name = tensor<string, []>("attn_3_cast_fp16")]; |
| tensor<bool, []> out_3_transpose_x_0 = const()[name = tensor<string, []>("out_3_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_3_transpose_y_0 = const()[name = tensor<string, []>("out_3_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_3_cast_fp16 = transpose(perm = v_3_perm_0, x = var_262_cast_fp16)[name = tensor<string, []>("transpose_233")]; |
| tensor<fp16, [1, 16, ?, 64]> out_3_cast_fp16 = matmul(transpose_x = out_3_transpose_x_0, transpose_y = out_3_transpose_y_0, x = attn_3_cast_fp16, y = v_3_cast_fp16)[name = tensor<string, []>("out_3_cast_fp16")]; |
| tensor<int32, [4]> var_270_perm_0 = const()[name = tensor<string, []>("op_270_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_9x = const()[name = tensor<string, []>("concat_9x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_270_cast_fp16 = transpose(perm = var_270_perm_0, x = out_3_cast_fp16)[name = tensor<string, []>("transpose_232")]; |
| tensor<fp16, [1, ?, 1024]> input_25_cast_fp16 = reshape(shape = concat_9x, x = var_270_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_1_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14686016))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15210368))), name = tensor<string, []>("layers_1_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15210496)))]; |
| tensor<fp16, [1, ?, 1024]> linear_10_cast_fp16 = linear(bias = layers_1_self_attn_out_proj_bias_to_fp16, weight = layers_1_self_attn_out_proj_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_27_cast_fp16 = add(x = input_21_cast_fp16, y = linear_10_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")]; |
| tensor<int32, [1]> input_29_axes_0 = const()[name = tensor<string, []>("input_29_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_1_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15212608)))]; |
| tensor<fp16, [1024]> layers_1_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15214720)))]; |
| tensor<fp16, [1, ?, 1024]> input_29_cast_fp16 = layer_norm(axes = input_29_axes_0, beta = layers_1_final_layer_norm_bias_to_fp16, epsilon = var_229_to_fp16, gamma = layers_1_final_layer_norm_weight_to_fp16, x = input_27_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_1_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15216832))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17314048))), name = tensor<string, []>("layers_1_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_1_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17314176)))]; |
| tensor<fp16, [1, ?, 4096]> linear_11_cast_fp16 = linear(bias = layers_1_fc1_bias_to_fp16, weight = layers_1_fc1_weight_to_fp16_palettized, x = input_29_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")]; |
| tensor<string, []> input_31_mode_0 = const()[name = tensor<string, []>("input_31_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = linear_11_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_1_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17322432))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19419648))), name = tensor<string, []>("layers_1_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_1_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19419776)))]; |
| tensor<fp16, [1, ?, 1024]> linear_12_cast_fp16 = linear(bias = layers_1_fc2_bias_to_fp16, weight = layers_1_fc2_weight_to_fp16_palettized, x = input_31_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_33_cast_fp16 = add(x = input_27_cast_fp16, y = linear_12_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")]; |
| tensor<int32, []> var_296 = const()[name = tensor<string, []>("op_296"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_17_axes_0 = const()[name = tensor<string, []>("x_17_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_2_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19421888)))]; |
| tensor<fp16, [1024]> layers_2_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19424000)))]; |
| tensor<fp16, []> var_299_to_fp16 = const()[name = tensor<string, []>("op_299_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_17_cast_fp16 = layer_norm(axes = x_17_axes_0, beta = layers_2_self_attn_layer_norm_bias_to_fp16, epsilon = var_299_to_fp16, gamma = layers_2_self_attn_layer_norm_weight_to_fp16, x = input_33_cast_fp16)[name = tensor<string, []>("x_17_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19426112))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19950464))), name = tensor<string, []>("layers_2_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19950592)))]; |
| tensor<fp16, [1, ?, 1024]> linear_13_cast_fp16 = linear(bias = layers_2_self_attn_q_proj_bias_to_fp16, weight = layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = x_17_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")]; |
| tensor<int32, [4]> concat_10x = const()[name = tensor<string, []>("concat_10x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_320_cast_fp16 = reshape(shape = concat_10x, x = linear_13_cast_fp16)[name = tensor<string, []>("op_320_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19952704))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20477056))), name = tensor<string, []>("layers_2_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_2_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20477184)))]; |
| tensor<fp16, [1, ?, 1024]> linear_14_cast_fp16 = linear(bias = layers_2_self_attn_k_proj_bias_to_fp16, weight = layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = x_17_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")]; |
| tensor<int32, [4]> concat_11x = const()[name = tensor<string, []>("concat_11x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_326_cast_fp16 = reshape(shape = concat_11x, x = linear_14_cast_fp16)[name = tensor<string, []>("op_326_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20479296))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21003648))), name = tensor<string, []>("layers_2_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21003776)))]; |
| tensor<fp16, [1, ?, 1024]> linear_15_cast_fp16 = linear(bias = layers_2_self_attn_v_proj_bias_to_fp16, weight = layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = x_17_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")]; |
| tensor<int32, [4]> concat_12x = const()[name = tensor<string, []>("concat_12x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_332_cast_fp16 = reshape(shape = concat_12x, x = linear_15_cast_fp16)[name = tensor<string, []>("op_332_cast_fp16")]; |
| tensor<int32, [4]> v_5_perm_0 = const()[name = tensor<string, []>("v_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_335_transpose_x_0 = const()[name = tensor<string, []>("op_335_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_335_transpose_y_0 = const()[name = tensor<string, []>("op_335_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_100_perm_0 = const()[name = tensor<string, []>("transpose_100_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_101_perm_0 = const()[name = tensor<string, []>("transpose_101_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_101 = transpose(perm = transpose_101_perm_0, x = var_326_cast_fp16)[name = tensor<string, []>("transpose_230")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_100 = transpose(perm = transpose_100_perm_0, x = var_320_cast_fp16)[name = tensor<string, []>("transpose_231")]; |
| tensor<fp16, [1, 16, ?, ?]> var_335_cast_fp16 = matmul(transpose_x = var_335_transpose_x_0, transpose_y = var_335_transpose_y_0, x = transpose_100, y = transpose_101)[name = tensor<string, []>("op_335_cast_fp16")]; |
| tensor<fp16, []> var_336_to_fp16 = const()[name = tensor<string, []>("op_336_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_35_cast_fp16 = mul(x = var_335_cast_fp16, y = var_336_to_fp16)[name = tensor<string, []>("input_35_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_5_cast_fp16 = softmax(axis = var_296, x = input_35_cast_fp16)[name = tensor<string, []>("attn_5_cast_fp16")]; |
| tensor<bool, []> out_5_transpose_x_0 = const()[name = tensor<string, []>("out_5_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_5_transpose_y_0 = const()[name = tensor<string, []>("out_5_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_5_cast_fp16 = transpose(perm = v_5_perm_0, x = var_332_cast_fp16)[name = tensor<string, []>("transpose_229")]; |
| tensor<fp16, [1, 16, ?, 64]> out_5_cast_fp16 = matmul(transpose_x = out_5_transpose_x_0, transpose_y = out_5_transpose_y_0, x = attn_5_cast_fp16, y = v_5_cast_fp16)[name = tensor<string, []>("out_5_cast_fp16")]; |
| tensor<int32, [4]> var_340_perm_0 = const()[name = tensor<string, []>("op_340_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_13x = const()[name = tensor<string, []>("concat_13x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_340_cast_fp16 = transpose(perm = var_340_perm_0, x = out_5_cast_fp16)[name = tensor<string, []>("transpose_228")]; |
| tensor<fp16, [1, ?, 1024]> input_37_cast_fp16 = reshape(shape = concat_13x, x = var_340_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_2_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21005888))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21530240))), name = tensor<string, []>("layers_2_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21530368)))]; |
| tensor<fp16, [1, ?, 1024]> linear_16_cast_fp16 = linear(bias = layers_2_self_attn_out_proj_bias_to_fp16, weight = layers_2_self_attn_out_proj_weight_to_fp16_palettized, x = input_37_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_39_cast_fp16 = add(x = input_33_cast_fp16, y = linear_16_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")]; |
| tensor<int32, [1]> input_41_axes_0 = const()[name = tensor<string, []>("input_41_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_2_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21532480)))]; |
| tensor<fp16, [1024]> layers_2_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21534592)))]; |
| tensor<fp16, [1, ?, 1024]> input_41_cast_fp16 = layer_norm(axes = input_41_axes_0, beta = layers_2_final_layer_norm_bias_to_fp16, epsilon = var_299_to_fp16, gamma = layers_2_final_layer_norm_weight_to_fp16, x = input_39_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_2_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21536704))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23633920))), name = tensor<string, []>("layers_2_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_2_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23634048)))]; |
| tensor<fp16, [1, ?, 4096]> linear_17_cast_fp16 = linear(bias = layers_2_fc1_bias_to_fp16, weight = layers_2_fc1_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")]; |
| tensor<string, []> input_43_mode_0 = const()[name = tensor<string, []>("input_43_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_43_cast_fp16 = gelu(mode = input_43_mode_0, x = linear_17_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_2_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23642304))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25739520))), name = tensor<string, []>("layers_2_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_2_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25739648)))]; |
| tensor<fp16, [1, ?, 1024]> linear_18_cast_fp16 = linear(bias = layers_2_fc2_bias_to_fp16, weight = layers_2_fc2_weight_to_fp16_palettized, x = input_43_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_45_cast_fp16 = add(x = input_39_cast_fp16, y = linear_18_cast_fp16)[name = tensor<string, []>("input_45_cast_fp16")]; |
| tensor<int32, []> var_366 = const()[name = tensor<string, []>("op_366"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_23_axes_0 = const()[name = tensor<string, []>("x_23_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_3_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25741760)))]; |
| tensor<fp16, [1024]> layers_3_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25743872)))]; |
| tensor<fp16, []> var_369_to_fp16 = const()[name = tensor<string, []>("op_369_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_23_cast_fp16 = layer_norm(axes = x_23_axes_0, beta = layers_3_self_attn_layer_norm_bias_to_fp16, epsilon = var_369_to_fp16, gamma = layers_3_self_attn_layer_norm_weight_to_fp16, x = input_45_cast_fp16)[name = tensor<string, []>("x_23_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25745984))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26270336))), name = tensor<string, []>("layers_3_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26270464)))]; |
| tensor<fp16, [1, ?, 1024]> linear_19_cast_fp16 = linear(bias = layers_3_self_attn_q_proj_bias_to_fp16, weight = layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = x_23_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")]; |
| tensor<int32, [4]> concat_14x = const()[name = tensor<string, []>("concat_14x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_390_cast_fp16 = reshape(shape = concat_14x, x = linear_19_cast_fp16)[name = tensor<string, []>("op_390_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26272576))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26796928))), name = tensor<string, []>("layers_3_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_3_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26797056)))]; |
| tensor<fp16, [1, ?, 1024]> linear_20_cast_fp16 = linear(bias = layers_3_self_attn_k_proj_bias_to_fp16, weight = layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = x_23_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")]; |
| tensor<int32, [4]> concat_15x = const()[name = tensor<string, []>("concat_15x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_396_cast_fp16 = reshape(shape = concat_15x, x = linear_20_cast_fp16)[name = tensor<string, []>("op_396_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26799168))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27323520))), name = tensor<string, []>("layers_3_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27323648)))]; |
| tensor<fp16, [1, ?, 1024]> linear_21_cast_fp16 = linear(bias = layers_3_self_attn_v_proj_bias_to_fp16, weight = layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = x_23_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")]; |
| tensor<int32, [4]> concat_16x = const()[name = tensor<string, []>("concat_16x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_402_cast_fp16 = reshape(shape = concat_16x, x = linear_21_cast_fp16)[name = tensor<string, []>("op_402_cast_fp16")]; |
| tensor<int32, [4]> v_7_perm_0 = const()[name = tensor<string, []>("v_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_405_transpose_x_0 = const()[name = tensor<string, []>("op_405_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_405_transpose_y_0 = const()[name = tensor<string, []>("op_405_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_102_perm_0 = const()[name = tensor<string, []>("transpose_102_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_103_perm_0 = const()[name = tensor<string, []>("transpose_103_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_103 = transpose(perm = transpose_103_perm_0, x = var_396_cast_fp16)[name = tensor<string, []>("transpose_226")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_102 = transpose(perm = transpose_102_perm_0, x = var_390_cast_fp16)[name = tensor<string, []>("transpose_227")]; |
| tensor<fp16, [1, 16, ?, ?]> var_405_cast_fp16 = matmul(transpose_x = var_405_transpose_x_0, transpose_y = var_405_transpose_y_0, x = transpose_102, y = transpose_103)[name = tensor<string, []>("op_405_cast_fp16")]; |
| tensor<fp16, []> var_406_to_fp16 = const()[name = tensor<string, []>("op_406_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_47_cast_fp16 = mul(x = var_405_cast_fp16, y = var_406_to_fp16)[name = tensor<string, []>("input_47_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_7_cast_fp16 = softmax(axis = var_366, x = input_47_cast_fp16)[name = tensor<string, []>("attn_7_cast_fp16")]; |
| tensor<bool, []> out_7_transpose_x_0 = const()[name = tensor<string, []>("out_7_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_7_transpose_y_0 = const()[name = tensor<string, []>("out_7_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_7_cast_fp16 = transpose(perm = v_7_perm_0, x = var_402_cast_fp16)[name = tensor<string, []>("transpose_225")]; |
| tensor<fp16, [1, 16, ?, 64]> out_7_cast_fp16 = matmul(transpose_x = out_7_transpose_x_0, transpose_y = out_7_transpose_y_0, x = attn_7_cast_fp16, y = v_7_cast_fp16)[name = tensor<string, []>("out_7_cast_fp16")]; |
| tensor<int32, [4]> var_410_perm_0 = const()[name = tensor<string, []>("op_410_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_17x = const()[name = tensor<string, []>("concat_17x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_410_cast_fp16 = transpose(perm = var_410_perm_0, x = out_7_cast_fp16)[name = tensor<string, []>("transpose_224")]; |
| tensor<fp16, [1, ?, 1024]> input_49_cast_fp16 = reshape(shape = concat_17x, x = var_410_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_3_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27325760))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27850112))), name = tensor<string, []>("layers_3_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27850240)))]; |
| tensor<fp16, [1, ?, 1024]> linear_22_cast_fp16 = linear(bias = layers_3_self_attn_out_proj_bias_to_fp16, weight = layers_3_self_attn_out_proj_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_51_cast_fp16 = add(x = input_45_cast_fp16, y = linear_22_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")]; |
| tensor<int32, [1]> input_53_axes_0 = const()[name = tensor<string, []>("input_53_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_3_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27852352)))]; |
| tensor<fp16, [1024]> layers_3_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27854464)))]; |
| tensor<fp16, [1, ?, 1024]> input_53_cast_fp16 = layer_norm(axes = input_53_axes_0, beta = layers_3_final_layer_norm_bias_to_fp16, epsilon = var_369_to_fp16, gamma = layers_3_final_layer_norm_weight_to_fp16, x = input_51_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_3_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27856576))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29953792))), name = tensor<string, []>("layers_3_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_3_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29953920)))]; |
| tensor<fp16, [1, ?, 4096]> linear_23_cast_fp16 = linear(bias = layers_3_fc1_bias_to_fp16, weight = layers_3_fc1_weight_to_fp16_palettized, x = input_53_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")]; |
| tensor<string, []> input_55_mode_0 = const()[name = tensor<string, []>("input_55_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_55_cast_fp16 = gelu(mode = input_55_mode_0, x = linear_23_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_3_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29962176))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32059392))), name = tensor<string, []>("layers_3_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_3_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32059520)))]; |
| tensor<fp16, [1, ?, 1024]> linear_24_cast_fp16 = linear(bias = layers_3_fc2_bias_to_fp16, weight = layers_3_fc2_weight_to_fp16_palettized, x = input_55_cast_fp16)[name = tensor<string, []>("linear_24_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_57_cast_fp16 = add(x = input_51_cast_fp16, y = linear_24_cast_fp16)[name = tensor<string, []>("input_57_cast_fp16")]; |
| tensor<int32, []> var_436 = const()[name = tensor<string, []>("op_436"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_29_axes_0 = const()[name = tensor<string, []>("x_29_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_4_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32061632)))]; |
| tensor<fp16, [1024]> layers_4_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32063744)))]; |
| tensor<fp16, []> var_439_to_fp16 = const()[name = tensor<string, []>("op_439_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_29_cast_fp16 = layer_norm(axes = x_29_axes_0, beta = layers_4_self_attn_layer_norm_bias_to_fp16, epsilon = var_439_to_fp16, gamma = layers_4_self_attn_layer_norm_weight_to_fp16, x = input_57_cast_fp16)[name = tensor<string, []>("x_29_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32065856))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32590208))), name = tensor<string, []>("layers_4_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32590336)))]; |
| tensor<fp16, [1, ?, 1024]> linear_25_cast_fp16 = linear(bias = layers_4_self_attn_q_proj_bias_to_fp16, weight = layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = x_29_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")]; |
| tensor<int32, [4]> concat_18x = const()[name = tensor<string, []>("concat_18x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_460_cast_fp16 = reshape(shape = concat_18x, x = linear_25_cast_fp16)[name = tensor<string, []>("op_460_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32592448))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33116800))), name = tensor<string, []>("layers_4_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_4_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33116928)))]; |
| tensor<fp16, [1, ?, 1024]> linear_26_cast_fp16 = linear(bias = layers_4_self_attn_k_proj_bias_to_fp16, weight = layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = x_29_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")]; |
| tensor<int32, [4]> concat_19x = const()[name = tensor<string, []>("concat_19x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_466_cast_fp16 = reshape(shape = concat_19x, x = linear_26_cast_fp16)[name = tensor<string, []>("op_466_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33119040))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33643392))), name = tensor<string, []>("layers_4_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33643520)))]; |
| tensor<fp16, [1, ?, 1024]> linear_27_cast_fp16 = linear(bias = layers_4_self_attn_v_proj_bias_to_fp16, weight = layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = x_29_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")]; |
| tensor<int32, [4]> concat_20x = const()[name = tensor<string, []>("concat_20x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_472_cast_fp16 = reshape(shape = concat_20x, x = linear_27_cast_fp16)[name = tensor<string, []>("op_472_cast_fp16")]; |
| tensor<int32, [4]> v_9_perm_0 = const()[name = tensor<string, []>("v_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_475_transpose_x_0 = const()[name = tensor<string, []>("op_475_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_475_transpose_y_0 = const()[name = tensor<string, []>("op_475_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_104_perm_0 = const()[name = tensor<string, []>("transpose_104_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_105_perm_0 = const()[name = tensor<string, []>("transpose_105_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_105 = transpose(perm = transpose_105_perm_0, x = var_466_cast_fp16)[name = tensor<string, []>("transpose_222")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_104 = transpose(perm = transpose_104_perm_0, x = var_460_cast_fp16)[name = tensor<string, []>("transpose_223")]; |
| tensor<fp16, [1, 16, ?, ?]> var_475_cast_fp16 = matmul(transpose_x = var_475_transpose_x_0, transpose_y = var_475_transpose_y_0, x = transpose_104, y = transpose_105)[name = tensor<string, []>("op_475_cast_fp16")]; |
| tensor<fp16, []> var_476_to_fp16 = const()[name = tensor<string, []>("op_476_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_59_cast_fp16 = mul(x = var_475_cast_fp16, y = var_476_to_fp16)[name = tensor<string, []>("input_59_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_9_cast_fp16 = softmax(axis = var_436, x = input_59_cast_fp16)[name = tensor<string, []>("attn_9_cast_fp16")]; |
| tensor<bool, []> out_9_transpose_x_0 = const()[name = tensor<string, []>("out_9_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_9_transpose_y_0 = const()[name = tensor<string, []>("out_9_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_9_cast_fp16 = transpose(perm = v_9_perm_0, x = var_472_cast_fp16)[name = tensor<string, []>("transpose_221")]; |
