program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}})]
{
    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_59_pad_type_0 = const()[name = tensor<string, []>("op_59_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_59_pad_0 = const()[name = tensor<string, []>("op_59_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> var_59_strides_0 = const()[name = tensor<string, []>("op_59_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_59_dilations_0 = const()[name = tensor<string, []>("op_59_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_59_groups_0 = const()[name = tensor<string, []>("op_59_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [480, 1, 3, 3]> conv2d1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4480))), 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, []>(5056)))];
            tensor<fp16, [1, 480, 64, ?]> var_59_cast_fp16 = conv(bias = conv2d1_bias_to_fp16, dilations = var_59_dilations_0, groups = var_59_groups_0, pad = var_59_pad_0, pad_type = var_59_pad_type_0, strides = var_59_strides_0, weight = conv2d1_weight_to_fp16_palettized, x = input_1_cast_fp16)[name = tensor<string, []>("op_59_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_59_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")];
            tensor<string, []> var_73_pad_type_0 = const()[name = tensor<string, []>("op_73_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_73_pad_0 = const()[name = tensor<string, []>("op_73_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> var_73_strides_0 = const()[name = tensor<string, []>("op_73_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_73_dilations_0 = const()[name = tensor<string, []>("op_73_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_73_groups_0 = const()[name = tensor<string, []>("op_73_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [480, 480, 3, 3]> conv2d2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2073600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6080))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2079744))), 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, []>(2080320)))];
            tensor<fp16, [1, 480, 32, ?]> var_73_cast_fp16 = conv(bias = conv2d2_bias_to_fp16, dilations = var_73_dilations_0, groups = var_73_groups_0, pad = var_73_pad_0, pad_type = var_73_pad_type_0, strides = var_73_strides_0, weight = conv2d2_weight_to_fp16_palettized, x = input_3_cast_fp16)[name = tensor<string, []>("op_73_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_73_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
            tensor<string, []> var_87_pad_type_0 = const()[name = tensor<string, []>("op_87_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_87_pad_0 = const()[name = tensor<string, []>("op_87_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> var_87_strides_0 = const()[name = tensor<string, []>("op_87_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_87_dilations_0 = const()[name = tensor<string, []>("op_87_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_87_groups_0 = const()[name = tensor<string, []>("op_87_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [480, 480, 3, 3]> conv2d3_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2073600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2081344))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4155008))), 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, []>(4155584)))];
            tensor<fp16, [1, 480, 16, ?]> var_87_cast_fp16 = conv(bias = conv2d3_bias_to_fp16, dilations = var_87_dilations_0, groups = var_87_groups_0, pad = var_87_pad_0, pad_type = var_87_pad_type_0, strides = var_87_strides_0, weight = conv2d3_weight_to_fp16_palettized, x = input_5_cast_fp16)[name = tensor<string, []>("op_87_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_87_cast_fp16)[name = tensor<string, []>("x_1_cast_fp16")];
            tensor<int32, [4]> var_108 = const()[name = tensor<string, []>("op_108"), 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_109_cast_fp16 = transpose(perm = var_108, x = x_1_cast_fp16)[name = tensor<string, []>("transpose_180")];
            tensor<fp16, [1, ?, 7680]> input_7_cast_fp16 = reshape(shape = concat_0x, x = var_109_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")];
            tensor<fp16, [896, 7680]> conv_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [6881280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4156608))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11037952))), name = tensor<string, []>("conv_out_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 7680])];
            tensor<fp16, [896]> linear_0_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_0_bias_0_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11038528)))];
            tensor<fp16, [1, ?, 896]> 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_118_shape_cast_fp16 = shape(x = linear_0_cast_fp16)[name = tensor<string, []>("op_118_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_118_shape_cast_fp16_to_uint16_dtype_0 = const()[name = tensor<string, []>("op_118_shape_cast_fp16_to_uint16_dtype_0"), val = tensor<string, []>("uint16")];
            tensor<uint16, []> gather_4_indices_0_to_uint16 = const()[name = tensor<string, []>("gather_4_indices_0_to_uint16"), val = tensor<uint16, []>(1)];
            tensor<uint16, [3]> var_118_shape_cast_fp16_to_uint16 = cast(dtype = var_118_shape_cast_fp16_to_uint16_dtype_0, x = var_118_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 = gather_4_indices_0_to_uint16, validate_indices = gather_4_validate_indices_0, x = var_118_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, []>(896)];