| tensor<fp16, [1, 16, ?, 64]> out_9_cast_fp16 = matmul(transpose_x = out_9_transpose_x_0, transpose_y = out_9_transpose_y_0, x = attn_9_cast_fp16, y = v_9_cast_fp16)[name = tensor<string, []>("out_9_cast_fp16")]; |
| tensor<int32, [4]> var_480_perm_0 = const()[name = tensor<string, []>("op_480_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_21x = const()[name = tensor<string, []>("concat_21x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_480_cast_fp16 = transpose(perm = var_480_perm_0, x = out_9_cast_fp16)[name = tensor<string, []>("transpose_220")]; |
| tensor<fp16, [1, ?, 1024]> input_61_cast_fp16 = reshape(shape = concat_21x, x = var_480_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_4_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33645632))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34169984))), name = tensor<string, []>("layers_4_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_4_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34170112)))]; |
| tensor<fp16, [1, ?, 1024]> linear_28_cast_fp16 = linear(bias = layers_4_self_attn_out_proj_bias_to_fp16, weight = layers_4_self_attn_out_proj_weight_to_fp16_palettized, x = input_61_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_63_cast_fp16 = add(x = input_57_cast_fp16, y = linear_28_cast_fp16)[name = tensor<string, []>("input_63_cast_fp16")]; |
| tensor<int32, [1]> input_65_axes_0 = const()[name = tensor<string, []>("input_65_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_4_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34172224)))]; |
| tensor<fp16, [1024]> layers_4_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34174336)))]; |
| tensor<fp16, [1, ?, 1024]> input_65_cast_fp16 = layer_norm(axes = input_65_axes_0, beta = layers_4_final_layer_norm_bias_to_fp16, epsilon = var_439_to_fp16, gamma = layers_4_final_layer_norm_weight_to_fp16, x = input_63_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_4_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34176448))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36273664))), name = tensor<string, []>("layers_4_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_4_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36273792)))]; |
| tensor<fp16, [1, ?, 4096]> linear_29_cast_fp16 = linear(bias = layers_4_fc1_bias_to_fp16, weight = layers_4_fc1_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")]; |
| tensor<string, []> input_67_mode_0 = const()[name = tensor<string, []>("input_67_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_67_cast_fp16 = gelu(mode = input_67_mode_0, x = linear_29_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_4_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36282048))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38379264))), name = tensor<string, []>("layers_4_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_4_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38379392)))]; |
| tensor<fp16, [1, ?, 1024]> linear_30_cast_fp16 = linear(bias = layers_4_fc2_bias_to_fp16, weight = layers_4_fc2_weight_to_fp16_palettized, x = input_67_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_69_cast_fp16 = add(x = input_63_cast_fp16, y = linear_30_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")]; |
| tensor<int32, []> var_506 = const()[name = tensor<string, []>("op_506"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_35_axes_0 = const()[name = tensor<string, []>("x_35_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_5_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38381504)))]; |
| tensor<fp16, [1024]> layers_5_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38383616)))]; |
| tensor<fp16, []> var_509_to_fp16 = const()[name = tensor<string, []>("op_509_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_35_cast_fp16 = layer_norm(axes = x_35_axes_0, beta = layers_5_self_attn_layer_norm_bias_to_fp16, epsilon = var_509_to_fp16, gamma = layers_5_self_attn_layer_norm_weight_to_fp16, x = input_69_cast_fp16)[name = tensor<string, []>("x_35_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38385728))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38910080))), name = tensor<string, []>("layers_5_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38910208)))]; |
| tensor<fp16, [1, ?, 1024]> linear_31_cast_fp16 = linear(bias = layers_5_self_attn_q_proj_bias_to_fp16, weight = layers_5_self_attn_q_proj_weight_to_fp16_palettized, x = x_35_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")]; |
| tensor<int32, [4]> concat_22x = const()[name = tensor<string, []>("concat_22x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_530_cast_fp16 = reshape(shape = concat_22x, x = linear_31_cast_fp16)[name = tensor<string, []>("op_530_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38912320))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39436672))), name = tensor<string, []>("layers_5_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_5_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39436800)))]; |
| tensor<fp16, [1, ?, 1024]> linear_32_cast_fp16 = linear(bias = layers_5_self_attn_k_proj_bias_to_fp16, weight = layers_5_self_attn_k_proj_weight_to_fp16_palettized, x = x_35_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")]; |
| tensor<int32, [4]> concat_23x = const()[name = tensor<string, []>("concat_23x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_536_cast_fp16 = reshape(shape = concat_23x, x = linear_32_cast_fp16)[name = tensor<string, []>("op_536_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39438912))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39963264))), name = tensor<string, []>("layers_5_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39963392)))]; |
| tensor<fp16, [1, ?, 1024]> linear_33_cast_fp16 = linear(bias = layers_5_self_attn_v_proj_bias_to_fp16, weight = layers_5_self_attn_v_proj_weight_to_fp16_palettized, x = x_35_cast_fp16)[name = tensor<string, []>("linear_33_cast_fp16")]; |
| tensor<int32, [4]> concat_24x = const()[name = tensor<string, []>("concat_24x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_542_cast_fp16 = reshape(shape = concat_24x, x = linear_33_cast_fp16)[name = tensor<string, []>("op_542_cast_fp16")]; |
| tensor<int32, [4]> v_11_perm_0 = const()[name = tensor<string, []>("v_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_545_transpose_x_0 = const()[name = tensor<string, []>("op_545_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_545_transpose_y_0 = const()[name = tensor<string, []>("op_545_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_106_perm_0 = const()[name = tensor<string, []>("transpose_106_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_107_perm_0 = const()[name = tensor<string, []>("transpose_107_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_107 = transpose(perm = transpose_107_perm_0, x = var_536_cast_fp16)[name = tensor<string, []>("transpose_218")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_106 = transpose(perm = transpose_106_perm_0, x = var_530_cast_fp16)[name = tensor<string, []>("transpose_219")]; |
| tensor<fp16, [1, 16, ?, ?]> var_545_cast_fp16 = matmul(transpose_x = var_545_transpose_x_0, transpose_y = var_545_transpose_y_0, x = transpose_106, y = transpose_107)[name = tensor<string, []>("op_545_cast_fp16")]; |
| tensor<fp16, []> var_546_to_fp16 = const()[name = tensor<string, []>("op_546_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_71_cast_fp16 = mul(x = var_545_cast_fp16, y = var_546_to_fp16)[name = tensor<string, []>("input_71_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_11_cast_fp16 = softmax(axis = var_506, x = input_71_cast_fp16)[name = tensor<string, []>("attn_11_cast_fp16")]; |
| tensor<bool, []> out_11_transpose_x_0 = const()[name = tensor<string, []>("out_11_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_11_transpose_y_0 = const()[name = tensor<string, []>("out_11_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_11_cast_fp16 = transpose(perm = v_11_perm_0, x = var_542_cast_fp16)[name = tensor<string, []>("transpose_217")]; |
| tensor<fp16, [1, 16, ?, 64]> out_11_cast_fp16 = matmul(transpose_x = out_11_transpose_x_0, transpose_y = out_11_transpose_y_0, x = attn_11_cast_fp16, y = v_11_cast_fp16)[name = tensor<string, []>("out_11_cast_fp16")]; |
| tensor<int32, [4]> var_550_perm_0 = const()[name = tensor<string, []>("op_550_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_25x = const()[name = tensor<string, []>("concat_25x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_550_cast_fp16 = transpose(perm = var_550_perm_0, x = out_11_cast_fp16)[name = tensor<string, []>("transpose_216")]; |
| tensor<fp16, [1, ?, 1024]> input_73_cast_fp16 = reshape(shape = concat_25x, x = var_550_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_5_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39965504))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40489856))), name = tensor<string, []>("layers_5_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_5_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40489984)))]; |
| tensor<fp16, [1, ?, 1024]> linear_34_cast_fp16 = linear(bias = layers_5_self_attn_out_proj_bias_to_fp16, weight = layers_5_self_attn_out_proj_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_75_cast_fp16 = add(x = input_69_cast_fp16, y = linear_34_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")]; |
| tensor<int32, [1]> input_77_axes_0 = const()[name = tensor<string, []>("input_77_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_5_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40492096)))]; |
| tensor<fp16, [1024]> layers_5_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40494208)))]; |
| tensor<fp16, [1, ?, 1024]> input_77_cast_fp16 = layer_norm(axes = input_77_axes_0, beta = layers_5_final_layer_norm_bias_to_fp16, epsilon = var_509_to_fp16, gamma = layers_5_final_layer_norm_weight_to_fp16, x = input_75_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_5_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40496320))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42593536))), name = tensor<string, []>("layers_5_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_5_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42593664)))]; |
| tensor<fp16, [1, ?, 4096]> linear_35_cast_fp16 = linear(bias = layers_5_fc1_bias_to_fp16, weight = layers_5_fc1_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")]; |
| tensor<string, []> input_79_mode_0 = const()[name = tensor<string, []>("input_79_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = linear_35_cast_fp16)[name = tensor<string, []>("input_79_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_5_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42601920))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44699136))), name = tensor<string, []>("layers_5_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_5_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44699264)))]; |
| tensor<fp16, [1, ?, 1024]> linear_36_cast_fp16 = linear(bias = layers_5_fc2_bias_to_fp16, weight = layers_5_fc2_weight_to_fp16_palettized, x = input_79_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_81_cast_fp16 = add(x = input_75_cast_fp16, y = linear_36_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")]; |
| tensor<int32, []> var_576 = const()[name = tensor<string, []>("op_576"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_41_axes_0 = const()[name = tensor<string, []>("x_41_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_6_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44701376)))]; |
| tensor<fp16, [1024]> layers_6_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44703488)))]; |
| tensor<fp16, []> var_579_to_fp16 = const()[name = tensor<string, []>("op_579_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_41_cast_fp16 = layer_norm(axes = x_41_axes_0, beta = layers_6_self_attn_layer_norm_bias_to_fp16, epsilon = var_579_to_fp16, gamma = layers_6_self_attn_layer_norm_weight_to_fp16, x = input_81_cast_fp16)[name = tensor<string, []>("x_41_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44705600))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45229952))), name = tensor<string, []>("layers_6_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45230080)))]; |
| tensor<fp16, [1, ?, 1024]> linear_37_cast_fp16 = linear(bias = layers_6_self_attn_q_proj_bias_to_fp16, weight = layers_6_self_attn_q_proj_weight_to_fp16_palettized, x = x_41_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")]; |
| tensor<int32, [4]> concat_26x = const()[name = tensor<string, []>("concat_26x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_600_cast_fp16 = reshape(shape = concat_26x, x = linear_37_cast_fp16)[name = tensor<string, []>("op_600_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45232192))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45756544))), name = tensor<string, []>("layers_6_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_6_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45756672)))]; |
| tensor<fp16, [1, ?, 1024]> linear_38_cast_fp16 = linear(bias = layers_6_self_attn_k_proj_bias_to_fp16, weight = layers_6_self_attn_k_proj_weight_to_fp16_palettized, x = x_41_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")]; |
| tensor<int32, [4]> concat_27x = const()[name = tensor<string, []>("concat_27x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_606_cast_fp16 = reshape(shape = concat_27x, x = linear_38_cast_fp16)[name = tensor<string, []>("op_606_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45758784))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46283136))), name = tensor<string, []>("layers_6_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46283264)))]; |
| tensor<fp16, [1, ?, 1024]> linear_39_cast_fp16 = linear(bias = layers_6_self_attn_v_proj_bias_to_fp16, weight = layers_6_self_attn_v_proj_weight_to_fp16_palettized, x = x_41_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")]; |
| tensor<int32, [4]> concat_28x = const()[name = tensor<string, []>("concat_28x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_612_cast_fp16 = reshape(shape = concat_28x, x = linear_39_cast_fp16)[name = tensor<string, []>("op_612_cast_fp16")]; |
| tensor<int32, [4]> v_13_perm_0 = const()[name = tensor<string, []>("v_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_615_transpose_x_0 = const()[name = tensor<string, []>("op_615_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_615_transpose_y_0 = const()[name = tensor<string, []>("op_615_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_108_perm_0 = const()[name = tensor<string, []>("transpose_108_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_109_perm_0 = const()[name = tensor<string, []>("transpose_109_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_109 = transpose(perm = transpose_109_perm_0, x = var_606_cast_fp16)[name = tensor<string, []>("transpose_214")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_108 = transpose(perm = transpose_108_perm_0, x = var_600_cast_fp16)[name = tensor<string, []>("transpose_215")]; |
| tensor<fp16, [1, 16, ?, ?]> var_615_cast_fp16 = matmul(transpose_x = var_615_transpose_x_0, transpose_y = var_615_transpose_y_0, x = transpose_108, y = transpose_109)[name = tensor<string, []>("op_615_cast_fp16")]; |
| tensor<fp16, []> var_616_to_fp16 = const()[name = tensor<string, []>("op_616_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_83_cast_fp16 = mul(x = var_615_cast_fp16, y = var_616_to_fp16)[name = tensor<string, []>("input_83_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_13_cast_fp16 = softmax(axis = var_576, x = input_83_cast_fp16)[name = tensor<string, []>("attn_13_cast_fp16")]; |
| tensor<bool, []> out_13_transpose_x_0 = const()[name = tensor<string, []>("out_13_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_13_transpose_y_0 = const()[name = tensor<string, []>("out_13_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_13_cast_fp16 = transpose(perm = v_13_perm_0, x = var_612_cast_fp16)[name = tensor<string, []>("transpose_213")]; |
| tensor<fp16, [1, 16, ?, 64]> out_13_cast_fp16 = matmul(transpose_x = out_13_transpose_x_0, transpose_y = out_13_transpose_y_0, x = attn_13_cast_fp16, y = v_13_cast_fp16)[name = tensor<string, []>("out_13_cast_fp16")]; |
| tensor<int32, [4]> var_620_perm_0 = const()[name = tensor<string, []>("op_620_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_29x = const()[name = tensor<string, []>("concat_29x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_620_cast_fp16 = transpose(perm = var_620_perm_0, x = out_13_cast_fp16)[name = tensor<string, []>("transpose_212")]; |
| tensor<fp16, [1, ?, 1024]> input_85_cast_fp16 = reshape(shape = concat_29x, x = var_620_cast_fp16)[name = tensor<string, []>("input_85_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_6_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46285376))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46809728))), name = tensor<string, []>("layers_6_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_6_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46809856)))]; |
| tensor<fp16, [1, ?, 1024]> linear_40_cast_fp16 = linear(bias = layers_6_self_attn_out_proj_bias_to_fp16, weight = layers_6_self_attn_out_proj_weight_to_fp16_palettized, x = input_85_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_87_cast_fp16 = add(x = input_81_cast_fp16, y = linear_40_cast_fp16)[name = tensor<string, []>("input_87_cast_fp16")]; |
| tensor<int32, [1]> input_89_axes_0 = const()[name = tensor<string, []>("input_89_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_6_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46811968)))]; |
| tensor<fp16, [1024]> layers_6_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46814080)))]; |
| tensor<fp16, [1, ?, 1024]> input_89_cast_fp16 = layer_norm(axes = input_89_axes_0, beta = layers_6_final_layer_norm_bias_to_fp16, epsilon = var_579_to_fp16, gamma = layers_6_final_layer_norm_weight_to_fp16, x = input_87_cast_fp16)[name = tensor<string, []>("input_89_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_6_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46816192))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48913408))), name = tensor<string, []>("layers_6_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_6_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48913536)))]; |
| tensor<fp16, [1, ?, 4096]> linear_41_cast_fp16 = linear(bias = layers_6_fc1_bias_to_fp16, weight = layers_6_fc1_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")]; |
| tensor<string, []> input_91_mode_0 = const()[name = tensor<string, []>("input_91_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_91_cast_fp16 = gelu(mode = input_91_mode_0, x = linear_41_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_6_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48921792))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51019008))), name = tensor<string, []>("layers_6_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_6_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51019136)))]; |
| tensor<fp16, [1, ?, 1024]> linear_42_cast_fp16 = linear(bias = layers_6_fc2_bias_to_fp16, weight = layers_6_fc2_weight_to_fp16_palettized, x = input_91_cast_fp16)[name = tensor<string, []>("linear_42_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_93_cast_fp16 = add(x = input_87_cast_fp16, y = linear_42_cast_fp16)[name = tensor<string, []>("input_93_cast_fp16")]; |
| tensor<int32, []> var_646 = const()[name = tensor<string, []>("op_646"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_47_axes_0 = const()[name = tensor<string, []>("x_47_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_7_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51021248)))]; |
| tensor<fp16, [1024]> layers_7_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51023360)))]; |
| tensor<fp16, []> var_649_to_fp16 = const()[name = tensor<string, []>("op_649_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_47_cast_fp16 = layer_norm(axes = x_47_axes_0, beta = layers_7_self_attn_layer_norm_bias_to_fp16, epsilon = var_649_to_fp16, gamma = layers_7_self_attn_layer_norm_weight_to_fp16, x = input_93_cast_fp16)[name = tensor<string, []>("x_47_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51025472))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51549824))), name = tensor<string, []>("layers_7_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51549952)))]; |
| tensor<fp16, [1, ?, 1024]> linear_43_cast_fp16 = linear(bias = layers_7_self_attn_q_proj_bias_to_fp16, weight = layers_7_self_attn_q_proj_weight_to_fp16_palettized, x = x_47_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")]; |
| tensor<int32, [4]> concat_30x = const()[name = tensor<string, []>("concat_30x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_670_cast_fp16 = reshape(shape = concat_30x, x = linear_43_cast_fp16)[name = tensor<string, []>("op_670_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51552064))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52076416))), name = tensor<string, []>("layers_7_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_7_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52076544)))]; |
| tensor<fp16, [1, ?, 1024]> linear_44_cast_fp16 = linear(bias = layers_7_self_attn_k_proj_bias_to_fp16, weight = layers_7_self_attn_k_proj_weight_to_fp16_palettized, x = x_47_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")]; |
| tensor<int32, [4]> concat_31x = const()[name = tensor<string, []>("concat_31x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_676_cast_fp16 = reshape(shape = concat_31x, x = linear_44_cast_fp16)[name = tensor<string, []>("op_676_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52078656))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52603008))), name = tensor<string, []>("layers_7_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52603136)))]; |
| tensor<fp16, [1, ?, 1024]> linear_45_cast_fp16 = linear(bias = layers_7_self_attn_v_proj_bias_to_fp16, weight = layers_7_self_attn_v_proj_weight_to_fp16_palettized, x = x_47_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")]; |
| tensor<int32, [4]> concat_32x = const()[name = tensor<string, []>("concat_32x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_682_cast_fp16 = reshape(shape = concat_32x, x = linear_45_cast_fp16)[name = tensor<string, []>("op_682_cast_fp16")]; |
| tensor<int32, [4]> v_15_perm_0 = const()[name = tensor<string, []>("v_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_685_transpose_x_0 = const()[name = tensor<string, []>("op_685_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_685_transpose_y_0 = const()[name = tensor<string, []>("op_685_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_110_perm_0 = const()[name = tensor<string, []>("transpose_110_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_111_perm_0 = const()[name = tensor<string, []>("transpose_111_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_111 = transpose(perm = transpose_111_perm_0, x = var_676_cast_fp16)[name = tensor<string, []>("transpose_210")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_110 = transpose(perm = transpose_110_perm_0, x = var_670_cast_fp16)[name = tensor<string, []>("transpose_211")]; |
| tensor<fp16, [1, 16, ?, ?]> var_685_cast_fp16 = matmul(transpose_x = var_685_transpose_x_0, transpose_y = var_685_transpose_y_0, x = transpose_110, y = transpose_111)[name = tensor<string, []>("op_685_cast_fp16")]; |
| tensor<fp16, []> var_686_to_fp16 = const()[name = tensor<string, []>("op_686_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_95_cast_fp16 = mul(x = var_685_cast_fp16, y = var_686_to_fp16)[name = tensor<string, []>("input_95_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_15_cast_fp16 = softmax(axis = var_646, x = input_95_cast_fp16)[name = tensor<string, []>("attn_15_cast_fp16")]; |
| tensor<bool, []> out_15_transpose_x_0 = const()[name = tensor<string, []>("out_15_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_15_transpose_y_0 = const()[name = tensor<string, []>("out_15_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_15_cast_fp16 = transpose(perm = v_15_perm_0, x = var_682_cast_fp16)[name = tensor<string, []>("transpose_209")]; |
| tensor<fp16, [1, 16, ?, 64]> out_15_cast_fp16 = matmul(transpose_x = out_15_transpose_x_0, transpose_y = out_15_transpose_y_0, x = attn_15_cast_fp16, y = v_15_cast_fp16)[name = tensor<string, []>("out_15_cast_fp16")]; |
| tensor<int32, [4]> var_690_perm_0 = const()[name = tensor<string, []>("op_690_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_33x = const()[name = tensor<string, []>("concat_33x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_690_cast_fp16 = transpose(perm = var_690_perm_0, x = out_15_cast_fp16)[name = tensor<string, []>("transpose_208")]; |
| tensor<fp16, [1, ?, 1024]> input_97_cast_fp16 = reshape(shape = concat_33x, x = var_690_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_7_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52605248))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53129600))), name = tensor<string, []>("layers_7_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_7_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53129728)))]; |
| tensor<fp16, [1, ?, 1024]> linear_46_cast_fp16 = linear(bias = layers_7_self_attn_out_proj_bias_to_fp16, weight = layers_7_self_attn_out_proj_weight_to_fp16_palettized, x = input_97_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_99_cast_fp16 = add(x = input_93_cast_fp16, y = linear_46_cast_fp16)[name = tensor<string, []>("input_99_cast_fp16")]; |
| tensor<int32, [1]> input_101_axes_0 = const()[name = tensor<string, []>("input_101_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_7_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53131840)))]; |
| tensor<fp16, [1024]> layers_7_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53133952)))]; |
| tensor<fp16, [1, ?, 1024]> input_101_cast_fp16 = layer_norm(axes = input_101_axes_0, beta = layers_7_final_layer_norm_bias_to_fp16, epsilon = var_649_to_fp16, gamma = layers_7_final_layer_norm_weight_to_fp16, x = input_99_cast_fp16)[name = tensor<string, []>("input_101_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_7_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53136064))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55233280))), name = tensor<string, []>("layers_7_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_7_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55233408)))]; |
| tensor<fp16, [1, ?, 4096]> linear_47_cast_fp16 = linear(bias = layers_7_fc1_bias_to_fp16, weight = layers_7_fc1_weight_to_fp16_palettized, x = input_101_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")]; |
| tensor<string, []> input_103_mode_0 = const()[name = tensor<string, []>("input_103_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_103_cast_fp16 = gelu(mode = input_103_mode_0, x = linear_47_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_7_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55241664))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57338880))), name = tensor<string, []>("layers_7_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_7_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57339008)))]; |
| tensor<fp16, [1, ?, 1024]> linear_48_cast_fp16 = linear(bias = layers_7_fc2_bias_to_fp16, weight = layers_7_fc2_weight_to_fp16_palettized, x = input_103_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_105_cast_fp16 = add(x = input_99_cast_fp16, y = linear_48_cast_fp16)[name = tensor<string, []>("input_105_cast_fp16")]; |
| tensor<int32, []> var_716 = const()[name = tensor<string, []>("op_716"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_53_axes_0 = const()[name = tensor<string, []>("x_53_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_8_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57341120)))]; |
| tensor<fp16, [1024]> layers_8_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57343232)))]; |
| tensor<fp16, []> var_719_to_fp16 = const()[name = tensor<string, []>("op_719_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_53_cast_fp16 = layer_norm(axes = x_53_axes_0, beta = layers_8_self_attn_layer_norm_bias_to_fp16, epsilon = var_719_to_fp16, gamma = layers_8_self_attn_layer_norm_weight_to_fp16, x = input_105_cast_fp16)[name = tensor<string, []>("x_53_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_8_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57345344))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57869696))), name = tensor<string, []>("layers_8_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57869824)))]; |
| tensor<fp16, [1, ?, 1024]> linear_49_cast_fp16 = linear(bias = layers_8_self_attn_q_proj_bias_to_fp16, weight = layers_8_self_attn_q_proj_weight_to_fp16_palettized, x = x_53_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")]; |
| tensor<int32, [4]> concat_34x = const()[name = tensor<string, []>("concat_34x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_740_cast_fp16 = reshape(shape = concat_34x, x = linear_49_cast_fp16)[name = tensor<string, []>("op_740_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57871936))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58396288))), name = tensor<string, []>("layers_8_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_8_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58396416)))]; |
| tensor<fp16, [1, ?, 1024]> linear_50_cast_fp16 = linear(bias = layers_8_self_attn_k_proj_bias_to_fp16, weight = layers_8_self_attn_k_proj_weight_to_fp16_palettized, x = x_53_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")]; |
| tensor<int32, [4]> concat_35x = const()[name = tensor<string, []>("concat_35x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_746_cast_fp16 = reshape(shape = concat_35x, x = linear_50_cast_fp16)[name = tensor<string, []>("op_746_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_8_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58398528))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58922880))), name = tensor<string, []>("layers_8_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58923008)))]; |
| tensor<fp16, [1, ?, 1024]> linear_51_cast_fp16 = linear(bias = layers_8_self_attn_v_proj_bias_to_fp16, weight = layers_8_self_attn_v_proj_weight_to_fp16_palettized, x = x_53_cast_fp16)[name = tensor<string, []>("linear_51_cast_fp16")]; |
| tensor<int32, [4]> concat_36x = const()[name = tensor<string, []>("concat_36x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_752_cast_fp16 = reshape(shape = concat_36x, x = linear_51_cast_fp16)[name = tensor<string, []>("op_752_cast_fp16")]; |
| tensor<int32, [4]> v_17_perm_0 = const()[name = tensor<string, []>("v_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_755_transpose_x_0 = const()[name = tensor<string, []>("op_755_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_755_transpose_y_0 = const()[name = tensor<string, []>("op_755_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_112_perm_0 = const()[name = tensor<string, []>("transpose_112_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_113_perm_0 = const()[name = tensor<string, []>("transpose_113_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_113 = transpose(perm = transpose_113_perm_0, x = var_746_cast_fp16)[name = tensor<string, []>("transpose_206")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_112 = transpose(perm = transpose_112_perm_0, x = var_740_cast_fp16)[name = tensor<string, []>("transpose_207")]; |
| tensor<fp16, [1, 16, ?, ?]> var_755_cast_fp16 = matmul(transpose_x = var_755_transpose_x_0, transpose_y = var_755_transpose_y_0, x = transpose_112, y = transpose_113)[name = tensor<string, []>("op_755_cast_fp16")]; |
| tensor<fp16, []> var_756_to_fp16 = const()[name = tensor<string, []>("op_756_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_107_cast_fp16 = mul(x = var_755_cast_fp16, y = var_756_to_fp16)[name = tensor<string, []>("input_107_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_17_cast_fp16 = softmax(axis = var_716, x = input_107_cast_fp16)[name = tensor<string, []>("attn_17_cast_fp16")]; |
| tensor<bool, []> out_17_transpose_x_0 = const()[name = tensor<string, []>("out_17_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_17_transpose_y_0 = const()[name = tensor<string, []>("out_17_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_17_cast_fp16 = transpose(perm = v_17_perm_0, x = var_752_cast_fp16)[name = tensor<string, []>("transpose_205")]; |
| tensor<fp16, [1, 16, ?, 64]> out_17_cast_fp16 = matmul(transpose_x = out_17_transpose_x_0, transpose_y = out_17_transpose_y_0, x = attn_17_cast_fp16, y = v_17_cast_fp16)[name = tensor<string, []>("out_17_cast_fp16")]; |
| tensor<int32, [4]> var_760_perm_0 = const()[name = tensor<string, []>("op_760_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_37x = const()[name = tensor<string, []>("concat_37x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_760_cast_fp16 = transpose(perm = var_760_perm_0, x = out_17_cast_fp16)[name = tensor<string, []>("transpose_204")]; |
| tensor<fp16, [1, ?, 1024]> input_109_cast_fp16 = reshape(shape = concat_37x, x = var_760_cast_fp16)[name = tensor<string, []>("input_109_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_8_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58925120))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59449472))), name = tensor<string, []>("layers_8_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_8_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59449600)))]; |
| tensor<fp16, [1, ?, 1024]> linear_52_cast_fp16 = linear(bias = layers_8_self_attn_out_proj_bias_to_fp16, weight = layers_8_self_attn_out_proj_weight_to_fp16_palettized, x = input_109_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_111_cast_fp16 = add(x = input_105_cast_fp16, y = linear_52_cast_fp16)[name = tensor<string, []>("input_111_cast_fp16")]; |
| tensor<int32, [1]> input_113_axes_0 = const()[name = tensor<string, []>("input_113_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_8_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59451712)))]; |
| tensor<fp16, [1024]> layers_8_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59453824)))]; |
| tensor<fp16, [1, ?, 1024]> input_113_cast_fp16 = layer_norm(axes = input_113_axes_0, beta = layers_8_final_layer_norm_bias_to_fp16, epsilon = var_719_to_fp16, gamma = layers_8_final_layer_norm_weight_to_fp16, x = input_111_cast_fp16)[name = tensor<string, []>("input_113_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_8_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59455936))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61553152))), name = tensor<string, []>("layers_8_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_8_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61553280)))]; |
| tensor<fp16, [1, ?, 4096]> linear_53_cast_fp16 = linear(bias = layers_8_fc1_bias_to_fp16, weight = layers_8_fc1_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")]; |
| tensor<string, []> input_115_mode_0 = const()[name = tensor<string, []>("input_115_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_115_cast_fp16 = gelu(mode = input_115_mode_0, x = linear_53_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_8_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61561536))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63658752))), name = tensor<string, []>("layers_8_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_8_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63658880)))]; |
| tensor<fp16, [1, ?, 1024]> linear_54_cast_fp16 = linear(bias = layers_8_fc2_bias_to_fp16, weight = layers_8_fc2_weight_to_fp16_palettized, x = input_115_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_117_cast_fp16 = add(x = input_111_cast_fp16, y = linear_54_cast_fp16)[name = tensor<string, []>("input_117_cast_fp16")]; |
| tensor<int32, []> var_786 = const()[name = tensor<string, []>("op_786"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_59_axes_0 = const()[name = tensor<string, []>("x_59_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_9_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63660992)))]; |
| tensor<fp16, [1024]> layers_9_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63663104)))]; |
| tensor<fp16, []> var_789_to_fp16 = const()[name = tensor<string, []>("op_789_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_59_cast_fp16 = layer_norm(axes = x_59_axes_0, beta = layers_9_self_attn_layer_norm_bias_to_fp16, epsilon = var_789_to_fp16, gamma = layers_9_self_attn_layer_norm_weight_to_fp16, x = input_117_cast_fp16)[name = tensor<string, []>("x_59_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_9_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63665216))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64189568))), name = tensor<string, []>("layers_9_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64189696)))]; |
| tensor<fp16, [1, ?, 1024]> linear_55_cast_fp16 = linear(bias = layers_9_self_attn_q_proj_bias_to_fp16, weight = layers_9_self_attn_q_proj_weight_to_fp16_palettized, x = x_59_cast_fp16)[name = tensor<string, []>("linear_55_cast_fp16")]; |
| tensor<int32, [4]> concat_38x = const()[name = tensor<string, []>("concat_38x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_810_cast_fp16 = reshape(shape = concat_38x, x = linear_55_cast_fp16)[name = tensor<string, []>("op_810_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64191808))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64716160))), name = tensor<string, []>("layers_9_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_9_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64716288)))]; |
| tensor<fp16, [1, ?, 1024]> linear_56_cast_fp16 = linear(bias = layers_9_self_attn_k_proj_bias_to_fp16, weight = layers_9_self_attn_k_proj_weight_to_fp16_palettized, x = x_59_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")]; |
| tensor<int32, [4]> concat_39x = const()[name = tensor<string, []>("concat_39x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_816_cast_fp16 = reshape(shape = concat_39x, x = linear_56_cast_fp16)[name = tensor<string, []>("op_816_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_9_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64718400))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65242752))), name = tensor<string, []>("layers_9_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65242880)))]; |
| tensor<fp16, [1, ?, 1024]> linear_57_cast_fp16 = linear(bias = layers_9_self_attn_v_proj_bias_to_fp16, weight = layers_9_self_attn_v_proj_weight_to_fp16_palettized, x = x_59_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")]; |
| tensor<int32, [4]> concat_40x = const()[name = tensor<string, []>("concat_40x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_822_cast_fp16 = reshape(shape = concat_40x, x = linear_57_cast_fp16)[name = tensor<string, []>("op_822_cast_fp16")]; |
| tensor<int32, [4]> v_19_perm_0 = const()[name = tensor<string, []>("v_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_825_transpose_x_0 = const()[name = tensor<string, []>("op_825_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_825_transpose_y_0 = const()[name = tensor<string, []>("op_825_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_114_perm_0 = const()[name = tensor<string, []>("transpose_114_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_115_perm_0 = const()[name = tensor<string, []>("transpose_115_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_115 = transpose(perm = transpose_115_perm_0, x = var_816_cast_fp16)[name = tensor<string, []>("transpose_202")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_114 = transpose(perm = transpose_114_perm_0, x = var_810_cast_fp16)[name = tensor<string, []>("transpose_203")]; |
| tensor<fp16, [1, 16, ?, ?]> var_825_cast_fp16 = matmul(transpose_x = var_825_transpose_x_0, transpose_y = var_825_transpose_y_0, x = transpose_114, y = transpose_115)[name = tensor<string, []>("op_825_cast_fp16")]; |
| tensor<fp16, []> var_826_to_fp16 = const()[name = tensor<string, []>("op_826_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_119_cast_fp16 = mul(x = var_825_cast_fp16, y = var_826_to_fp16)[name = tensor<string, []>("input_119_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_19_cast_fp16 = softmax(axis = var_786, x = input_119_cast_fp16)[name = tensor<string, []>("attn_19_cast_fp16")]; |
| tensor<bool, []> out_19_transpose_x_0 = const()[name = tensor<string, []>("out_19_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_19_transpose_y_0 = const()[name = tensor<string, []>("out_19_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_19_cast_fp16 = transpose(perm = v_19_perm_0, x = var_822_cast_fp16)[name = tensor<string, []>("transpose_201")]; |
| tensor<fp16, [1, 16, ?, 64]> out_19_cast_fp16 = matmul(transpose_x = out_19_transpose_x_0, transpose_y = out_19_transpose_y_0, x = attn_19_cast_fp16, y = v_19_cast_fp16)[name = tensor<string, []>("out_19_cast_fp16")]; |
| tensor<int32, [4]> var_830_perm_0 = const()[name = tensor<string, []>("op_830_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_41x = const()[name = tensor<string, []>("concat_41x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_830_cast_fp16 = transpose(perm = var_830_perm_0, x = out_19_cast_fp16)[name = tensor<string, []>("transpose_200")]; |
| tensor<fp16, [1, ?, 1024]> input_121_cast_fp16 = reshape(shape = concat_41x, x = var_830_cast_fp16)[name = tensor<string, []>("input_121_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_9_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65244992))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65769344))), name = tensor<string, []>("layers_9_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_9_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65769472)))]; |
| tensor<fp16, [1, ?, 1024]> linear_58_cast_fp16 = linear(bias = layers_9_self_attn_out_proj_bias_to_fp16, weight = layers_9_self_attn_out_proj_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_123_cast_fp16 = add(x = input_117_cast_fp16, y = linear_58_cast_fp16)[name = tensor<string, []>("input_123_cast_fp16")]; |
| tensor<int32, [1]> input_125_axes_0 = const()[name = tensor<string, []>("input_125_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_9_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65771584)))]; |
| tensor<fp16, [1024]> layers_9_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65773696)))]; |
| tensor<fp16, [1, ?, 1024]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = layers_9_final_layer_norm_bias_to_fp16, epsilon = var_789_to_fp16, gamma = layers_9_final_layer_norm_weight_to_fp16, x = input_123_cast_fp16)[name = tensor<string, []>("input_125_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_9_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65775808))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67873024))), name = tensor<string, []>("layers_9_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_9_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67873152)))]; |
| tensor<fp16, [1, ?, 4096]> linear_59_cast_fp16 = linear(bias = layers_9_fc1_bias_to_fp16, weight = layers_9_fc1_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")]; |
| tensor<string, []> input_127_mode_0 = const()[name = tensor<string, []>("input_127_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_127_cast_fp16 = gelu(mode = input_127_mode_0, x = linear_59_cast_fp16)[name = tensor<string, []>("input_127_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_9_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67881408))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69978624))), name = tensor<string, []>("layers_9_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_9_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69978752)))]; |
| tensor<fp16, [1, ?, 1024]> linear_60_cast_fp16 = linear(bias = layers_9_fc2_bias_to_fp16, weight = layers_9_fc2_weight_to_fp16_palettized, x = input_127_cast_fp16)[name = tensor<string, []>("linear_60_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_129_cast_fp16 = add(x = input_123_cast_fp16, y = linear_60_cast_fp16)[name = tensor<string, []>("input_129_cast_fp16")]; |
| tensor<int32, []> var_856 = const()[name = tensor<string, []>("op_856"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_65_axes_0 = const()[name = tensor<string, []>("x_65_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_10_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69980864)))]; |
| tensor<fp16, [1024]> layers_10_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69982976)))]; |
| tensor<fp16, []> var_859_to_fp16 = const()[name = tensor<string, []>("op_859_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_65_cast_fp16 = layer_norm(axes = x_65_axes_0, beta = layers_10_self_attn_layer_norm_bias_to_fp16, epsilon = var_859_to_fp16, gamma = layers_10_self_attn_layer_norm_weight_to_fp16, x = input_129_cast_fp16)[name = tensor<string, []>("x_65_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_10_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69985088))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70509440))), name = tensor<string, []>("layers_10_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70509568)))]; |
| tensor<fp16, [1, ?, 1024]> linear_61_cast_fp16 = linear(bias = layers_10_self_attn_q_proj_bias_to_fp16, weight = layers_10_self_attn_q_proj_weight_to_fp16_palettized, x = x_65_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")]; |
| tensor<int32, [4]> concat_42x = const()[name = tensor<string, []>("concat_42x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_880_cast_fp16 = reshape(shape = concat_42x, x = linear_61_cast_fp16)[name = tensor<string, []>("op_880_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70511680))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71036032))), name = tensor<string, []>("layers_10_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_10_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71036160)))]; |
| tensor<fp16, [1, ?, 1024]> linear_62_cast_fp16 = linear(bias = layers_10_self_attn_k_proj_bias_to_fp16, weight = layers_10_self_attn_k_proj_weight_to_fp16_palettized, x = x_65_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")]; |
| tensor<int32, [4]> concat_43x = const()[name = tensor<string, []>("concat_43x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_886_cast_fp16 = reshape(shape = concat_43x, x = linear_62_cast_fp16)[name = tensor<string, []>("op_886_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_10_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71038272))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71562624))), name = tensor<string, []>("layers_10_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71562752)))]; |
| tensor<fp16, [1, ?, 1024]> linear_63_cast_fp16 = linear(bias = layers_10_self_attn_v_proj_bias_to_fp16, weight = layers_10_self_attn_v_proj_weight_to_fp16_palettized, x = x_65_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")]; |
| tensor<int32, [4]> concat_44x = const()[name = tensor<string, []>("concat_44x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_892_cast_fp16 = reshape(shape = concat_44x, x = linear_63_cast_fp16)[name = tensor<string, []>("op_892_cast_fp16")]; |
| tensor<int32, [4]> v_21_perm_0 = const()[name = tensor<string, []>("v_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_895_transpose_x_0 = const()[name = tensor<string, []>("op_895_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_895_transpose_y_0 = const()[name = tensor<string, []>("op_895_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_116_perm_0 = const()[name = tensor<string, []>("transpose_116_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_117_perm_0 = const()[name = tensor<string, []>("transpose_117_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_117 = transpose(perm = transpose_117_perm_0, x = var_886_cast_fp16)[name = tensor<string, []>("transpose_198")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_116 = transpose(perm = transpose_116_perm_0, x = var_880_cast_fp16)[name = tensor<string, []>("transpose_199")]; |
| tensor<fp16, [1, 16, ?, ?]> var_895_cast_fp16 = matmul(transpose_x = var_895_transpose_x_0, transpose_y = var_895_transpose_y_0, x = transpose_116, y = transpose_117)[name = tensor<string, []>("op_895_cast_fp16")]; |
| tensor<fp16, []> var_896_to_fp16 = const()[name = tensor<string, []>("op_896_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_131_cast_fp16 = mul(x = var_895_cast_fp16, y = var_896_to_fp16)[name = tensor<string, []>("input_131_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_21_cast_fp16 = softmax(axis = var_856, x = input_131_cast_fp16)[name = tensor<string, []>("attn_21_cast_fp16")]; |