            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_129_begin_0 = const()[name = tensor<string, []>("op_129_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<bool, [3]> var_129_end_mask_0 = const()[name = tensor<string, []>("op_129_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1500, 896]> pos_embed_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1344000]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11040384))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12384448))), name = tensor<string, []>("pos_embed_to_fp16_palettized"), shape = tensor<uint32, [3]>([1, 1500, 896])];
            tensor<fp16, [1, ?, 896]> var_129_cast_fp16 = slice_by_index(begin = var_129_begin_0, end = concat_1, end_mask = var_129_end_mask_0, x = pos_embed_to_fp16_palettized)[name = tensor<string, []>("op_129_cast_fp16")];
            tensor<fp16, [1, ?, 896]> input_9_cast_fp16 = add(x = linear_0_cast_fp16, y = var_129_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")];
            tensor<int32, []> var_144 = const()[name = tensor<string, []>("op_144"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12385024)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12386880)))];
            tensor<fp16, []> var_147_to_fp16 = const()[name = tensor<string, []>("op_147_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_5_cast_fp16 = layer_norm(axes = x_5_axes_0, beta = layers_0_self_attn_layer_norm_bias_to_fp16, epsilon = var_147_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, [896, 896]> layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12388736))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13191616))), name = tensor<string, []>("layers_0_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13192192)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_168_cast_fp16 = reshape(shape = concat_2x, x = linear_1_cast_fp16)[name = tensor<string, []>("op_168_cast_fp16")];
            tensor<fp16, [896, 896]> layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13194048))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13996928))), name = tensor<string, []>("layers_0_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13997504)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_174_cast_fp16 = reshape(shape = concat_3x, x = linear_2_cast_fp16)[name = tensor<string, []>("op_174_cast_fp16")];
            tensor<fp16, [896, 896]> layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13999360))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14802240))), name = tensor<string, []>("layers_0_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14802816)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_180_cast_fp16 = reshape(shape = concat_4x, x = linear_3_cast_fp16)[name = tensor<string, []>("op_180_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_183_transpose_x_0 = const()[name = tensor<string, []>("op_183_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_183_transpose_y_0 = const()[name = tensor<string, []>("op_183_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_72_perm_0 = const()[name = tensor<string, []>("transpose_72_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_73_perm_0 = const()[name = tensor<string, []>("transpose_73_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 14, 64, ?]> transpose_73 = transpose(perm = transpose_73_perm_0, x = var_174_cast_fp16)[name = tensor<string, []>("transpose_178")];
            tensor<fp16, [1, 14, ?, 64]> transpose_72 = transpose(perm = transpose_72_perm_0, x = var_168_cast_fp16)[name = tensor<string, []>("transpose_179")];
            tensor<fp16, [1, 14, ?, ?]> var_183_cast_fp16 = matmul(transpose_x = var_183_transpose_x_0, transpose_y = var_183_transpose_y_0, x = transpose_72, y = transpose_73)[name = tensor<string, []>("op_183_cast_fp16")];
            tensor<fp16, []> var_184_to_fp16 = const()[name = tensor<string, []>("op_184_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_11_cast_fp16 = mul(x = var_183_cast_fp16, y = var_184_to_fp16)[name = tensor<string, []>("input_11_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_1_cast_fp16 = softmax(axis = var_144, 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, 14, ?, 64]> v_1_cast_fp16 = transpose(perm = v_1_perm_0, x = var_180_cast_fp16)[name = tensor<string, []>("transpose_177")];
            tensor<fp16, [1, 14, ?, 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_188_perm_0 = const()[name = tensor<string, []>("op_188_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_188_cast_fp16 = transpose(perm = var_188_perm_0, x = out_1_cast_fp16)[name = tensor<string, []>("transpose_176")];
            tensor<fp16, [1, ?, 896]> input_13_cast_fp16 = reshape(shape = concat_5x, x = var_188_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")];
            tensor<fp16, [896, 896]> layers_0_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14804672))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15607552))), name = tensor<string, []>("layers_0_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15608128)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_0_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15609984)))];
            tensor<fp16, [896]> layers_0_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15611840)))];