| tensor<bool, []> out_21_transpose_x_0 = const()[name = tensor<string, []>("out_21_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_21_transpose_y_0 = const()[name = tensor<string, []>("out_21_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_21_cast_fp16 = transpose(perm = v_21_perm_0, x = var_892_cast_fp16)[name = tensor<string, []>("transpose_197")]; |
| tensor<fp16, [1, 16, ?, 64]> out_21_cast_fp16 = matmul(transpose_x = out_21_transpose_x_0, transpose_y = out_21_transpose_y_0, x = attn_21_cast_fp16, y = v_21_cast_fp16)[name = tensor<string, []>("out_21_cast_fp16")]; |
| tensor<int32, [4]> var_900_perm_0 = const()[name = tensor<string, []>("op_900_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_45x = const()[name = tensor<string, []>("concat_45x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_900_cast_fp16 = transpose(perm = var_900_perm_0, x = out_21_cast_fp16)[name = tensor<string, []>("transpose_196")]; |
| tensor<fp16, [1, ?, 1024]> input_133_cast_fp16 = reshape(shape = concat_45x, x = var_900_cast_fp16)[name = tensor<string, []>("input_133_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_10_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71564864))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72089216))), name = tensor<string, []>("layers_10_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_10_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72089344)))]; |
| tensor<fp16, [1, ?, 1024]> linear_64_cast_fp16 = linear(bias = layers_10_self_attn_out_proj_bias_to_fp16, weight = layers_10_self_attn_out_proj_weight_to_fp16_palettized, x = input_133_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_135_cast_fp16 = add(x = input_129_cast_fp16, y = linear_64_cast_fp16)[name = tensor<string, []>("input_135_cast_fp16")]; |
| tensor<int32, [1]> input_137_axes_0 = const()[name = tensor<string, []>("input_137_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_10_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72091456)))]; |
| tensor<fp16, [1024]> layers_10_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72093568)))]; |
| tensor<fp16, [1, ?, 1024]> input_137_cast_fp16 = layer_norm(axes = input_137_axes_0, beta = layers_10_final_layer_norm_bias_to_fp16, epsilon = var_859_to_fp16, gamma = layers_10_final_layer_norm_weight_to_fp16, x = input_135_cast_fp16)[name = tensor<string, []>("input_137_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_10_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72095680))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74192896))), name = tensor<string, []>("layers_10_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_10_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74193024)))]; |
| tensor<fp16, [1, ?, 4096]> linear_65_cast_fp16 = linear(bias = layers_10_fc1_bias_to_fp16, weight = layers_10_fc1_weight_to_fp16_palettized, x = input_137_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")]; |
| tensor<string, []> input_139_mode_0 = const()[name = tensor<string, []>("input_139_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_139_cast_fp16 = gelu(mode = input_139_mode_0, x = linear_65_cast_fp16)[name = tensor<string, []>("input_139_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_10_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74201280))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76298496))), name = tensor<string, []>("layers_10_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_10_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76298624)))]; |
| tensor<fp16, [1, ?, 1024]> linear_66_cast_fp16 = linear(bias = layers_10_fc2_bias_to_fp16, weight = layers_10_fc2_weight_to_fp16_palettized, x = input_139_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_141_cast_fp16 = add(x = input_135_cast_fp16, y = linear_66_cast_fp16)[name = tensor<string, []>("input_141_cast_fp16")]; |
| tensor<int32, []> var_926 = const()[name = tensor<string, []>("op_926"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_71_axes_0 = const()[name = tensor<string, []>("x_71_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_11_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76300736)))]; |
| tensor<fp16, [1024]> layers_11_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76302848)))]; |
| tensor<fp16, []> var_929_to_fp16 = const()[name = tensor<string, []>("op_929_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_71_cast_fp16 = layer_norm(axes = x_71_axes_0, beta = layers_11_self_attn_layer_norm_bias_to_fp16, epsilon = var_929_to_fp16, gamma = layers_11_self_attn_layer_norm_weight_to_fp16, x = input_141_cast_fp16)[name = tensor<string, []>("x_71_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_11_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76304960))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76829312))), name = tensor<string, []>("layers_11_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76829440)))]; |
| tensor<fp16, [1, ?, 1024]> linear_67_cast_fp16 = linear(bias = layers_11_self_attn_q_proj_bias_to_fp16, weight = layers_11_self_attn_q_proj_weight_to_fp16_palettized, x = x_71_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")]; |
| tensor<int32, [4]> concat_46x = const()[name = tensor<string, []>("concat_46x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_950_cast_fp16 = reshape(shape = concat_46x, x = linear_67_cast_fp16)[name = tensor<string, []>("op_950_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76831552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77355904))), name = tensor<string, []>("layers_11_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_11_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77356032)))]; |
| tensor<fp16, [1, ?, 1024]> linear_68_cast_fp16 = linear(bias = layers_11_self_attn_k_proj_bias_to_fp16, weight = layers_11_self_attn_k_proj_weight_to_fp16_palettized, x = x_71_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")]; |
| tensor<int32, [4]> concat_47x = const()[name = tensor<string, []>("concat_47x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_956_cast_fp16 = reshape(shape = concat_47x, x = linear_68_cast_fp16)[name = tensor<string, []>("op_956_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_11_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77358144))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77882496))), name = tensor<string, []>("layers_11_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77882624)))]; |
| tensor<fp16, [1, ?, 1024]> linear_69_cast_fp16 = linear(bias = layers_11_self_attn_v_proj_bias_to_fp16, weight = layers_11_self_attn_v_proj_weight_to_fp16_palettized, x = x_71_cast_fp16)[name = tensor<string, []>("linear_69_cast_fp16")]; |
| tensor<int32, [4]> concat_48x = const()[name = tensor<string, []>("concat_48x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_962_cast_fp16 = reshape(shape = concat_48x, x = linear_69_cast_fp16)[name = tensor<string, []>("op_962_cast_fp16")]; |
| tensor<int32, [4]> v_23_perm_0 = const()[name = tensor<string, []>("v_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_965_transpose_x_0 = const()[name = tensor<string, []>("op_965_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_965_transpose_y_0 = const()[name = tensor<string, []>("op_965_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_118_perm_0 = const()[name = tensor<string, []>("transpose_118_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_119_perm_0 = const()[name = tensor<string, []>("transpose_119_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_119 = transpose(perm = transpose_119_perm_0, x = var_956_cast_fp16)[name = tensor<string, []>("transpose_194")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_118 = transpose(perm = transpose_118_perm_0, x = var_950_cast_fp16)[name = tensor<string, []>("transpose_195")]; |
| tensor<fp16, [1, 16, ?, ?]> var_965_cast_fp16 = matmul(transpose_x = var_965_transpose_x_0, transpose_y = var_965_transpose_y_0, x = transpose_118, y = transpose_119)[name = tensor<string, []>("op_965_cast_fp16")]; |
| tensor<fp16, []> var_966_to_fp16 = const()[name = tensor<string, []>("op_966_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_143_cast_fp16 = mul(x = var_965_cast_fp16, y = var_966_to_fp16)[name = tensor<string, []>("input_143_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_23_cast_fp16 = softmax(axis = var_926, x = input_143_cast_fp16)[name = tensor<string, []>("attn_23_cast_fp16")]; |
| tensor<bool, []> out_23_transpose_x_0 = const()[name = tensor<string, []>("out_23_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_23_transpose_y_0 = const()[name = tensor<string, []>("out_23_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_23_cast_fp16 = transpose(perm = v_23_perm_0, x = var_962_cast_fp16)[name = tensor<string, []>("transpose_193")]; |
| tensor<fp16, [1, 16, ?, 64]> out_23_cast_fp16 = matmul(transpose_x = out_23_transpose_x_0, transpose_y = out_23_transpose_y_0, x = attn_23_cast_fp16, y = v_23_cast_fp16)[name = tensor<string, []>("out_23_cast_fp16")]; |
| tensor<int32, [4]> var_970_perm_0 = const()[name = tensor<string, []>("op_970_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_49x = const()[name = tensor<string, []>("concat_49x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_970_cast_fp16 = transpose(perm = var_970_perm_0, x = out_23_cast_fp16)[name = tensor<string, []>("transpose_192")]; |
| tensor<fp16, [1, ?, 1024]> input_145_cast_fp16 = reshape(shape = concat_49x, x = var_970_cast_fp16)[name = tensor<string, []>("input_145_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_11_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77884736))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78409088))), name = tensor<string, []>("layers_11_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_11_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78409216)))]; |
| tensor<fp16, [1, ?, 1024]> linear_70_cast_fp16 = linear(bias = layers_11_self_attn_out_proj_bias_to_fp16, weight = layers_11_self_attn_out_proj_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_147_cast_fp16 = add(x = input_141_cast_fp16, y = linear_70_cast_fp16)[name = tensor<string, []>("input_147_cast_fp16")]; |
| tensor<int32, [1]> input_149_axes_0 = const()[name = tensor<string, []>("input_149_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_11_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78411328)))]; |
| tensor<fp16, [1024]> layers_11_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78413440)))]; |
| tensor<fp16, [1, ?, 1024]> input_149_cast_fp16 = layer_norm(axes = input_149_axes_0, beta = layers_11_final_layer_norm_bias_to_fp16, epsilon = var_929_to_fp16, gamma = layers_11_final_layer_norm_weight_to_fp16, x = input_147_cast_fp16)[name = tensor<string, []>("input_149_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_11_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78415552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80512768))), name = tensor<string, []>("layers_11_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_11_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80512896)))]; |
| tensor<fp16, [1, ?, 4096]> linear_71_cast_fp16 = linear(bias = layers_11_fc1_bias_to_fp16, weight = layers_11_fc1_weight_to_fp16_palettized, x = input_149_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")]; |
| tensor<string, []> input_151_mode_0 = const()[name = tensor<string, []>("input_151_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_151_cast_fp16 = gelu(mode = input_151_mode_0, x = linear_71_cast_fp16)[name = tensor<string, []>("input_151_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_11_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80521152))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82618368))), name = tensor<string, []>("layers_11_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_11_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82618496)))]; |
| tensor<fp16, [1, ?, 1024]> linear_72_cast_fp16 = linear(bias = layers_11_fc2_bias_to_fp16, weight = layers_11_fc2_weight_to_fp16_palettized, x = input_151_cast_fp16)[name = tensor<string, []>("linear_72_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_153_cast_fp16 = add(x = input_147_cast_fp16, y = linear_72_cast_fp16)[name = tensor<string, []>("input_153_cast_fp16")]; |
| tensor<int32, []> var_996 = const()[name = tensor<string, []>("op_996"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_77_axes_0 = const()[name = tensor<string, []>("x_77_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_12_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_12_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82620608)))]; |
| tensor<fp16, [1024]> layers_12_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82622720)))]; |
| tensor<fp16, []> var_999_to_fp16 = const()[name = tensor<string, []>("op_999_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_77_cast_fp16 = layer_norm(axes = x_77_axes_0, beta = layers_12_self_attn_layer_norm_bias_to_fp16, epsilon = var_999_to_fp16, gamma = layers_12_self_attn_layer_norm_weight_to_fp16, x = input_153_cast_fp16)[name = tensor<string, []>("x_77_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_12_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82624832))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83149184))), name = tensor<string, []>("layers_12_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83149312)))]; |
| tensor<fp16, [1, ?, 1024]> linear_73_cast_fp16 = linear(bias = layers_12_self_attn_q_proj_bias_to_fp16, weight = layers_12_self_attn_q_proj_weight_to_fp16_palettized, x = x_77_cast_fp16)[name = tensor<string, []>("linear_73_cast_fp16")]; |
| tensor<int32, [4]> concat_50x = const()[name = tensor<string, []>("concat_50x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1020_cast_fp16 = reshape(shape = concat_50x, x = linear_73_cast_fp16)[name = tensor<string, []>("op_1020_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_12_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83151424))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83675776))), name = tensor<string, []>("layers_12_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_12_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83675904)))]; |
| tensor<fp16, [1, ?, 1024]> linear_74_cast_fp16 = linear(bias = layers_12_self_attn_k_proj_bias_to_fp16, weight = layers_12_self_attn_k_proj_weight_to_fp16_palettized, x = x_77_cast_fp16)[name = tensor<string, []>("linear_74_cast_fp16")]; |
| tensor<int32, [4]> concat_51x = const()[name = tensor<string, []>("concat_51x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1026_cast_fp16 = reshape(shape = concat_51x, x = linear_74_cast_fp16)[name = tensor<string, []>("op_1026_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_12_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83678016))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84202368))), name = tensor<string, []>("layers_12_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84202496)))]; |
| tensor<fp16, [1, ?, 1024]> linear_75_cast_fp16 = linear(bias = layers_12_self_attn_v_proj_bias_to_fp16, weight = layers_12_self_attn_v_proj_weight_to_fp16_palettized, x = x_77_cast_fp16)[name = tensor<string, []>("linear_75_cast_fp16")]; |
| tensor<int32, [4]> concat_52x = const()[name = tensor<string, []>("concat_52x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1032_cast_fp16 = reshape(shape = concat_52x, x = linear_75_cast_fp16)[name = tensor<string, []>("op_1032_cast_fp16")]; |
| tensor<int32, [4]> v_25_perm_0 = const()[name = tensor<string, []>("v_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_1035_transpose_x_0 = const()[name = tensor<string, []>("op_1035_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1035_transpose_y_0 = const()[name = tensor<string, []>("op_1035_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_120_perm_0 = const()[name = tensor<string, []>("transpose_120_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_121_perm_0 = const()[name = tensor<string, []>("transpose_121_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_121 = transpose(perm = transpose_121_perm_0, x = var_1026_cast_fp16)[name = tensor<string, []>("transpose_190")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_120 = transpose(perm = transpose_120_perm_0, x = var_1020_cast_fp16)[name = tensor<string, []>("transpose_191")]; |
| tensor<fp16, [1, 16, ?, ?]> var_1035_cast_fp16 = matmul(transpose_x = var_1035_transpose_x_0, transpose_y = var_1035_transpose_y_0, x = transpose_120, y = transpose_121)[name = tensor<string, []>("op_1035_cast_fp16")]; |
| tensor<fp16, []> var_1036_to_fp16 = const()[name = tensor<string, []>("op_1036_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_155_cast_fp16 = mul(x = var_1035_cast_fp16, y = var_1036_to_fp16)[name = tensor<string, []>("input_155_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_25_cast_fp16 = softmax(axis = var_996, x = input_155_cast_fp16)[name = tensor<string, []>("attn_25_cast_fp16")]; |
| tensor<bool, []> out_25_transpose_x_0 = const()[name = tensor<string, []>("out_25_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_25_transpose_y_0 = const()[name = tensor<string, []>("out_25_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_25_cast_fp16 = transpose(perm = v_25_perm_0, x = var_1032_cast_fp16)[name = tensor<string, []>("transpose_189")]; |
| tensor<fp16, [1, 16, ?, 64]> out_25_cast_fp16 = matmul(transpose_x = out_25_transpose_x_0, transpose_y = out_25_transpose_y_0, x = attn_25_cast_fp16, y = v_25_cast_fp16)[name = tensor<string, []>("out_25_cast_fp16")]; |
| tensor<int32, [4]> var_1040_perm_0 = const()[name = tensor<string, []>("op_1040_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_53x = const()[name = tensor<string, []>("concat_53x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1040_cast_fp16 = transpose(perm = var_1040_perm_0, x = out_25_cast_fp16)[name = tensor<string, []>("transpose_188")]; |
| tensor<fp16, [1, ?, 1024]> input_157_cast_fp16 = reshape(shape = concat_53x, x = var_1040_cast_fp16)[name = tensor<string, []>("input_157_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_12_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84204608))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84728960))), name = tensor<string, []>("layers_12_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_12_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84729088)))]; |
| tensor<fp16, [1, ?, 1024]> linear_76_cast_fp16 = linear(bias = layers_12_self_attn_out_proj_bias_to_fp16, weight = layers_12_self_attn_out_proj_weight_to_fp16_palettized, x = input_157_cast_fp16)[name = tensor<string, []>("linear_76_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_159_cast_fp16 = add(x = input_153_cast_fp16, y = linear_76_cast_fp16)[name = tensor<string, []>("input_159_cast_fp16")]; |
| tensor<int32, [1]> input_161_axes_0 = const()[name = tensor<string, []>("input_161_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_12_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_12_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84731200)))]; |
| tensor<fp16, [1024]> layers_12_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84733312)))]; |
| tensor<fp16, [1, ?, 1024]> input_161_cast_fp16 = layer_norm(axes = input_161_axes_0, beta = layers_12_final_layer_norm_bias_to_fp16, epsilon = var_999_to_fp16, gamma = layers_12_final_layer_norm_weight_to_fp16, x = input_159_cast_fp16)[name = tensor<string, []>("input_161_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_12_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84735424))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86832640))), name = tensor<string, []>("layers_12_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_12_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86832768)))]; |
| tensor<fp16, [1, ?, 4096]> linear_77_cast_fp16 = linear(bias = layers_12_fc1_bias_to_fp16, weight = layers_12_fc1_weight_to_fp16_palettized, x = input_161_cast_fp16)[name = tensor<string, []>("linear_77_cast_fp16")]; |
| tensor<string, []> input_163_mode_0 = const()[name = tensor<string, []>("input_163_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_163_cast_fp16 = gelu(mode = input_163_mode_0, x = linear_77_cast_fp16)[name = tensor<string, []>("input_163_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_12_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86841024))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88938240))), name = tensor<string, []>("layers_12_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_12_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88938368)))]; |
| tensor<fp16, [1, ?, 1024]> linear_78_cast_fp16 = linear(bias = layers_12_fc2_bias_to_fp16, weight = layers_12_fc2_weight_to_fp16_palettized, x = input_163_cast_fp16)[name = tensor<string, []>("linear_78_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_165_cast_fp16 = add(x = input_159_cast_fp16, y = linear_78_cast_fp16)[name = tensor<string, []>("input_165_cast_fp16")]; |
| tensor<int32, []> var_1066 = const()[name = tensor<string, []>("op_1066"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_83_axes_0 = const()[name = tensor<string, []>("x_83_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_13_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_13_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88940480)))]; |
| tensor<fp16, [1024]> layers_13_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88942592)))]; |
| tensor<fp16, []> var_1069_to_fp16 = const()[name = tensor<string, []>("op_1069_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_83_cast_fp16 = layer_norm(axes = x_83_axes_0, beta = layers_13_self_attn_layer_norm_bias_to_fp16, epsilon = var_1069_to_fp16, gamma = layers_13_self_attn_layer_norm_weight_to_fp16, x = input_165_cast_fp16)[name = tensor<string, []>("x_83_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_13_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88944704))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89469056))), name = tensor<string, []>("layers_13_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89469184)))]; |
| tensor<fp16, [1, ?, 1024]> linear_79_cast_fp16 = linear(bias = layers_13_self_attn_q_proj_bias_to_fp16, weight = layers_13_self_attn_q_proj_weight_to_fp16_palettized, x = x_83_cast_fp16)[name = tensor<string, []>("linear_79_cast_fp16")]; |
| tensor<int32, [4]> concat_54x = const()[name = tensor<string, []>("concat_54x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1090_cast_fp16 = reshape(shape = concat_54x, x = linear_79_cast_fp16)[name = tensor<string, []>("op_1090_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_13_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89471296))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89995648))), name = tensor<string, []>("layers_13_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_13_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89995776)))]; |
| tensor<fp16, [1, ?, 1024]> linear_80_cast_fp16 = linear(bias = layers_13_self_attn_k_proj_bias_to_fp16, weight = layers_13_self_attn_k_proj_weight_to_fp16_palettized, x = x_83_cast_fp16)[name = tensor<string, []>("linear_80_cast_fp16")]; |
| tensor<int32, [4]> concat_55x = const()[name = tensor<string, []>("concat_55x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1096_cast_fp16 = reshape(shape = concat_55x, x = linear_80_cast_fp16)[name = tensor<string, []>("op_1096_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_13_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89997888))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90522240))), name = tensor<string, []>("layers_13_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90522368)))]; |
| tensor<fp16, [1, ?, 1024]> linear_81_cast_fp16 = linear(bias = layers_13_self_attn_v_proj_bias_to_fp16, weight = layers_13_self_attn_v_proj_weight_to_fp16_palettized, x = x_83_cast_fp16)[name = tensor<string, []>("linear_81_cast_fp16")]; |
| tensor<int32, [4]> concat_56x = const()[name = tensor<string, []>("concat_56x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1102_cast_fp16 = reshape(shape = concat_56x, x = linear_81_cast_fp16)[name = tensor<string, []>("op_1102_cast_fp16")]; |
| tensor<int32, [4]> v_27_perm_0 = const()[name = tensor<string, []>("v_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_1105_transpose_x_0 = const()[name = tensor<string, []>("op_1105_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1105_transpose_y_0 = const()[name = tensor<string, []>("op_1105_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_122_perm_0 = const()[name = tensor<string, []>("transpose_122_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_123_perm_0 = const()[name = tensor<string, []>("transpose_123_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_123 = transpose(perm = transpose_123_perm_0, x = var_1096_cast_fp16)[name = tensor<string, []>("transpose_186")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_122 = transpose(perm = transpose_122_perm_0, x = var_1090_cast_fp16)[name = tensor<string, []>("transpose_187")]; |
| tensor<fp16, [1, 16, ?, ?]> var_1105_cast_fp16 = matmul(transpose_x = var_1105_transpose_x_0, transpose_y = var_1105_transpose_y_0, x = transpose_122, y = transpose_123)[name = tensor<string, []>("op_1105_cast_fp16")]; |
| tensor<fp16, []> var_1106_to_fp16 = const()[name = tensor<string, []>("op_1106_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_167_cast_fp16 = mul(x = var_1105_cast_fp16, y = var_1106_to_fp16)[name = tensor<string, []>("input_167_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_27_cast_fp16 = softmax(axis = var_1066, x = input_167_cast_fp16)[name = tensor<string, []>("attn_27_cast_fp16")]; |
| tensor<bool, []> out_27_transpose_x_0 = const()[name = tensor<string, []>("out_27_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_27_transpose_y_0 = const()[name = tensor<string, []>("out_27_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_27_cast_fp16 = transpose(perm = v_27_perm_0, x = var_1102_cast_fp16)[name = tensor<string, []>("transpose_185")]; |
| tensor<fp16, [1, 16, ?, 64]> out_27_cast_fp16 = matmul(transpose_x = out_27_transpose_x_0, transpose_y = out_27_transpose_y_0, x = attn_27_cast_fp16, y = v_27_cast_fp16)[name = tensor<string, []>("out_27_cast_fp16")]; |
| tensor<int32, [4]> var_1110_perm_0 = const()[name = tensor<string, []>("op_1110_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_57x = const()[name = tensor<string, []>("concat_57x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1110_cast_fp16 = transpose(perm = var_1110_perm_0, x = out_27_cast_fp16)[name = tensor<string, []>("transpose_184")]; |
| tensor<fp16, [1, ?, 1024]> input_169_cast_fp16 = reshape(shape = concat_57x, x = var_1110_cast_fp16)[name = tensor<string, []>("input_169_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_13_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90524480))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91048832))), name = tensor<string, []>("layers_13_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_13_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91048960)))]; |
| tensor<fp16, [1, ?, 1024]> linear_82_cast_fp16 = linear(bias = layers_13_self_attn_out_proj_bias_to_fp16, weight = layers_13_self_attn_out_proj_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = tensor<string, []>("linear_82_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_171_cast_fp16 = add(x = input_165_cast_fp16, y = linear_82_cast_fp16)[name = tensor<string, []>("input_171_cast_fp16")]; |
| tensor<int32, [1]> input_173_axes_0 = const()[name = tensor<string, []>("input_173_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_13_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_13_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91051072)))]; |
| tensor<fp16, [1024]> layers_13_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91053184)))]; |
| tensor<fp16, [1, ?, 1024]> input_173_cast_fp16 = layer_norm(axes = input_173_axes_0, beta = layers_13_final_layer_norm_bias_to_fp16, epsilon = var_1069_to_fp16, gamma = layers_13_final_layer_norm_weight_to_fp16, x = input_171_cast_fp16)[name = tensor<string, []>("input_173_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_13_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91055296))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93152512))), name = tensor<string, []>("layers_13_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_13_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93152640)))]; |
| tensor<fp16, [1, ?, 4096]> linear_83_cast_fp16 = linear(bias = layers_13_fc1_bias_to_fp16, weight = layers_13_fc1_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = tensor<string, []>("linear_83_cast_fp16")]; |
| tensor<string, []> input_175_mode_0 = const()[name = tensor<string, []>("input_175_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_175_cast_fp16 = gelu(mode = input_175_mode_0, x = linear_83_cast_fp16)[name = tensor<string, []>("input_175_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_13_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93160896))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95258112))), name = tensor<string, []>("layers_13_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_13_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95258240)))]; |
| tensor<fp16, [1, ?, 1024]> linear_84_cast_fp16 = linear(bias = layers_13_fc2_bias_to_fp16, weight = layers_13_fc2_weight_to_fp16_palettized, x = input_175_cast_fp16)[name = tensor<string, []>("linear_84_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_177_cast_fp16 = add(x = input_171_cast_fp16, y = linear_84_cast_fp16)[name = tensor<string, []>("input_177_cast_fp16")]; |
| tensor<int32, []> var_1136 = const()[name = tensor<string, []>("op_1136"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_89_axes_0 = const()[name = tensor<string, []>("x_89_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_14_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_14_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95260352)))]; |
| tensor<fp16, [1024]> layers_14_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95262464)))]; |
| tensor<fp16, []> var_1139_to_fp16 = const()[name = tensor<string, []>("op_1139_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_89_cast_fp16 = layer_norm(axes = x_89_axes_0, beta = layers_14_self_attn_layer_norm_bias_to_fp16, epsilon = var_1139_to_fp16, gamma = layers_14_self_attn_layer_norm_weight_to_fp16, x = input_177_cast_fp16)[name = tensor<string, []>("x_89_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_14_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95264576))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95788928))), name = tensor<string, []>("layers_14_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95789056)))]; |
| tensor<fp16, [1, ?, 1024]> linear_85_cast_fp16 = linear(bias = layers_14_self_attn_q_proj_bias_to_fp16, weight = layers_14_self_attn_q_proj_weight_to_fp16_palettized, x = x_89_cast_fp16)[name = tensor<string, []>("linear_85_cast_fp16")]; |
| tensor<int32, [4]> concat_58x = const()[name = tensor<string, []>("concat_58x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1160_cast_fp16 = reshape(shape = concat_58x, x = linear_85_cast_fp16)[name = tensor<string, []>("op_1160_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_14_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95791168))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96315520))), name = tensor<string, []>("layers_14_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_14_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96315648)))]; |
| tensor<fp16, [1, ?, 1024]> linear_86_cast_fp16 = linear(bias = layers_14_self_attn_k_proj_bias_to_fp16, weight = layers_14_self_attn_k_proj_weight_to_fp16_palettized, x = x_89_cast_fp16)[name = tensor<string, []>("linear_86_cast_fp16")]; |
| tensor<int32, [4]> concat_59x = const()[name = tensor<string, []>("concat_59x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1166_cast_fp16 = reshape(shape = concat_59x, x = linear_86_cast_fp16)[name = tensor<string, []>("op_1166_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_14_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96317760))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96842112))), name = tensor<string, []>("layers_14_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96842240)))]; |
| tensor<fp16, [1, ?, 1024]> linear_87_cast_fp16 = linear(bias = layers_14_self_attn_v_proj_bias_to_fp16, weight = layers_14_self_attn_v_proj_weight_to_fp16_palettized, x = x_89_cast_fp16)[name = tensor<string, []>("linear_87_cast_fp16")]; |
| tensor<int32, [4]> concat_60x = const()[name = tensor<string, []>("concat_60x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1172_cast_fp16 = reshape(shape = concat_60x, x = linear_87_cast_fp16)[name = tensor<string, []>("op_1172_cast_fp16")]; |
| tensor<int32, [4]> v_29_perm_0 = const()[name = tensor<string, []>("v_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_1175_transpose_x_0 = const()[name = tensor<string, []>("op_1175_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1175_transpose_y_0 = const()[name = tensor<string, []>("op_1175_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_124_perm_0 = const()[name = tensor<string, []>("transpose_124_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_125_perm_0 = const()[name = tensor<string, []>("transpose_125_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_125 = transpose(perm = transpose_125_perm_0, x = var_1166_cast_fp16)[name = tensor<string, []>("transpose_182")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_124 = transpose(perm = transpose_124_perm_0, x = var_1160_cast_fp16)[name = tensor<string, []>("transpose_183")]; |
| tensor<fp16, [1, 16, ?, ?]> var_1175_cast_fp16 = matmul(transpose_x = var_1175_transpose_x_0, transpose_y = var_1175_transpose_y_0, x = transpose_124, y = transpose_125)[name = tensor<string, []>("op_1175_cast_fp16")]; |
| tensor<fp16, []> var_1176_to_fp16 = const()[name = tensor<string, []>("op_1176_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_179_cast_fp16 = mul(x = var_1175_cast_fp16, y = var_1176_to_fp16)[name = tensor<string, []>("input_179_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_29_cast_fp16 = softmax(axis = var_1136, x = input_179_cast_fp16)[name = tensor<string, []>("attn_29_cast_fp16")]; |
| tensor<bool, []> out_29_transpose_x_0 = const()[name = tensor<string, []>("out_29_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_29_transpose_y_0 = const()[name = tensor<string, []>("out_29_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_29_cast_fp16 = transpose(perm = v_29_perm_0, x = var_1172_cast_fp16)[name = tensor<string, []>("transpose_181")]; |
| tensor<fp16, [1, 16, ?, 64]> out_29_cast_fp16 = matmul(transpose_x = out_29_transpose_x_0, transpose_y = out_29_transpose_y_0, x = attn_29_cast_fp16, y = v_29_cast_fp16)[name = tensor<string, []>("out_29_cast_fp16")]; |
| tensor<int32, [4]> var_1180_perm_0 = const()[name = tensor<string, []>("op_1180_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_61x = const()[name = tensor<string, []>("concat_61x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1180_cast_fp16 = transpose(perm = var_1180_perm_0, x = out_29_cast_fp16)[name = tensor<string, []>("transpose_180")]; |
| tensor<fp16, [1, ?, 1024]> input_181_cast_fp16 = reshape(shape = concat_61x, x = var_1180_cast_fp16)[name = tensor<string, []>("input_181_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_14_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96844352))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97368704))), name = tensor<string, []>("layers_14_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_14_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97368832)))]; |
| tensor<fp16, [1, ?, 1024]> linear_88_cast_fp16 = linear(bias = layers_14_self_attn_out_proj_bias_to_fp16, weight = layers_14_self_attn_out_proj_weight_to_fp16_palettized, x = input_181_cast_fp16)[name = tensor<string, []>("linear_88_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_183_cast_fp16 = add(x = input_177_cast_fp16, y = linear_88_cast_fp16)[name = tensor<string, []>("input_183_cast_fp16")]; |
| tensor<int32, [1]> input_185_axes_0 = const()[name = tensor<string, []>("input_185_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_14_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_14_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97370944)))]; |
| tensor<fp16, [1024]> layers_14_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97373056)))]; |
| tensor<fp16, [1, ?, 1024]> input_185_cast_fp16 = layer_norm(axes = input_185_axes_0, beta = layers_14_final_layer_norm_bias_to_fp16, epsilon = var_1139_to_fp16, gamma = layers_14_final_layer_norm_weight_to_fp16, x = input_183_cast_fp16)[name = tensor<string, []>("input_185_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_14_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97375168))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99472384))), name = tensor<string, []>("layers_14_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_14_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99472512)))]; |
| tensor<fp16, [1, ?, 4096]> linear_89_cast_fp16 = linear(bias = layers_14_fc1_bias_to_fp16, weight = layers_14_fc1_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = tensor<string, []>("linear_89_cast_fp16")]; |
| tensor<string, []> input_187_mode_0 = const()[name = tensor<string, []>("input_187_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_187_cast_fp16 = gelu(mode = input_187_mode_0, x = linear_89_cast_fp16)[name = tensor<string, []>("input_187_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_14_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99480768))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101577984))), name = tensor<string, []>("layers_14_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_14_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101578112)))]; |
| tensor<fp16, [1, ?, 1024]> linear_90_cast_fp16 = linear(bias = layers_14_fc2_bias_to_fp16, weight = layers_14_fc2_weight_to_fp16_palettized, x = input_187_cast_fp16)[name = tensor<string, []>("linear_90_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_189_cast_fp16 = add(x = input_183_cast_fp16, y = linear_90_cast_fp16)[name = tensor<string, []>("input_189_cast_fp16")]; |
| tensor<int32, []> var_1206 = const()[name = tensor<string, []>("op_1206"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_95_axes_0 = const()[name = tensor<string, []>("x_95_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_15_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_15_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101580224)))]; |
| tensor<fp16, [1024]> layers_15_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101582336)))]; |
| tensor<fp16, []> var_1209_to_fp16 = const()[name = tensor<string, []>("op_1209_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_95_cast_fp16 = layer_norm(axes = x_95_axes_0, beta = layers_15_self_attn_layer_norm_bias_to_fp16, epsilon = var_1209_to_fp16, gamma = layers_15_self_attn_layer_norm_weight_to_fp16, x = input_189_cast_fp16)[name = tensor<string, []>("x_95_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_15_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101584448))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102108800))), name = tensor<string, []>("layers_15_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102108928)))]; |
| tensor<fp16, [1, ?, 1024]> linear_91_cast_fp16 = linear(bias = layers_15_self_attn_q_proj_bias_to_fp16, weight = layers_15_self_attn_q_proj_weight_to_fp16_palettized, x = x_95_cast_fp16)[name = tensor<string, []>("linear_91_cast_fp16")]; |
| tensor<int32, [4]> concat_62x = const()[name = tensor<string, []>("concat_62x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1230_cast_fp16 = reshape(shape = concat_62x, x = linear_91_cast_fp16)[name = tensor<string, []>("op_1230_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_15_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102111040))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102635392))), name = tensor<string, []>("layers_15_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_15_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102635520)))]; |
| tensor<fp16, [1, ?, 1024]> linear_92_cast_fp16 = linear(bias = layers_15_self_attn_k_proj_bias_to_fp16, weight = layers_15_self_attn_k_proj_weight_to_fp16_palettized, x = x_95_cast_fp16)[name = tensor<string, []>("linear_92_cast_fp16")]; |
| tensor<int32, [4]> concat_63x = const()[name = tensor<string, []>("concat_63x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1236_cast_fp16 = reshape(shape = concat_63x, x = linear_92_cast_fp16)[name = tensor<string, []>("op_1236_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_15_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102637632))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103161984))), name = tensor<string, []>("layers_15_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103162112)))]; |
| tensor<fp16, [1, ?, 1024]> linear_93_cast_fp16 = linear(bias = layers_15_self_attn_v_proj_bias_to_fp16, weight = layers_15_self_attn_v_proj_weight_to_fp16_palettized, x = x_95_cast_fp16)[name = tensor<string, []>("linear_93_cast_fp16")]; |
| tensor<int32, [4]> concat_64x = const()[name = tensor<string, []>("concat_64x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1242_cast_fp16 = reshape(shape = concat_64x, x = linear_93_cast_fp16)[name = tensor<string, []>("op_1242_cast_fp16")]; |
| tensor<int32, [4]> v_31_perm_0 = const()[name = tensor<string, []>("v_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_1245_transpose_x_0 = const()[name = tensor<string, []>("op_1245_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1245_transpose_y_0 = const()[name = tensor<string, []>("op_1245_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_126_perm_0 = const()[name = tensor<string, []>("transpose_126_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_127_perm_0 = const()[name = tensor<string, []>("transpose_127_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_127 = transpose(perm = transpose_127_perm_0, x = var_1236_cast_fp16)[name = tensor<string, []>("transpose_178")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_126 = transpose(perm = transpose_126_perm_0, x = var_1230_cast_fp16)[name = tensor<string, []>("transpose_179")]; |
| tensor<fp16, [1, 16, ?, ?]> var_1245_cast_fp16 = matmul(transpose_x = var_1245_transpose_x_0, transpose_y = var_1245_transpose_y_0, x = transpose_126, y = transpose_127)[name = tensor<string, []>("op_1245_cast_fp16")]; |
| tensor<fp16, []> var_1246_to_fp16 = const()[name = tensor<string, []>("op_1246_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_191_cast_fp16 = mul(x = var_1245_cast_fp16, y = var_1246_to_fp16)[name = tensor<string, []>("input_191_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_31_cast_fp16 = softmax(axis = var_1206, x = input_191_cast_fp16)[name = tensor<string, []>("attn_31_cast_fp16")]; |
| tensor<bool, []> out_31_transpose_x_0 = const()[name = tensor<string, []>("out_31_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_31_transpose_y_0 = const()[name = tensor<string, []>("out_31_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_31_cast_fp16 = transpose(perm = v_31_perm_0, x = var_1242_cast_fp16)[name = tensor<string, []>("transpose_177")]; |
| tensor<fp16, [1, 16, ?, 64]> out_31_cast_fp16 = matmul(transpose_x = out_31_transpose_x_0, transpose_y = out_31_transpose_y_0, x = attn_31_cast_fp16, y = v_31_cast_fp16)[name = tensor<string, []>("out_31_cast_fp16")]; |
| tensor<int32, [4]> var_1250_perm_0 = const()[name = tensor<string, []>("op_1250_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_65x = const()[name = tensor<string, []>("concat_65x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1250_cast_fp16 = transpose(perm = var_1250_perm_0, x = out_31_cast_fp16)[name = tensor<string, []>("transpose_176")]; |
| tensor<fp16, [1, ?, 1024]> input_193_cast_fp16 = reshape(shape = concat_65x, x = var_1250_cast_fp16)[name = tensor<string, []>("input_193_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_15_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103164224))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103688576))), name = tensor<string, []>("layers_15_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_15_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103688704)))]; |
| tensor<fp16, [1, ?, 1024]> linear_94_cast_fp16 = linear(bias = layers_15_self_attn_out_proj_bias_to_fp16, weight = layers_15_self_attn_out_proj_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = tensor<string, []>("linear_94_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_195_cast_fp16 = add(x = input_189_cast_fp16, y = linear_94_cast_fp16)[name = tensor<string, []>("input_195_cast_fp16")]; |
| tensor<int32, [1]> input_197_axes_0 = const()[name = tensor<string, []>("input_197_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_15_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_15_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103690816)))]; |
| tensor<fp16, [1024]> layers_15_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103692928)))]; |
| tensor<fp16, [1, ?, 1024]> input_197_cast_fp16 = layer_norm(axes = input_197_axes_0, beta = layers_15_final_layer_norm_bias_to_fp16, epsilon = var_1209_to_fp16, gamma = layers_15_final_layer_norm_weight_to_fp16, x = input_195_cast_fp16)[name = tensor<string, []>("input_197_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_15_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103695040))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105792256))), name = tensor<string, []>("layers_15_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_15_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105792384)))]; |
| tensor<fp16, [1, ?, 4096]> linear_95_cast_fp16 = linear(bias = layers_15_fc1_bias_to_fp16, weight = layers_15_fc1_weight_to_fp16_palettized, x = input_197_cast_fp16)[name = tensor<string, []>("linear_95_cast_fp16")]; |
| tensor<string, []> input_199_mode_0 = const()[name = tensor<string, []>("input_199_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_199_cast_fp16 = gelu(mode = input_199_mode_0, x = linear_95_cast_fp16)[name = tensor<string, []>("input_199_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_15_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105800640))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107897856))), name = tensor<string, []>("layers_15_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_15_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107897984)))]; |
| tensor<fp16, [1, ?, 1024]> linear_96_cast_fp16 = linear(bias = layers_15_fc2_bias_to_fp16, weight = layers_15_fc2_weight_to_fp16_palettized, x = input_199_cast_fp16)[name = tensor<string, []>("linear_96_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_201_cast_fp16 = add(x = input_195_cast_fp16, y = linear_96_cast_fp16)[name = tensor<string, []>("input_201_cast_fp16")]; |
| tensor<int32, []> var_1276 = const()[name = tensor<string, []>("op_1276"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_101_axes_0 = const()[name = tensor<string, []>("x_101_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_16_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_16_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107900096)))]; |
| tensor<fp16, [1024]> layers_16_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107902208)))]; |
| tensor<fp16, []> var_1279_to_fp16 = const()[name = tensor<string, []>("op_1279_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_101_cast_fp16 = layer_norm(axes = x_101_axes_0, beta = layers_16_self_attn_layer_norm_bias_to_fp16, epsilon = var_1279_to_fp16, gamma = layers_16_self_attn_layer_norm_weight_to_fp16, x = input_201_cast_fp16)[name = tensor<string, []>("x_101_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_16_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107904320))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108428672))), name = tensor<string, []>("layers_16_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108428800)))]; |
| tensor<fp16, [1, ?, 1024]> linear_97_cast_fp16 = linear(bias = layers_16_self_attn_q_proj_bias_to_fp16, weight = layers_16_self_attn_q_proj_weight_to_fp16_palettized, x = x_101_cast_fp16)[name = tensor<string, []>("linear_97_cast_fp16")]; |
| tensor<int32, [4]> concat_66x = const()[name = tensor<string, []>("concat_66x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1300_cast_fp16 = reshape(shape = concat_66x, x = linear_97_cast_fp16)[name = tensor<string, []>("op_1300_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_16_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108430912))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108955264))), name = tensor<string, []>("layers_16_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_16_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108955392)))]; |