            tensor<fp16, [1, ?, 896]> input_17_cast_fp16 = layer_norm(axes = input_17_axes_0, beta = layers_0_final_layer_norm_bias_to_fp16, epsilon = var_147_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, [3584, 896]> layers_0_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15613696))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18825024))), name = tensor<string, []>("layers_0_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_0_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18825600)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = linear_5_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_0_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18832832))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22044160))), name = tensor<string, []>("layers_0_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_0_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22044736)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_214 = const()[name = tensor<string, []>("op_214"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22046592)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22048448)))];
            tensor<fp16, []> var_217_to_fp16 = const()[name = tensor<string, []>("op_217_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_11_cast_fp16 = layer_norm(axes = x_11_axes_0, beta = layers_1_self_attn_layer_norm_bias_to_fp16, epsilon = var_217_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, [896, 896]> layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22050304))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22853184))), name = tensor<string, []>("layers_1_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22853760)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_238_cast_fp16 = reshape(shape = concat_6x, x = linear_7_cast_fp16)[name = tensor<string, []>("op_238_cast_fp16")];
            tensor<fp16, [896, 896]> layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22855616))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23658496))), name = tensor<string, []>("layers_1_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23659072)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_244_cast_fp16 = reshape(shape = concat_7x, x = linear_8_cast_fp16)[name = tensor<string, []>("op_244_cast_fp16")];
            tensor<fp16, [896, 896]> layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23660928))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24463808))), name = tensor<string, []>("layers_1_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24464384)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_250_cast_fp16 = reshape(shape = concat_8x, x = linear_9_cast_fp16)[name = tensor<string, []>("op_250_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_253_transpose_x_0 = const()[name = tensor<string, []>("op_253_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_253_transpose_y_0 = const()[name = tensor<string, []>("op_253_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_74_perm_0 = const()[name = tensor<string, []>("transpose_74_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_75_perm_0 = const()[name = tensor<string, []>("transpose_75_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 14, 64, ?]> transpose_75 = transpose(perm = transpose_75_perm_0, x = var_244_cast_fp16)[name = tensor<string, []>("transpose_174")];
            tensor<fp16, [1, 14, ?, 64]> transpose_74 = transpose(perm = transpose_74_perm_0, x = var_238_cast_fp16)[name = tensor<string, []>("transpose_175")];
            tensor<fp16, [1, 14, ?, ?]> var_253_cast_fp16 = matmul(transpose_x = var_253_transpose_x_0, transpose_y = var_253_transpose_y_0, x = transpose_74, y = transpose_75)[name = tensor<string, []>("op_253_cast_fp16")];
            tensor<fp16, []> var_254_to_fp16 = const()[name = tensor<string, []>("op_254_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_23_cast_fp16 = mul(x = var_253_cast_fp16, y = var_254_to_fp16)[name = tensor<string, []>("input_23_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_3_cast_fp16 = softmax(axis = var_214, 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, 14, ?, 64]> v_3_cast_fp16 = transpose(perm = v_3_perm_0, x = var_250_cast_fp16)[name = tensor<string, []>("transpose_173")];
            tensor<fp16, [1, 14, ?, 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_258_perm_0 = const()[name = tensor<string, []>("op_258_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_258_cast_fp16 = transpose(perm = var_258_perm_0, x = out_3_cast_fp16)[name = tensor<string, []>("transpose_172")];
            tensor<fp16, [1, ?, 896]> input_25_cast_fp16 = reshape(shape = concat_9x, x = var_258_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")];
            tensor<fp16, [896, 896]> layers_1_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24466240))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25269120))), name = tensor<string, []>("layers_1_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25269696)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_1_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25271552)))];
            tensor<fp16, [896]> layers_1_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25273408)))];
            tensor<fp16, [1, ?, 896]> input_29_cast_fp16 = layer_norm(axes = input_29_axes_0, beta = layers_1_final_layer_norm_bias_to_fp16, epsilon = var_217_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, [3584, 896]> layers_1_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25275264))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28486592))), name = tensor<string, []>("layers_1_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_1_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28487168)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = linear_11_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_1_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28494400))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31705728))), name = tensor<string, []>("layers_1_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_1_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31706304)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_284 = const()[name = tensor<string, []>("op_284"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31708160)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31710016)))];