| tensor<fp16, [1, ?, 1024]> linear_98_cast_fp16 = linear(bias = layers_16_self_attn_k_proj_bias_to_fp16, weight = layers_16_self_attn_k_proj_weight_to_fp16_palettized, x = x_101_cast_fp16)[name = tensor<string, []>("linear_98_cast_fp16")]; |
| tensor<int32, [4]> concat_67x = const()[name = tensor<string, []>("concat_67x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1306_cast_fp16 = reshape(shape = concat_67x, x = linear_98_cast_fp16)[name = tensor<string, []>("op_1306_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_16_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108957504))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109481856))), name = tensor<string, []>("layers_16_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109481984)))]; |
| tensor<fp16, [1, ?, 1024]> linear_99_cast_fp16 = linear(bias = layers_16_self_attn_v_proj_bias_to_fp16, weight = layers_16_self_attn_v_proj_weight_to_fp16_palettized, x = x_101_cast_fp16)[name = tensor<string, []>("linear_99_cast_fp16")]; |
| tensor<int32, [4]> concat_68x = const()[name = tensor<string, []>("concat_68x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1312_cast_fp16 = reshape(shape = concat_68x, x = linear_99_cast_fp16)[name = tensor<string, []>("op_1312_cast_fp16")]; |
| tensor<int32, [4]> v_33_perm_0 = const()[name = tensor<string, []>("v_33_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_1315_transpose_x_0 = const()[name = tensor<string, []>("op_1315_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1315_transpose_y_0 = const()[name = tensor<string, []>("op_1315_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_128_perm_0 = const()[name = tensor<string, []>("transpose_128_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_129_perm_0 = const()[name = tensor<string, []>("transpose_129_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_129 = transpose(perm = transpose_129_perm_0, x = var_1306_cast_fp16)[name = tensor<string, []>("transpose_174")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_128 = transpose(perm = transpose_128_perm_0, x = var_1300_cast_fp16)[name = tensor<string, []>("transpose_175")]; |
| tensor<fp16, [1, 16, ?, ?]> var_1315_cast_fp16 = matmul(transpose_x = var_1315_transpose_x_0, transpose_y = var_1315_transpose_y_0, x = transpose_128, y = transpose_129)[name = tensor<string, []>("op_1315_cast_fp16")]; |
| tensor<fp16, []> var_1316_to_fp16 = const()[name = tensor<string, []>("op_1316_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_203_cast_fp16 = mul(x = var_1315_cast_fp16, y = var_1316_to_fp16)[name = tensor<string, []>("input_203_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_33_cast_fp16 = softmax(axis = var_1276, x = input_203_cast_fp16)[name = tensor<string, []>("attn_33_cast_fp16")]; |
| tensor<bool, []> out_33_transpose_x_0 = const()[name = tensor<string, []>("out_33_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_33_transpose_y_0 = const()[name = tensor<string, []>("out_33_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_33_cast_fp16 = transpose(perm = v_33_perm_0, x = var_1312_cast_fp16)[name = tensor<string, []>("transpose_173")]; |
| tensor<fp16, [1, 16, ?, 64]> out_33_cast_fp16 = matmul(transpose_x = out_33_transpose_x_0, transpose_y = out_33_transpose_y_0, x = attn_33_cast_fp16, y = v_33_cast_fp16)[name = tensor<string, []>("out_33_cast_fp16")]; |
| tensor<int32, [4]> var_1320_perm_0 = const()[name = tensor<string, []>("op_1320_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_69x = const()[name = tensor<string, []>("concat_69x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1320_cast_fp16 = transpose(perm = var_1320_perm_0, x = out_33_cast_fp16)[name = tensor<string, []>("transpose_172")]; |
| tensor<fp16, [1, ?, 1024]> input_205_cast_fp16 = reshape(shape = concat_69x, x = var_1320_cast_fp16)[name = tensor<string, []>("input_205_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_16_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109484096))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110008448))), name = tensor<string, []>("layers_16_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_16_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110008576)))]; |
| tensor<fp16, [1, ?, 1024]> linear_100_cast_fp16 = linear(bias = layers_16_self_attn_out_proj_bias_to_fp16, weight = layers_16_self_attn_out_proj_weight_to_fp16_palettized, x = input_205_cast_fp16)[name = tensor<string, []>("linear_100_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_207_cast_fp16 = add(x = input_201_cast_fp16, y = linear_100_cast_fp16)[name = tensor<string, []>("input_207_cast_fp16")]; |
| tensor<int32, [1]> input_209_axes_0 = const()[name = tensor<string, []>("input_209_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_16_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_16_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110010688)))]; |
| tensor<fp16, [1024]> layers_16_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110012800)))]; |
| tensor<fp16, [1, ?, 1024]> input_209_cast_fp16 = layer_norm(axes = input_209_axes_0, beta = layers_16_final_layer_norm_bias_to_fp16, epsilon = var_1279_to_fp16, gamma = layers_16_final_layer_norm_weight_to_fp16, x = input_207_cast_fp16)[name = tensor<string, []>("input_209_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_16_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110014912))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112112128))), name = tensor<string, []>("layers_16_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_16_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112112256)))]; |
| tensor<fp16, [1, ?, 4096]> linear_101_cast_fp16 = linear(bias = layers_16_fc1_bias_to_fp16, weight = layers_16_fc1_weight_to_fp16_palettized, x = input_209_cast_fp16)[name = tensor<string, []>("linear_101_cast_fp16")]; |
| tensor<string, []> input_211_mode_0 = const()[name = tensor<string, []>("input_211_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_211_cast_fp16 = gelu(mode = input_211_mode_0, x = linear_101_cast_fp16)[name = tensor<string, []>("input_211_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_16_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112120512))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114217728))), name = tensor<string, []>("layers_16_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_16_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114217856)))]; |
| tensor<fp16, [1, ?, 1024]> linear_102_cast_fp16 = linear(bias = layers_16_fc2_bias_to_fp16, weight = layers_16_fc2_weight_to_fp16_palettized, x = input_211_cast_fp16)[name = tensor<string, []>("linear_102_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_213_cast_fp16 = add(x = input_207_cast_fp16, y = linear_102_cast_fp16)[name = tensor<string, []>("input_213_cast_fp16")]; |
| tensor<int32, []> var_1346 = const()[name = tensor<string, []>("op_1346"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_107_axes_0 = const()[name = tensor<string, []>("x_107_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_17_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_17_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114219968)))]; |
| tensor<fp16, [1024]> layers_17_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114222080)))]; |
| tensor<fp16, []> var_1349_to_fp16 = const()[name = tensor<string, []>("op_1349_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_107_cast_fp16 = layer_norm(axes = x_107_axes_0, beta = layers_17_self_attn_layer_norm_bias_to_fp16, epsilon = var_1349_to_fp16, gamma = layers_17_self_attn_layer_norm_weight_to_fp16, x = input_213_cast_fp16)[name = tensor<string, []>("x_107_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_17_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114224192))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114748544))), name = tensor<string, []>("layers_17_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114748672)))]; |
| tensor<fp16, [1, ?, 1024]> linear_103_cast_fp16 = linear(bias = layers_17_self_attn_q_proj_bias_to_fp16, weight = layers_17_self_attn_q_proj_weight_to_fp16_palettized, x = x_107_cast_fp16)[name = tensor<string, []>("linear_103_cast_fp16")]; |
| tensor<int32, [4]> concat_70x = const()[name = tensor<string, []>("concat_70x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1370_cast_fp16 = reshape(shape = concat_70x, x = linear_103_cast_fp16)[name = tensor<string, []>("op_1370_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_17_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114750784))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115275136))), name = tensor<string, []>("layers_17_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_17_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115275264)))]; |
| tensor<fp16, [1, ?, 1024]> linear_104_cast_fp16 = linear(bias = layers_17_self_attn_k_proj_bias_to_fp16, weight = layers_17_self_attn_k_proj_weight_to_fp16_palettized, x = x_107_cast_fp16)[name = tensor<string, []>("linear_104_cast_fp16")]; |
| tensor<int32, [4]> concat_71x = const()[name = tensor<string, []>("concat_71x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1376_cast_fp16 = reshape(shape = concat_71x, x = linear_104_cast_fp16)[name = tensor<string, []>("op_1376_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_17_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115277376))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115801728))), name = tensor<string, []>("layers_17_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115801856)))]; |
| tensor<fp16, [1, ?, 1024]> linear_105_cast_fp16 = linear(bias = layers_17_self_attn_v_proj_bias_to_fp16, weight = layers_17_self_attn_v_proj_weight_to_fp16_palettized, x = x_107_cast_fp16)[name = tensor<string, []>("linear_105_cast_fp16")]; |
| tensor<int32, [4]> concat_72x = const()[name = tensor<string, []>("concat_72x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1382_cast_fp16 = reshape(shape = concat_72x, x = linear_105_cast_fp16)[name = tensor<string, []>("op_1382_cast_fp16")]; |
| tensor<int32, [4]> v_35_perm_0 = const()[name = tensor<string, []>("v_35_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_1385_transpose_x_0 = const()[name = tensor<string, []>("op_1385_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1385_transpose_y_0 = const()[name = tensor<string, []>("op_1385_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_130_perm_0 = const()[name = tensor<string, []>("transpose_130_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_131_perm_0 = const()[name = tensor<string, []>("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_131 = transpose(perm = transpose_131_perm_0, x = var_1376_cast_fp16)[name = tensor<string, []>("transpose_170")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_130 = transpose(perm = transpose_130_perm_0, x = var_1370_cast_fp16)[name = tensor<string, []>("transpose_171")]; |
| tensor<fp16, [1, 16, ?, ?]> var_1385_cast_fp16 = matmul(transpose_x = var_1385_transpose_x_0, transpose_y = var_1385_transpose_y_0, x = transpose_130, y = transpose_131)[name = tensor<string, []>("op_1385_cast_fp16")]; |
| tensor<fp16, []> var_1386_to_fp16 = const()[name = tensor<string, []>("op_1386_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_215_cast_fp16 = mul(x = var_1385_cast_fp16, y = var_1386_to_fp16)[name = tensor<string, []>("input_215_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_35_cast_fp16 = softmax(axis = var_1346, x = input_215_cast_fp16)[name = tensor<string, []>("attn_35_cast_fp16")]; |
| tensor<bool, []> out_35_transpose_x_0 = const()[name = tensor<string, []>("out_35_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_35_transpose_y_0 = const()[name = tensor<string, []>("out_35_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_35_cast_fp16 = transpose(perm = v_35_perm_0, x = var_1382_cast_fp16)[name = tensor<string, []>("transpose_169")]; |
| tensor<fp16, [1, 16, ?, 64]> out_35_cast_fp16 = matmul(transpose_x = out_35_transpose_x_0, transpose_y = out_35_transpose_y_0, x = attn_35_cast_fp16, y = v_35_cast_fp16)[name = tensor<string, []>("out_35_cast_fp16")]; |
| tensor<int32, [4]> var_1390_perm_0 = const()[name = tensor<string, []>("op_1390_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_73x = const()[name = tensor<string, []>("concat_73x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1390_cast_fp16 = transpose(perm = var_1390_perm_0, x = out_35_cast_fp16)[name = tensor<string, []>("transpose_168")]; |
| tensor<fp16, [1, ?, 1024]> input_217_cast_fp16 = reshape(shape = concat_73x, x = var_1390_cast_fp16)[name = tensor<string, []>("input_217_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_17_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115803968))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116328320))), name = tensor<string, []>("layers_17_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_17_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116328448)))]; |
| tensor<fp16, [1, ?, 1024]> linear_106_cast_fp16 = linear(bias = layers_17_self_attn_out_proj_bias_to_fp16, weight = layers_17_self_attn_out_proj_weight_to_fp16_palettized, x = input_217_cast_fp16)[name = tensor<string, []>("linear_106_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_219_cast_fp16 = add(x = input_213_cast_fp16, y = linear_106_cast_fp16)[name = tensor<string, []>("input_219_cast_fp16")]; |
| tensor<int32, [1]> input_221_axes_0 = const()[name = tensor<string, []>("input_221_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_17_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_17_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116330560)))]; |
| tensor<fp16, [1024]> layers_17_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116332672)))]; |
| tensor<fp16, [1, ?, 1024]> input_221_cast_fp16 = layer_norm(axes = input_221_axes_0, beta = layers_17_final_layer_norm_bias_to_fp16, epsilon = var_1349_to_fp16, gamma = layers_17_final_layer_norm_weight_to_fp16, x = input_219_cast_fp16)[name = tensor<string, []>("input_221_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_17_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116334784))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118432000))), name = tensor<string, []>("layers_17_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_17_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118432128)))]; |
| tensor<fp16, [1, ?, 4096]> linear_107_cast_fp16 = linear(bias = layers_17_fc1_bias_to_fp16, weight = layers_17_fc1_weight_to_fp16_palettized, x = input_221_cast_fp16)[name = tensor<string, []>("linear_107_cast_fp16")]; |
| tensor<string, []> input_223_mode_0 = const()[name = tensor<string, []>("input_223_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_223_cast_fp16 = gelu(mode = input_223_mode_0, x = linear_107_cast_fp16)[name = tensor<string, []>("input_223_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_17_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118440384))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120537600))), name = tensor<string, []>("layers_17_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_17_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120537728)))]; |
| tensor<fp16, [1, ?, 1024]> linear_108_cast_fp16 = linear(bias = layers_17_fc2_bias_to_fp16, weight = layers_17_fc2_weight_to_fp16_palettized, x = input_223_cast_fp16)[name = tensor<string, []>("linear_108_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_225_cast_fp16 = add(x = input_219_cast_fp16, y = linear_108_cast_fp16)[name = tensor<string, []>("input_225_cast_fp16")]; |
| tensor<int32, []> var_1416 = const()[name = tensor<string, []>("op_1416"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_113_axes_0 = const()[name = tensor<string, []>("x_113_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_18_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_18_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120539840)))]; |
| tensor<fp16, [1024]> layers_18_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_18_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120541952)))]; |
| tensor<fp16, []> var_1419_to_fp16 = const()[name = tensor<string, []>("op_1419_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_113_cast_fp16 = layer_norm(axes = x_113_axes_0, beta = layers_18_self_attn_layer_norm_bias_to_fp16, epsilon = var_1419_to_fp16, gamma = layers_18_self_attn_layer_norm_weight_to_fp16, x = input_225_cast_fp16)[name = tensor<string, []>("x_113_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_18_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120544064))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121068416))), name = tensor<string, []>("layers_18_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121068544)))]; |
| tensor<fp16, [1, ?, 1024]> linear_109_cast_fp16 = linear(bias = layers_18_self_attn_q_proj_bias_to_fp16, weight = layers_18_self_attn_q_proj_weight_to_fp16_palettized, x = x_113_cast_fp16)[name = tensor<string, []>("linear_109_cast_fp16")]; |
| tensor<int32, [4]> concat_74x = const()[name = tensor<string, []>("concat_74x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1440_cast_fp16 = reshape(shape = concat_74x, x = linear_109_cast_fp16)[name = tensor<string, []>("op_1440_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_18_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121070656))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121595008))), name = tensor<string, []>("layers_18_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_18_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_18_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121595136)))]; |
| tensor<fp16, [1, ?, 1024]> linear_110_cast_fp16 = linear(bias = layers_18_self_attn_k_proj_bias_to_fp16, weight = layers_18_self_attn_k_proj_weight_to_fp16_palettized, x = x_113_cast_fp16)[name = tensor<string, []>("linear_110_cast_fp16")]; |
| tensor<int32, [4]> concat_75x = const()[name = tensor<string, []>("concat_75x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1446_cast_fp16 = reshape(shape = concat_75x, x = linear_110_cast_fp16)[name = tensor<string, []>("op_1446_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_18_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121597248))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122121600))), name = tensor<string, []>("layers_18_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122121728)))]; |
| tensor<fp16, [1, ?, 1024]> linear_111_cast_fp16 = linear(bias = layers_18_self_attn_v_proj_bias_to_fp16, weight = layers_18_self_attn_v_proj_weight_to_fp16_palettized, x = x_113_cast_fp16)[name = tensor<string, []>("linear_111_cast_fp16")]; |
| tensor<int32, [4]> concat_76x = const()[name = tensor<string, []>("concat_76x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1452_cast_fp16 = reshape(shape = concat_76x, x = linear_111_cast_fp16)[name = tensor<string, []>("op_1452_cast_fp16")]; |
| tensor<int32, [4]> v_37_perm_0 = const()[name = tensor<string, []>("v_37_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_1455_transpose_x_0 = const()[name = tensor<string, []>("op_1455_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1455_transpose_y_0 = const()[name = tensor<string, []>("op_1455_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_132_perm_0 = const()[name = tensor<string, []>("transpose_132_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_133_perm_0 = const()[name = tensor<string, []>("transpose_133_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_133 = transpose(perm = transpose_133_perm_0, x = var_1446_cast_fp16)[name = tensor<string, []>("transpose_166")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_132 = transpose(perm = transpose_132_perm_0, x = var_1440_cast_fp16)[name = tensor<string, []>("transpose_167")]; |
| tensor<fp16, [1, 16, ?, ?]> var_1455_cast_fp16 = matmul(transpose_x = var_1455_transpose_x_0, transpose_y = var_1455_transpose_y_0, x = transpose_132, y = transpose_133)[name = tensor<string, []>("op_1455_cast_fp16")]; |
| tensor<fp16, []> var_1456_to_fp16 = const()[name = tensor<string, []>("op_1456_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_227_cast_fp16 = mul(x = var_1455_cast_fp16, y = var_1456_to_fp16)[name = tensor<string, []>("input_227_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_37_cast_fp16 = softmax(axis = var_1416, x = input_227_cast_fp16)[name = tensor<string, []>("attn_37_cast_fp16")]; |
| tensor<bool, []> out_37_transpose_x_0 = const()[name = tensor<string, []>("out_37_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_37_transpose_y_0 = const()[name = tensor<string, []>("out_37_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_37_cast_fp16 = transpose(perm = v_37_perm_0, x = var_1452_cast_fp16)[name = tensor<string, []>("transpose_165")]; |
| tensor<fp16, [1, 16, ?, 64]> out_37_cast_fp16 = matmul(transpose_x = out_37_transpose_x_0, transpose_y = out_37_transpose_y_0, x = attn_37_cast_fp16, y = v_37_cast_fp16)[name = tensor<string, []>("out_37_cast_fp16")]; |
| tensor<int32, [4]> var_1460_perm_0 = const()[name = tensor<string, []>("op_1460_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_77x = const()[name = tensor<string, []>("concat_77x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1460_cast_fp16 = transpose(perm = var_1460_perm_0, x = out_37_cast_fp16)[name = tensor<string, []>("transpose_164")]; |
| tensor<fp16, [1, ?, 1024]> input_229_cast_fp16 = reshape(shape = concat_77x, x = var_1460_cast_fp16)[name = tensor<string, []>("input_229_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_18_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122123840))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122648192))), name = tensor<string, []>("layers_18_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_18_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_18_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122648320)))]; |
| tensor<fp16, [1, ?, 1024]> linear_112_cast_fp16 = linear(bias = layers_18_self_attn_out_proj_bias_to_fp16, weight = layers_18_self_attn_out_proj_weight_to_fp16_palettized, x = input_229_cast_fp16)[name = tensor<string, []>("linear_112_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_231_cast_fp16 = add(x = input_225_cast_fp16, y = linear_112_cast_fp16)[name = tensor<string, []>("input_231_cast_fp16")]; |
| tensor<int32, [1]> input_233_axes_0 = const()[name = tensor<string, []>("input_233_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_18_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_18_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122650432)))]; |
| tensor<fp16, [1024]> layers_18_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_18_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122652544)))]; |
| tensor<fp16, [1, ?, 1024]> input_233_cast_fp16 = layer_norm(axes = input_233_axes_0, beta = layers_18_final_layer_norm_bias_to_fp16, epsilon = var_1419_to_fp16, gamma = layers_18_final_layer_norm_weight_to_fp16, x = input_231_cast_fp16)[name = tensor<string, []>("input_233_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_18_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122654656))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124751872))), name = tensor<string, []>("layers_18_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_18_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_18_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124752000)))]; |
| tensor<fp16, [1, ?, 4096]> linear_113_cast_fp16 = linear(bias = layers_18_fc1_bias_to_fp16, weight = layers_18_fc1_weight_to_fp16_palettized, x = input_233_cast_fp16)[name = tensor<string, []>("linear_113_cast_fp16")]; |
| tensor<string, []> input_235_mode_0 = const()[name = tensor<string, []>("input_235_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_235_cast_fp16 = gelu(mode = input_235_mode_0, x = linear_113_cast_fp16)[name = tensor<string, []>("input_235_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_18_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124760256))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126857472))), name = tensor<string, []>("layers_18_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_18_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_18_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126857600)))]; |
| tensor<fp16, [1, ?, 1024]> linear_114_cast_fp16 = linear(bias = layers_18_fc2_bias_to_fp16, weight = layers_18_fc2_weight_to_fp16_palettized, x = input_235_cast_fp16)[name = tensor<string, []>("linear_114_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_237_cast_fp16 = add(x = input_231_cast_fp16, y = linear_114_cast_fp16)[name = tensor<string, []>("input_237_cast_fp16")]; |
| tensor<int32, []> var_1486 = const()[name = tensor<string, []>("op_1486"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_119_axes_0 = const()[name = tensor<string, []>("x_119_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_19_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_19_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126859712)))]; |
| tensor<fp16, [1024]> layers_19_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_19_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126861824)))]; |
| tensor<fp16, []> var_1489_to_fp16 = const()[name = tensor<string, []>("op_1489_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_119_cast_fp16 = layer_norm(axes = x_119_axes_0, beta = layers_19_self_attn_layer_norm_bias_to_fp16, epsilon = var_1489_to_fp16, gamma = layers_19_self_attn_layer_norm_weight_to_fp16, x = input_237_cast_fp16)[name = tensor<string, []>("x_119_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_19_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126863936))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127388288))), name = tensor<string, []>("layers_19_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127388416)))]; |
| tensor<fp16, [1, ?, 1024]> linear_115_cast_fp16 = linear(bias = layers_19_self_attn_q_proj_bias_to_fp16, weight = layers_19_self_attn_q_proj_weight_to_fp16_palettized, x = x_119_cast_fp16)[name = tensor<string, []>("linear_115_cast_fp16")]; |
| tensor<int32, [4]> concat_78x = const()[name = tensor<string, []>("concat_78x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1510_cast_fp16 = reshape(shape = concat_78x, x = linear_115_cast_fp16)[name = tensor<string, []>("op_1510_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_19_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127390528))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127914880))), name = tensor<string, []>("layers_19_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_19_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_19_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127915008)))]; |
| tensor<fp16, [1, ?, 1024]> linear_116_cast_fp16 = linear(bias = layers_19_self_attn_k_proj_bias_to_fp16, weight = layers_19_self_attn_k_proj_weight_to_fp16_palettized, x = x_119_cast_fp16)[name = tensor<string, []>("linear_116_cast_fp16")]; |
| tensor<int32, [4]> concat_79x = const()[name = tensor<string, []>("concat_79x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1516_cast_fp16 = reshape(shape = concat_79x, x = linear_116_cast_fp16)[name = tensor<string, []>("op_1516_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_19_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127917120))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128441472))), name = tensor<string, []>("layers_19_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128441600)))]; |
| tensor<fp16, [1, ?, 1024]> linear_117_cast_fp16 = linear(bias = layers_19_self_attn_v_proj_bias_to_fp16, weight = layers_19_self_attn_v_proj_weight_to_fp16_palettized, x = x_119_cast_fp16)[name = tensor<string, []>("linear_117_cast_fp16")]; |
| tensor<int32, [4]> concat_80x = const()[name = tensor<string, []>("concat_80x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1522_cast_fp16 = reshape(shape = concat_80x, x = linear_117_cast_fp16)[name = tensor<string, []>("op_1522_cast_fp16")]; |
| tensor<int32, [4]> v_39_perm_0 = const()[name = tensor<string, []>("v_39_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_1525_transpose_x_0 = const()[name = tensor<string, []>("op_1525_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1525_transpose_y_0 = const()[name = tensor<string, []>("op_1525_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_134_perm_0 = const()[name = tensor<string, []>("transpose_134_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_135_perm_0 = const()[name = tensor<string, []>("transpose_135_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_135 = transpose(perm = transpose_135_perm_0, x = var_1516_cast_fp16)[name = tensor<string, []>("transpose_162")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_134 = transpose(perm = transpose_134_perm_0, x = var_1510_cast_fp16)[name = tensor<string, []>("transpose_163")]; |
| tensor<fp16, [1, 16, ?, ?]> var_1525_cast_fp16 = matmul(transpose_x = var_1525_transpose_x_0, transpose_y = var_1525_transpose_y_0, x = transpose_134, y = transpose_135)[name = tensor<string, []>("op_1525_cast_fp16")]; |
| tensor<fp16, []> var_1526_to_fp16 = const()[name = tensor<string, []>("op_1526_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_239_cast_fp16 = mul(x = var_1525_cast_fp16, y = var_1526_to_fp16)[name = tensor<string, []>("input_239_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_39_cast_fp16 = softmax(axis = var_1486, x = input_239_cast_fp16)[name = tensor<string, []>("attn_39_cast_fp16")]; |
| tensor<bool, []> out_39_transpose_x_0 = const()[name = tensor<string, []>("out_39_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_39_transpose_y_0 = const()[name = tensor<string, []>("out_39_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_39_cast_fp16 = transpose(perm = v_39_perm_0, x = var_1522_cast_fp16)[name = tensor<string, []>("transpose_161")]; |
| tensor<fp16, [1, 16, ?, 64]> out_39_cast_fp16 = matmul(transpose_x = out_39_transpose_x_0, transpose_y = out_39_transpose_y_0, x = attn_39_cast_fp16, y = v_39_cast_fp16)[name = tensor<string, []>("out_39_cast_fp16")]; |
| tensor<int32, [4]> var_1530_perm_0 = const()[name = tensor<string, []>("op_1530_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_81x = const()[name = tensor<string, []>("concat_81x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1530_cast_fp16 = transpose(perm = var_1530_perm_0, x = out_39_cast_fp16)[name = tensor<string, []>("transpose_160")]; |
| tensor<fp16, [1, ?, 1024]> input_241_cast_fp16 = reshape(shape = concat_81x, x = var_1530_cast_fp16)[name = tensor<string, []>("input_241_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_19_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128443712))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128968064))), name = tensor<string, []>("layers_19_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_19_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_19_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128968192)))]; |
| tensor<fp16, [1, ?, 1024]> linear_118_cast_fp16 = linear(bias = layers_19_self_attn_out_proj_bias_to_fp16, weight = layers_19_self_attn_out_proj_weight_to_fp16_palettized, x = input_241_cast_fp16)[name = tensor<string, []>("linear_118_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_243_cast_fp16 = add(x = input_237_cast_fp16, y = linear_118_cast_fp16)[name = tensor<string, []>("input_243_cast_fp16")]; |
| tensor<int32, [1]> input_245_axes_0 = const()[name = tensor<string, []>("input_245_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_19_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_19_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128970304)))]; |
| tensor<fp16, [1024]> layers_19_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_19_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128972416)))]; |
| tensor<fp16, [1, ?, 1024]> input_245_cast_fp16 = layer_norm(axes = input_245_axes_0, beta = layers_19_final_layer_norm_bias_to_fp16, epsilon = var_1489_to_fp16, gamma = layers_19_final_layer_norm_weight_to_fp16, x = input_243_cast_fp16)[name = tensor<string, []>("input_245_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_19_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128974528))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131071744))), name = tensor<string, []>("layers_19_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_19_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_19_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131071872)))]; |
| tensor<fp16, [1, ?, 4096]> linear_119_cast_fp16 = linear(bias = layers_19_fc1_bias_to_fp16, weight = layers_19_fc1_weight_to_fp16_palettized, x = input_245_cast_fp16)[name = tensor<string, []>("linear_119_cast_fp16")]; |
| tensor<string, []> input_247_mode_0 = const()[name = tensor<string, []>("input_247_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_247_cast_fp16 = gelu(mode = input_247_mode_0, x = linear_119_cast_fp16)[name = tensor<string, []>("input_247_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_19_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131080128))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133177344))), name = tensor<string, []>("layers_19_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_19_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_19_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133177472)))]; |
| tensor<fp16, [1, ?, 1024]> linear_120_cast_fp16 = linear(bias = layers_19_fc2_bias_to_fp16, weight = layers_19_fc2_weight_to_fp16_palettized, x = input_247_cast_fp16)[name = tensor<string, []>("linear_120_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_249_cast_fp16 = add(x = input_243_cast_fp16, y = linear_120_cast_fp16)[name = tensor<string, []>("input_249_cast_fp16")]; |
| tensor<int32, []> var_1556 = const()[name = tensor<string, []>("op_1556"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_125_axes_0 = const()[name = tensor<string, []>("x_125_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_20_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_20_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133179584)))]; |
| tensor<fp16, [1024]> layers_20_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_20_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133181696)))]; |
| tensor<fp16, []> var_1559_to_fp16 = const()[name = tensor<string, []>("op_1559_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_125_cast_fp16 = layer_norm(axes = x_125_axes_0, beta = layers_20_self_attn_layer_norm_bias_to_fp16, epsilon = var_1559_to_fp16, gamma = layers_20_self_attn_layer_norm_weight_to_fp16, x = input_249_cast_fp16)[name = tensor<string, []>("x_125_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_20_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133183808))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133708160))), name = tensor<string, []>("layers_20_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133708288)))]; |
| tensor<fp16, [1, ?, 1024]> linear_121_cast_fp16 = linear(bias = layers_20_self_attn_q_proj_bias_to_fp16, weight = layers_20_self_attn_q_proj_weight_to_fp16_palettized, x = x_125_cast_fp16)[name = tensor<string, []>("linear_121_cast_fp16")]; |
| tensor<int32, [4]> concat_82x = const()[name = tensor<string, []>("concat_82x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1580_cast_fp16 = reshape(shape = concat_82x, x = linear_121_cast_fp16)[name = tensor<string, []>("op_1580_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_20_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133710400))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134234752))), name = tensor<string, []>("layers_20_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_20_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_20_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134234880)))]; |
| tensor<fp16, [1, ?, 1024]> linear_122_cast_fp16 = linear(bias = layers_20_self_attn_k_proj_bias_to_fp16, weight = layers_20_self_attn_k_proj_weight_to_fp16_palettized, x = x_125_cast_fp16)[name = tensor<string, []>("linear_122_cast_fp16")]; |
| tensor<int32, [4]> concat_83x = const()[name = tensor<string, []>("concat_83x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1586_cast_fp16 = reshape(shape = concat_83x, x = linear_122_cast_fp16)[name = tensor<string, []>("op_1586_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_20_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134236992))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134761344))), name = tensor<string, []>("layers_20_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134761472)))]; |
| tensor<fp16, [1, ?, 1024]> linear_123_cast_fp16 = linear(bias = layers_20_self_attn_v_proj_bias_to_fp16, weight = layers_20_self_attn_v_proj_weight_to_fp16_palettized, x = x_125_cast_fp16)[name = tensor<string, []>("linear_123_cast_fp16")]; |
| tensor<int32, [4]> concat_84x = const()[name = tensor<string, []>("concat_84x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1592_cast_fp16 = reshape(shape = concat_84x, x = linear_123_cast_fp16)[name = tensor<string, []>("op_1592_cast_fp16")]; |
| tensor<int32, [4]> v_41_perm_0 = const()[name = tensor<string, []>("v_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_1595_transpose_x_0 = const()[name = tensor<string, []>("op_1595_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1595_transpose_y_0 = const()[name = tensor<string, []>("op_1595_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_136_perm_0 = const()[name = tensor<string, []>("transpose_136_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_137_perm_0 = const()[name = tensor<string, []>("transpose_137_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_137 = transpose(perm = transpose_137_perm_0, x = var_1586_cast_fp16)[name = tensor<string, []>("transpose_158")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_136 = transpose(perm = transpose_136_perm_0, x = var_1580_cast_fp16)[name = tensor<string, []>("transpose_159")]; |
| tensor<fp16, [1, 16, ?, ?]> var_1595_cast_fp16 = matmul(transpose_x = var_1595_transpose_x_0, transpose_y = var_1595_transpose_y_0, x = transpose_136, y = transpose_137)[name = tensor<string, []>("op_1595_cast_fp16")]; |
| tensor<fp16, []> var_1596_to_fp16 = const()[name = tensor<string, []>("op_1596_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_251_cast_fp16 = mul(x = var_1595_cast_fp16, y = var_1596_to_fp16)[name = tensor<string, []>("input_251_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_41_cast_fp16 = softmax(axis = var_1556, x = input_251_cast_fp16)[name = tensor<string, []>("attn_41_cast_fp16")]; |
| tensor<bool, []> out_41_transpose_x_0 = const()[name = tensor<string, []>("out_41_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_41_transpose_y_0 = const()[name = tensor<string, []>("out_41_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_41_cast_fp16 = transpose(perm = v_41_perm_0, x = var_1592_cast_fp16)[name = tensor<string, []>("transpose_157")]; |
| tensor<fp16, [1, 16, ?, 64]> out_41_cast_fp16 = matmul(transpose_x = out_41_transpose_x_0, transpose_y = out_41_transpose_y_0, x = attn_41_cast_fp16, y = v_41_cast_fp16)[name = tensor<string, []>("out_41_cast_fp16")]; |
| tensor<int32, [4]> var_1600_perm_0 = const()[name = tensor<string, []>("op_1600_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_85x = const()[name = tensor<string, []>("concat_85x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1600_cast_fp16 = transpose(perm = var_1600_perm_0, x = out_41_cast_fp16)[name = tensor<string, []>("transpose_156")]; |
| tensor<fp16, [1, ?, 1024]> input_253_cast_fp16 = reshape(shape = concat_85x, x = var_1600_cast_fp16)[name = tensor<string, []>("input_253_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_20_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134763584))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135287936))), name = tensor<string, []>("layers_20_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_20_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_20_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135288064)))]; |
| tensor<fp16, [1, ?, 1024]> linear_124_cast_fp16 = linear(bias = layers_20_self_attn_out_proj_bias_to_fp16, weight = layers_20_self_attn_out_proj_weight_to_fp16_palettized, x = input_253_cast_fp16)[name = tensor<string, []>("linear_124_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_255_cast_fp16 = add(x = input_249_cast_fp16, y = linear_124_cast_fp16)[name = tensor<string, []>("input_255_cast_fp16")]; |
| tensor<int32, [1]> input_257_axes_0 = const()[name = tensor<string, []>("input_257_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_20_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_20_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135290176)))]; |
| tensor<fp16, [1024]> layers_20_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_20_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135292288)))]; |
| tensor<fp16, [1, ?, 1024]> input_257_cast_fp16 = layer_norm(axes = input_257_axes_0, beta = layers_20_final_layer_norm_bias_to_fp16, epsilon = var_1559_to_fp16, gamma = layers_20_final_layer_norm_weight_to_fp16, x = input_255_cast_fp16)[name = tensor<string, []>("input_257_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_20_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135294400))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137391616))), name = tensor<string, []>("layers_20_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_20_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_20_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137391744)))]; |
| tensor<fp16, [1, ?, 4096]> linear_125_cast_fp16 = linear(bias = layers_20_fc1_bias_to_fp16, weight = layers_20_fc1_weight_to_fp16_palettized, x = input_257_cast_fp16)[name = tensor<string, []>("linear_125_cast_fp16")]; |
| tensor<string, []> input_259_mode_0 = const()[name = tensor<string, []>("input_259_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_259_cast_fp16 = gelu(mode = input_259_mode_0, x = linear_125_cast_fp16)[name = tensor<string, []>("input_259_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_20_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137400000))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139497216))), name = tensor<string, []>("layers_20_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_20_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_20_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139497344)))]; |
| tensor<fp16, [1, ?, 1024]> linear_126_cast_fp16 = linear(bias = layers_20_fc2_bias_to_fp16, weight = layers_20_fc2_weight_to_fp16_palettized, x = input_259_cast_fp16)[name = tensor<string, []>("linear_126_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_261_cast_fp16 = add(x = input_255_cast_fp16, y = linear_126_cast_fp16)[name = tensor<string, []>("input_261_cast_fp16")]; |
| tensor<int32, []> var_1626 = const()[name = tensor<string, []>("op_1626"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_131_axes_0 = const()[name = tensor<string, []>("x_131_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_21_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_21_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139499456)))]; |
| tensor<fp16, [1024]> layers_21_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_21_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139501568)))]; |
| tensor<fp16, []> var_1629_to_fp16 = const()[name = tensor<string, []>("op_1629_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_131_cast_fp16 = layer_norm(axes = x_131_axes_0, beta = layers_21_self_attn_layer_norm_bias_to_fp16, epsilon = var_1629_to_fp16, gamma = layers_21_self_attn_layer_norm_weight_to_fp16, x = input_261_cast_fp16)[name = tensor<string, []>("x_131_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_21_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139503680))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140028032))), name = tensor<string, []>("layers_21_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140028160)))]; |
| tensor<fp16, [1, ?, 1024]> linear_127_cast_fp16 = linear(bias = layers_21_self_attn_q_proj_bias_to_fp16, weight = layers_21_self_attn_q_proj_weight_to_fp16_palettized, x = x_131_cast_fp16)[name = tensor<string, []>("linear_127_cast_fp16")]; |
| tensor<int32, [4]> concat_86x = const()[name = tensor<string, []>("concat_86x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1650_cast_fp16 = reshape(shape = concat_86x, x = linear_127_cast_fp16)[name = tensor<string, []>("op_1650_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_21_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140030272))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140554624))), name = tensor<string, []>("layers_21_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_21_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_21_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140554752)))]; |
| tensor<fp16, [1, ?, 1024]> linear_128_cast_fp16 = linear(bias = layers_21_self_attn_k_proj_bias_to_fp16, weight = layers_21_self_attn_k_proj_weight_to_fp16_palettized, x = x_131_cast_fp16)[name = tensor<string, []>("linear_128_cast_fp16")]; |
| tensor<int32, [4]> concat_87x = const()[name = tensor<string, []>("concat_87x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1656_cast_fp16 = reshape(shape = concat_87x, x = linear_128_cast_fp16)[name = tensor<string, []>("op_1656_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_21_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140556864))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141081216))), name = tensor<string, []>("layers_21_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141081344)))]; |
| tensor<fp16, [1, ?, 1024]> linear_129_cast_fp16 = linear(bias = layers_21_self_attn_v_proj_bias_to_fp16, weight = layers_21_self_attn_v_proj_weight_to_fp16_palettized, x = x_131_cast_fp16)[name = tensor<string, []>("linear_129_cast_fp16")]; |
| tensor<int32, [4]> concat_88x = const()[name = tensor<string, []>("concat_88x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1662_cast_fp16 = reshape(shape = concat_88x, x = linear_129_cast_fp16)[name = tensor<string, []>("op_1662_cast_fp16")]; |
| tensor<int32, [4]> v_43_perm_0 = const()[name = tensor<string, []>("v_43_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_1665_transpose_x_0 = const()[name = tensor<string, []>("op_1665_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1665_transpose_y_0 = const()[name = tensor<string, []>("op_1665_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_138_perm_0 = const()[name = tensor<string, []>("transpose_138_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_139_perm_0 = const()[name = tensor<string, []>("transpose_139_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_139 = transpose(perm = transpose_139_perm_0, x = var_1656_cast_fp16)[name = tensor<string, []>("transpose_154")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_138 = transpose(perm = transpose_138_perm_0, x = var_1650_cast_fp16)[name = tensor<string, []>("transpose_155")]; |
| tensor<fp16, [1, 16, ?, ?]> var_1665_cast_fp16 = matmul(transpose_x = var_1665_transpose_x_0, transpose_y = var_1665_transpose_y_0, x = transpose_138, y = transpose_139)[name = tensor<string, []>("op_1665_cast_fp16")]; |
| tensor<fp16, []> var_1666_to_fp16 = const()[name = tensor<string, []>("op_1666_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_263_cast_fp16 = mul(x = var_1665_cast_fp16, y = var_1666_to_fp16)[name = tensor<string, []>("input_263_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_43_cast_fp16 = softmax(axis = var_1626, x = input_263_cast_fp16)[name = tensor<string, []>("attn_43_cast_fp16")]; |
| tensor<bool, []> out_43_transpose_x_0 = const()[name = tensor<string, []>("out_43_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_43_transpose_y_0 = const()[name = tensor<string, []>("out_43_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_43_cast_fp16 = transpose(perm = v_43_perm_0, x = var_1662_cast_fp16)[name = tensor<string, []>("transpose_153")]; |
| tensor<fp16, [1, 16, ?, 64]> out_43_cast_fp16 = matmul(transpose_x = out_43_transpose_x_0, transpose_y = out_43_transpose_y_0, x = attn_43_cast_fp16, y = v_43_cast_fp16)[name = tensor<string, []>("out_43_cast_fp16")]; |
| tensor<int32, [4]> var_1670_perm_0 = const()[name = tensor<string, []>("op_1670_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_89x = const()[name = tensor<string, []>("concat_89x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1670_cast_fp16 = transpose(perm = var_1670_perm_0, x = out_43_cast_fp16)[name = tensor<string, []>("transpose_152")]; |