            tensor<fp16, []> var_287_to_fp16 = const()[name = tensor<string, []>("op_287_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_17_cast_fp16 = layer_norm(axes = x_17_axes_0, beta = layers_2_self_attn_layer_norm_bias_to_fp16, epsilon = var_287_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, [896, 896]> layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31711872))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32514752))), name = tensor<string, []>("layers_2_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32515328)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_308_cast_fp16 = reshape(shape = concat_10x, x = linear_13_cast_fp16)[name = tensor<string, []>("op_308_cast_fp16")];
            tensor<fp16, [896, 896]> layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32517184))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33320064))), name = tensor<string, []>("layers_2_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33320640)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_314_cast_fp16 = reshape(shape = concat_11x, x = linear_14_cast_fp16)[name = tensor<string, []>("op_314_cast_fp16")];
            tensor<fp16, [896, 896]> layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33322496))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34125376))), name = tensor<string, []>("layers_2_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34125952)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_320_cast_fp16 = reshape(shape = concat_12x, x = linear_15_cast_fp16)[name = tensor<string, []>("op_320_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_323_transpose_x_0 = const()[name = tensor<string, []>("op_323_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_323_transpose_y_0 = const()[name = tensor<string, []>("op_323_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_76_perm_0 = const()[name = tensor<string, []>("transpose_76_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_77_perm_0 = const()[name = tensor<string, []>("transpose_77_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 14, 64, ?]> transpose_77 = transpose(perm = transpose_77_perm_0, x = var_314_cast_fp16)[name = tensor<string, []>("transpose_170")];
            tensor<fp16, [1, 14, ?, 64]> transpose_76 = transpose(perm = transpose_76_perm_0, x = var_308_cast_fp16)[name = tensor<string, []>("transpose_171")];
            tensor<fp16, [1, 14, ?, ?]> var_323_cast_fp16 = matmul(transpose_x = var_323_transpose_x_0, transpose_y = var_323_transpose_y_0, x = transpose_76, y = transpose_77)[name = tensor<string, []>("op_323_cast_fp16")];
            tensor<fp16, []> var_324_to_fp16 = const()[name = tensor<string, []>("op_324_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_35_cast_fp16 = mul(x = var_323_cast_fp16, y = var_324_to_fp16)[name = tensor<string, []>("input_35_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_5_cast_fp16 = softmax(axis = var_284, 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, 14, ?, 64]> v_5_cast_fp16 = transpose(perm = v_5_perm_0, x = var_320_cast_fp16)[name = tensor<string, []>("transpose_169")];
            tensor<fp16, [1, 14, ?, 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_328_perm_0 = const()[name = tensor<string, []>("op_328_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_328_cast_fp16 = transpose(perm = var_328_perm_0, x = out_5_cast_fp16)[name = tensor<string, []>("transpose_168")];
            tensor<fp16, [1, ?, 896]> input_37_cast_fp16 = reshape(shape = concat_13x, x = var_328_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")];
            tensor<fp16, [896, 896]> layers_2_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34127808))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34930688))), name = tensor<string, []>("layers_2_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34931264)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_2_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34933120)))];
            tensor<fp16, [896]> layers_2_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34934976)))];
            tensor<fp16, [1, ?, 896]> input_41_cast_fp16 = layer_norm(axes = input_41_axes_0, beta = layers_2_final_layer_norm_bias_to_fp16, epsilon = var_287_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, [3584, 896]> layers_2_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34936832))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38148160))), name = tensor<string, []>("layers_2_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_2_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38148736)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_43_cast_fp16 = gelu(mode = input_43_mode_0, x = linear_17_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_2_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38155968))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41367296))), name = tensor<string, []>("layers_2_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_2_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41367872)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_354 = const()[name = tensor<string, []>("op_354"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41369728)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41371584)))];
            tensor<fp16, []> var_357_to_fp16 = const()[name = tensor<string, []>("op_357_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_23_cast_fp16 = layer_norm(axes = x_23_axes_0, beta = layers_3_self_attn_layer_norm_bias_to_fp16, epsilon = var_357_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, [896, 896]> layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41373440))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42176320))), name = tensor<string, []>("layers_3_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42176896)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_378_cast_fp16 = reshape(shape = concat_14x, x = linear_19_cast_fp16)[name = tensor<string, []>("op_378_cast_fp16")];
            tensor<fp16, [896, 896]> layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42178752))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42981632))), name = tensor<string, []>("layers_3_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42982208)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_384_cast_fp16 = reshape(shape = concat_15x, x = linear_20_cast_fp16)[name = tensor<string, []>("op_384_cast_fp16")];
            tensor<fp16, [896, 896]> layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42984064))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43786944))), name = tensor<string, []>("layers_3_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43787520)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_390_cast_fp16 = reshape(shape = concat_16x, x = linear_21_cast_fp16)[name = tensor<string, []>("op_390_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_393_transpose_x_0 = const()[name = tensor<string, []>("op_393_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_393_transpose_y_0 = const()[name = tensor<string, []>("op_393_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_78_perm_0 = const()[name = tensor<string, []>("transpose_78_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_79_perm_0 = const()[name = tensor<string, []>("transpose_79_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 14, 64, ?]> transpose_79 = transpose(perm = transpose_79_perm_0, x = var_384_cast_fp16)[name = tensor<string, []>("transpose_166")];
            tensor<fp16, [1, 14, ?, 64]> transpose_78 = transpose(perm = transpose_78_perm_0, x = var_378_cast_fp16)[name = tensor<string, []>("transpose_167")];
            tensor<fp16, [1, 14, ?, ?]> var_393_cast_fp16 = matmul(transpose_x = var_393_transpose_x_0, transpose_y = var_393_transpose_y_0, x = transpose_78, y = transpose_79)[name = tensor<string, []>("op_393_cast_fp16")];
            tensor<fp16, []> var_394_to_fp16 = const()[name = tensor<string, []>("op_394_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_47_cast_fp16 = mul(x = var_393_cast_fp16, y = var_394_to_fp16)[name = tensor<string, []>("input_47_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_7_cast_fp16 = softmax(axis = var_354, 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, 14, ?, 64]> v_7_cast_fp16 = transpose(perm = v_7_perm_0, x = var_390_cast_fp16)[name = tensor<string, []>("transpose_165")];
            tensor<fp16, [1, 14, ?, 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_398_perm_0 = const()[name = tensor<string, []>("op_398_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_398_cast_fp16 = transpose(perm = var_398_perm_0, x = out_7_cast_fp16)[name = tensor<string, []>("transpose_164")];
            tensor<fp16, [1, ?, 896]> input_49_cast_fp16 = reshape(shape = concat_17x, x = var_398_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")];
            tensor<fp16, [896, 896]> layers_3_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43789376))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44592256))), name = tensor<string, []>("layers_3_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44592832)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_3_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44594688)))];
            tensor<fp16, [896]> layers_3_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44596544)))];
            tensor<fp16, [1, ?, 896]> input_53_cast_fp16 = layer_norm(axes = input_53_axes_0, beta = layers_3_final_layer_norm_bias_to_fp16, epsilon = var_357_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, [3584, 896]> layers_3_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44598400))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47809728))), name = tensor<string, []>("layers_3_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_3_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47810304)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_55_cast_fp16 = gelu(mode = input_55_mode_0, x = linear_23_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_3_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47817536))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51028864))), name = tensor<string, []>("layers_3_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_3_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51029440)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_424 = const()[name = tensor<string, []>("op_424"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51031296)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51033152)))];
            tensor<fp16, []> var_427_to_fp16 = const()[name = tensor<string, []>("op_427_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_29_cast_fp16 = layer_norm(axes = x_29_axes_0, beta = layers_4_self_attn_layer_norm_bias_to_fp16, epsilon = var_427_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, [896, 896]> layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51035008))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51837888))), name = tensor<string, []>("layers_4_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51838464)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_448_cast_fp16 = reshape(shape = concat_18x, x = linear_25_cast_fp16)[name = tensor<string, []>("op_448_cast_fp16")];