| tensor<fp16, [1, ?, 1024]> input_265_cast_fp16 = reshape(shape = concat_89x, x = var_1670_cast_fp16)[name = tensor<string, []>("input_265_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_21_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141083456))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141607808))), name = tensor<string, []>("layers_21_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_21_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_21_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141607936)))]; |
| tensor<fp16, [1, ?, 1024]> linear_130_cast_fp16 = linear(bias = layers_21_self_attn_out_proj_bias_to_fp16, weight = layers_21_self_attn_out_proj_weight_to_fp16_palettized, x = input_265_cast_fp16)[name = tensor<string, []>("linear_130_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_267_cast_fp16 = add(x = input_261_cast_fp16, y = linear_130_cast_fp16)[name = tensor<string, []>("input_267_cast_fp16")]; |
| tensor<int32, [1]> input_269_axes_0 = const()[name = tensor<string, []>("input_269_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_21_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_21_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141610048)))]; |
| tensor<fp16, [1024]> layers_21_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_21_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141612160)))]; |
| tensor<fp16, [1, ?, 1024]> input_269_cast_fp16 = layer_norm(axes = input_269_axes_0, beta = layers_21_final_layer_norm_bias_to_fp16, epsilon = var_1629_to_fp16, gamma = layers_21_final_layer_norm_weight_to_fp16, x = input_267_cast_fp16)[name = tensor<string, []>("input_269_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_21_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141614272))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143711488))), name = tensor<string, []>("layers_21_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_21_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_21_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143711616)))]; |
| tensor<fp16, [1, ?, 4096]> linear_131_cast_fp16 = linear(bias = layers_21_fc1_bias_to_fp16, weight = layers_21_fc1_weight_to_fp16_palettized, x = input_269_cast_fp16)[name = tensor<string, []>("linear_131_cast_fp16")]; |
| tensor<string, []> input_271_mode_0 = const()[name = tensor<string, []>("input_271_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_271_cast_fp16 = gelu(mode = input_271_mode_0, x = linear_131_cast_fp16)[name = tensor<string, []>("input_271_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_21_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143719872))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145817088))), name = tensor<string, []>("layers_21_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_21_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_21_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145817216)))]; |
| tensor<fp16, [1, ?, 1024]> linear_132_cast_fp16 = linear(bias = layers_21_fc2_bias_to_fp16, weight = layers_21_fc2_weight_to_fp16_palettized, x = input_271_cast_fp16)[name = tensor<string, []>("linear_132_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_273_cast_fp16 = add(x = input_267_cast_fp16, y = linear_132_cast_fp16)[name = tensor<string, []>("input_273_cast_fp16")]; |
| tensor<int32, []> var_1696 = const()[name = tensor<string, []>("op_1696"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_137_axes_0 = const()[name = tensor<string, []>("x_137_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_22_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_22_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145819328)))]; |
| tensor<fp16, [1024]> layers_22_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_22_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145821440)))]; |
| tensor<fp16, []> var_1699_to_fp16 = const()[name = tensor<string, []>("op_1699_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_137_cast_fp16 = layer_norm(axes = x_137_axes_0, beta = layers_22_self_attn_layer_norm_bias_to_fp16, epsilon = var_1699_to_fp16, gamma = layers_22_self_attn_layer_norm_weight_to_fp16, x = input_273_cast_fp16)[name = tensor<string, []>("x_137_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_22_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145823552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146347904))), name = tensor<string, []>("layers_22_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146348032)))]; |
| tensor<fp16, [1, ?, 1024]> linear_133_cast_fp16 = linear(bias = layers_22_self_attn_q_proj_bias_to_fp16, weight = layers_22_self_attn_q_proj_weight_to_fp16_palettized, x = x_137_cast_fp16)[name = tensor<string, []>("linear_133_cast_fp16")]; |
| tensor<int32, [4]> concat_90x = const()[name = tensor<string, []>("concat_90x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1720_cast_fp16 = reshape(shape = concat_90x, x = linear_133_cast_fp16)[name = tensor<string, []>("op_1720_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_22_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146350144))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146874496))), name = tensor<string, []>("layers_22_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_22_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_22_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146874624)))]; |
| tensor<fp16, [1, ?, 1024]> linear_134_cast_fp16 = linear(bias = layers_22_self_attn_k_proj_bias_to_fp16, weight = layers_22_self_attn_k_proj_weight_to_fp16_palettized, x = x_137_cast_fp16)[name = tensor<string, []>("linear_134_cast_fp16")]; |
| tensor<int32, [4]> concat_91x = const()[name = tensor<string, []>("concat_91x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1726_cast_fp16 = reshape(shape = concat_91x, x = linear_134_cast_fp16)[name = tensor<string, []>("op_1726_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_22_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146876736))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147401088))), name = tensor<string, []>("layers_22_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147401216)))]; |
| tensor<fp16, [1, ?, 1024]> linear_135_cast_fp16 = linear(bias = layers_22_self_attn_v_proj_bias_to_fp16, weight = layers_22_self_attn_v_proj_weight_to_fp16_palettized, x = x_137_cast_fp16)[name = tensor<string, []>("linear_135_cast_fp16")]; |
| tensor<int32, [4]> concat_92x = const()[name = tensor<string, []>("concat_92x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1732_cast_fp16 = reshape(shape = concat_92x, x = linear_135_cast_fp16)[name = tensor<string, []>("op_1732_cast_fp16")]; |
| tensor<int32, [4]> v_45_perm_0 = const()[name = tensor<string, []>("v_45_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_1735_transpose_x_0 = const()[name = tensor<string, []>("op_1735_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1735_transpose_y_0 = const()[name = tensor<string, []>("op_1735_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_140_perm_0 = const()[name = tensor<string, []>("transpose_140_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_141_perm_0 = const()[name = tensor<string, []>("transpose_141_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_141 = transpose(perm = transpose_141_perm_0, x = var_1726_cast_fp16)[name = tensor<string, []>("transpose_150")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_140 = transpose(perm = transpose_140_perm_0, x = var_1720_cast_fp16)[name = tensor<string, []>("transpose_151")]; |
| tensor<fp16, [1, 16, ?, ?]> var_1735_cast_fp16 = matmul(transpose_x = var_1735_transpose_x_0, transpose_y = var_1735_transpose_y_0, x = transpose_140, y = transpose_141)[name = tensor<string, []>("op_1735_cast_fp16")]; |
| tensor<fp16, []> var_1736_to_fp16 = const()[name = tensor<string, []>("op_1736_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_275_cast_fp16 = mul(x = var_1735_cast_fp16, y = var_1736_to_fp16)[name = tensor<string, []>("input_275_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_45_cast_fp16 = softmax(axis = var_1696, x = input_275_cast_fp16)[name = tensor<string, []>("attn_45_cast_fp16")]; |
| tensor<bool, []> out_45_transpose_x_0 = const()[name = tensor<string, []>("out_45_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_45_transpose_y_0 = const()[name = tensor<string, []>("out_45_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_45_cast_fp16 = transpose(perm = v_45_perm_0, x = var_1732_cast_fp16)[name = tensor<string, []>("transpose_149")]; |
| tensor<fp16, [1, 16, ?, 64]> out_45_cast_fp16 = matmul(transpose_x = out_45_transpose_x_0, transpose_y = out_45_transpose_y_0, x = attn_45_cast_fp16, y = v_45_cast_fp16)[name = tensor<string, []>("out_45_cast_fp16")]; |
| tensor<int32, [4]> var_1740_perm_0 = const()[name = tensor<string, []>("op_1740_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_93x = const()[name = tensor<string, []>("concat_93x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1740_cast_fp16 = transpose(perm = var_1740_perm_0, x = out_45_cast_fp16)[name = tensor<string, []>("transpose_148")]; |
| tensor<fp16, [1, ?, 1024]> input_277_cast_fp16 = reshape(shape = concat_93x, x = var_1740_cast_fp16)[name = tensor<string, []>("input_277_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_22_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147403328))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147927680))), name = tensor<string, []>("layers_22_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_22_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_22_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147927808)))]; |
| tensor<fp16, [1, ?, 1024]> linear_136_cast_fp16 = linear(bias = layers_22_self_attn_out_proj_bias_to_fp16, weight = layers_22_self_attn_out_proj_weight_to_fp16_palettized, x = input_277_cast_fp16)[name = tensor<string, []>("linear_136_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_279_cast_fp16 = add(x = input_273_cast_fp16, y = linear_136_cast_fp16)[name = tensor<string, []>("input_279_cast_fp16")]; |
| tensor<int32, [1]> input_281_axes_0 = const()[name = tensor<string, []>("input_281_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_22_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_22_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147929920)))]; |
| tensor<fp16, [1024]> layers_22_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_22_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147932032)))]; |
| tensor<fp16, [1, ?, 1024]> input_281_cast_fp16 = layer_norm(axes = input_281_axes_0, beta = layers_22_final_layer_norm_bias_to_fp16, epsilon = var_1699_to_fp16, gamma = layers_22_final_layer_norm_weight_to_fp16, x = input_279_cast_fp16)[name = tensor<string, []>("input_281_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_22_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147934144))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150031360))), name = tensor<string, []>("layers_22_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_22_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_22_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150031488)))]; |
| tensor<fp16, [1, ?, 4096]> linear_137_cast_fp16 = linear(bias = layers_22_fc1_bias_to_fp16, weight = layers_22_fc1_weight_to_fp16_palettized, x = input_281_cast_fp16)[name = tensor<string, []>("linear_137_cast_fp16")]; |
| tensor<string, []> input_283_mode_0 = const()[name = tensor<string, []>("input_283_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_283_cast_fp16 = gelu(mode = input_283_mode_0, x = linear_137_cast_fp16)[name = tensor<string, []>("input_283_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_22_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150039744))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152136960))), name = tensor<string, []>("layers_22_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_22_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_22_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152137088)))]; |
| tensor<fp16, [1, ?, 1024]> linear_138_cast_fp16 = linear(bias = layers_22_fc2_bias_to_fp16, weight = layers_22_fc2_weight_to_fp16_palettized, x = input_283_cast_fp16)[name = tensor<string, []>("linear_138_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_285_cast_fp16 = add(x = input_279_cast_fp16, y = linear_138_cast_fp16)[name = tensor<string, []>("input_285_cast_fp16")]; |
| tensor<int32, []> var_1766 = const()[name = tensor<string, []>("op_1766"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> x_143_axes_0 = const()[name = tensor<string, []>("x_143_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_23_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_23_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152139200)))]; |
| tensor<fp16, [1024]> layers_23_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_23_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152141312)))]; |
| tensor<fp16, []> var_1769_to_fp16 = const()[name = tensor<string, []>("op_1769_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> x_143_cast_fp16 = layer_norm(axes = x_143_axes_0, beta = layers_23_self_attn_layer_norm_bias_to_fp16, epsilon = var_1769_to_fp16, gamma = layers_23_self_attn_layer_norm_weight_to_fp16, x = input_285_cast_fp16)[name = tensor<string, []>("x_143_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_23_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152143424))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152667776))), name = tensor<string, []>("layers_23_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_23_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_23_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152667904)))]; |
| tensor<fp16, [1, ?, 1024]> linear_139_cast_fp16 = linear(bias = layers_23_self_attn_q_proj_bias_to_fp16, weight = layers_23_self_attn_q_proj_weight_to_fp16_palettized, x = x_143_cast_fp16)[name = tensor<string, []>("linear_139_cast_fp16")]; |
| tensor<int32, [4]> concat_94x = const()[name = tensor<string, []>("concat_94x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1790_cast_fp16 = reshape(shape = concat_94x, x = linear_139_cast_fp16)[name = tensor<string, []>("op_1790_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_23_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152670016))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153194368))), name = tensor<string, []>("layers_23_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_23_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_23_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153194496)))]; |
| tensor<fp16, [1, ?, 1024]> linear_140_cast_fp16 = linear(bias = layers_23_self_attn_k_proj_bias_to_fp16, weight = layers_23_self_attn_k_proj_weight_to_fp16_palettized, x = x_143_cast_fp16)[name = tensor<string, []>("linear_140_cast_fp16")]; |
| tensor<int32, [4]> concat_95x = const()[name = tensor<string, []>("concat_95x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1796_cast_fp16 = reshape(shape = concat_95x, x = linear_140_cast_fp16)[name = tensor<string, []>("op_1796_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_23_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153196608))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153720960))), name = tensor<string, []>("layers_23_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_23_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_23_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153721088)))]; |
| tensor<fp16, [1, ?, 1024]> linear_141_cast_fp16 = linear(bias = layers_23_self_attn_v_proj_bias_to_fp16, weight = layers_23_self_attn_v_proj_weight_to_fp16_palettized, x = x_143_cast_fp16)[name = tensor<string, []>("linear_141_cast_fp16")]; |
| tensor<int32, [4]> concat_96x = const()[name = tensor<string, []>("concat_96x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1802_cast_fp16 = reshape(shape = concat_96x, x = linear_141_cast_fp16)[name = tensor<string, []>("op_1802_cast_fp16")]; |
| tensor<int32, [4]> v_perm_0 = const()[name = tensor<string, []>("v_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<bool, []> var_1805_transpose_x_0 = const()[name = tensor<string, []>("op_1805_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1805_transpose_y_0 = const()[name = tensor<string, []>("op_1805_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_142_perm_0 = const()[name = tensor<string, []>("transpose_142_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_143_perm_0 = const()[name = tensor<string, []>("transpose_143_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; |
| tensor<fp16, [1, 16, 64, ?]> transpose_143 = transpose(perm = transpose_143_perm_0, x = var_1796_cast_fp16)[name = tensor<string, []>("transpose_146")]; |
| tensor<fp16, [1, 16, ?, 64]> transpose_142 = transpose(perm = transpose_142_perm_0, x = var_1790_cast_fp16)[name = tensor<string, []>("transpose_147")]; |
| tensor<fp16, [1, 16, ?, ?]> var_1805_cast_fp16 = matmul(transpose_x = var_1805_transpose_x_0, transpose_y = var_1805_transpose_y_0, x = transpose_142, y = transpose_143)[name = tensor<string, []>("op_1805_cast_fp16")]; |
| tensor<fp16, []> var_1806_to_fp16 = const()[name = tensor<string, []>("op_1806_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 16, ?, ?]> input_287_cast_fp16 = mul(x = var_1805_cast_fp16, y = var_1806_to_fp16)[name = tensor<string, []>("input_287_cast_fp16")]; |
| tensor<fp16, [1, 16, ?, ?]> attn_cast_fp16 = softmax(axis = var_1766, x = input_287_cast_fp16)[name = tensor<string, []>("attn_cast_fp16")]; |
| tensor<bool, []> out_transpose_x_0 = const()[name = tensor<string, []>("out_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> out_transpose_y_0 = const()[name = tensor<string, []>("out_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 16, ?, 64]> v_cast_fp16 = transpose(perm = v_perm_0, x = var_1802_cast_fp16)[name = tensor<string, []>("transpose_145")]; |
| tensor<fp16, [1, 16, ?, 64]> out_cast_fp16 = matmul(transpose_x = out_transpose_x_0, transpose_y = out_transpose_y_0, x = attn_cast_fp16, y = v_cast_fp16)[name = tensor<string, []>("out_cast_fp16")]; |
| tensor<int32, [4]> var_1810_perm_0 = const()[name = tensor<string, []>("op_1810_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_97x = const()[name = tensor<string, []>("concat_97x"), val = tensor<int32, [3]>([1, -1, 1024])]; |
| tensor<fp16, [1, ?, 16, 64]> var_1810_cast_fp16 = transpose(perm = var_1810_perm_0, x = out_cast_fp16)[name = tensor<string, []>("transpose_144")]; |
| tensor<fp16, [1, ?, 1024]> input_289_cast_fp16 = reshape(shape = concat_97x, x = var_1810_cast_fp16)[name = tensor<string, []>("input_289_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> layers_23_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153723200))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154247552))), name = tensor<string, []>("layers_23_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> layers_23_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_23_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154247680)))]; |
| tensor<fp16, [1, ?, 1024]> linear_142_cast_fp16 = linear(bias = layers_23_self_attn_out_proj_bias_to_fp16, weight = layers_23_self_attn_out_proj_weight_to_fp16_palettized, x = input_289_cast_fp16)[name = tensor<string, []>("linear_142_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_291_cast_fp16 = add(x = input_285_cast_fp16, y = linear_142_cast_fp16)[name = tensor<string, []>("input_291_cast_fp16")]; |
| tensor<int32, [1]> input_293_axes_0 = const()[name = tensor<string, []>("input_293_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> layers_23_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_23_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154249792)))]; |
| tensor<fp16, [1024]> layers_23_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_23_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154251904)))]; |
| tensor<fp16, [1, ?, 1024]> input_293_cast_fp16 = layer_norm(axes = input_293_axes_0, beta = layers_23_final_layer_norm_bias_to_fp16, epsilon = var_1769_to_fp16, gamma = layers_23_final_layer_norm_weight_to_fp16, x = input_291_cast_fp16)[name = tensor<string, []>("input_293_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> layers_23_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154254016))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156351232))), name = tensor<string, []>("layers_23_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> layers_23_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_23_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156351360)))]; |
| tensor<fp16, [1, ?, 4096]> linear_143_cast_fp16 = linear(bias = layers_23_fc1_bias_to_fp16, weight = layers_23_fc1_weight_to_fp16_palettized, x = input_293_cast_fp16)[name = tensor<string, []>("linear_143_cast_fp16")]; |
| tensor<string, []> input_295_mode_0 = const()[name = tensor<string, []>("input_295_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 4096]> input_295_cast_fp16 = gelu(mode = input_295_mode_0, x = linear_143_cast_fp16)[name = tensor<string, []>("input_295_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> layers_23_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156359616))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158456832))), name = tensor<string, []>("layers_23_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> layers_23_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_23_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158456960)))]; |
| tensor<fp16, [1, ?, 1024]> linear_144_cast_fp16 = linear(bias = layers_23_fc2_bias_to_fp16, weight = layers_23_fc2_weight_to_fp16_palettized, x = input_295_cast_fp16)[name = tensor<string, []>("linear_144_cast_fp16")]; |
| tensor<fp16, [1, ?, 1024]> input_297_cast_fp16 = add(x = input_291_cast_fp16, y = linear_144_cast_fp16)[name = tensor<string, []>("input_297_cast_fp16")]; |
| tensor<int32, [1]> input_299_axes_0 = const()[name = tensor<string, []>("input_299_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> ln_post_weight_to_fp16 = const()[name = tensor<string, []>("ln_post_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158459072)))]; |
| tensor<fp16, [1024]> ln_post_bias_to_fp16 = const()[name = tensor<string, []>("ln_post_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158461184)))]; |
| tensor<fp16, []> var_1830_to_fp16 = const()[name = tensor<string, []>("op_1830_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, ?, 1024]> input_299_cast_fp16 = layer_norm(axes = input_299_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_1830_to_fp16, gamma = ln_post_weight_to_fp16, x = input_297_cast_fp16)[name = tensor<string, []>("input_299_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> proj1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158463296))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158987648))), name = tensor<string, []>("proj1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> proj1_bias_to_fp16 = const()[name = tensor<string, []>("proj1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158987776)))]; |
| tensor<fp16, [1, ?, 1024]> linear_145_cast_fp16 = linear(bias = proj1_bias_to_fp16, weight = proj1_weight_to_fp16_palettized, x = input_299_cast_fp16)[name = tensor<string, []>("linear_145_cast_fp16")]; |
| tensor<string, []> input_mode_0 = const()[name = tensor<string, []>("input_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, ?, 1024]> input_cast_fp16 = gelu(mode = input_mode_0, x = linear_145_cast_fp16)[name = tensor<string, []>("input_cast_fp16")]; |
| tensor<fp16, [2048, 1024]> proj2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1048576]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158989888))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160038528))), name = tensor<string, []>("proj2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 1024])]; |
| tensor<fp16, [2048]> proj2_bias_to_fp16 = const()[name = tensor<string, []>("proj2_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160038656)))]; |
| tensor<fp16, [1, ?, 2048]> audio_embeddings = linear(bias = proj2_bias_to_fp16, weight = proj2_weight_to_fp16_palettized, x = input_cast_fp16)[name = tensor<string, []>("linear_146_cast_fp16")]; |
| } -> (audio_embeddings); |
| } |