            tensor<fp16, [896, 896]> layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51840320))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52643200))), name = tensor<string, []>("layers_4_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52643776)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_454_cast_fp16 = reshape(shape = concat_19x, x = linear_26_cast_fp16)[name = tensor<string, []>("op_454_cast_fp16")];
            tensor<fp16, [896, 896]> layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52645632))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53448512))), name = tensor<string, []>("layers_4_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53449088)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_460_cast_fp16 = reshape(shape = concat_20x, x = linear_27_cast_fp16)[name = tensor<string, []>("op_460_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_463_transpose_x_0 = const()[name = tensor<string, []>("op_463_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_463_transpose_y_0 = const()[name = tensor<string, []>("op_463_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_80_perm_0 = const()[name = tensor<string, []>("transpose_80_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_81_perm_0 = const()[name = tensor<string, []>("transpose_81_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 14, 64, ?]> transpose_81 = transpose(perm = transpose_81_perm_0, x = var_454_cast_fp16)[name = tensor<string, []>("transpose_162")];
            tensor<fp16, [1, 14, ?, 64]> transpose_80 = transpose(perm = transpose_80_perm_0, x = var_448_cast_fp16)[name = tensor<string, []>("transpose_163")];
            tensor<fp16, [1, 14, ?, ?]> var_463_cast_fp16 = matmul(transpose_x = var_463_transpose_x_0, transpose_y = var_463_transpose_y_0, x = transpose_80, y = transpose_81)[name = tensor<string, []>("op_463_cast_fp16")];
            tensor<fp16, []> var_464_to_fp16 = const()[name = tensor<string, []>("op_464_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_59_cast_fp16 = mul(x = var_463_cast_fp16, y = var_464_to_fp16)[name = tensor<string, []>("input_59_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_9_cast_fp16 = softmax(axis = var_424, 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, 14, ?, 64]> v_9_cast_fp16 = transpose(perm = v_9_perm_0, x = var_460_cast_fp16)[name = tensor<string, []>("transpose_161")];
            tensor<fp16, [1, 14, ?, 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_468_perm_0 = const()[name = tensor<string, []>("op_468_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_468_cast_fp16 = transpose(perm = var_468_perm_0, x = out_9_cast_fp16)[name = tensor<string, []>("transpose_160")];
            tensor<fp16, [1, ?, 896]> input_61_cast_fp16 = reshape(shape = concat_21x, x = var_468_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")];
            tensor<fp16, [896, 896]> layers_4_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53450944))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54253824))), name = tensor<string, []>("layers_4_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54254400)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_4_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54256256)))];
            tensor<fp16, [896]> layers_4_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54258112)))];
            tensor<fp16, [1, ?, 896]> input_65_cast_fp16 = layer_norm(axes = input_65_axes_0, beta = layers_4_final_layer_norm_bias_to_fp16, epsilon = var_427_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, [3584, 896]> layers_4_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54259968))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57471296))), name = tensor<string, []>("layers_4_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_4_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57471872)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_67_cast_fp16 = gelu(mode = input_67_mode_0, x = linear_29_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_4_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57479104))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60690432))), name = tensor<string, []>("layers_4_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_4_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60691008)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_494 = const()[name = tensor<string, []>("op_494"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60692864)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60694720)))];
            tensor<fp16, []> var_497_to_fp16 = const()[name = tensor<string, []>("op_497_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_35_cast_fp16 = layer_norm(axes = x_35_axes_0, beta = layers_5_self_attn_layer_norm_bias_to_fp16, epsilon = var_497_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, [896, 896]> layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60696576))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61499456))), name = tensor<string, []>("layers_5_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61500032)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_518_cast_fp16 = reshape(shape = concat_22x, x = linear_31_cast_fp16)[name = tensor<string, []>("op_518_cast_fp16")];
            tensor<fp16, [896, 896]> layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61501888))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62304768))), name = tensor<string, []>("layers_5_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62305344)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_524_cast_fp16 = reshape(shape = concat_23x, x = linear_32_cast_fp16)[name = tensor<string, []>("op_524_cast_fp16")];
            tensor<fp16, [896, 896]> layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62307200))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63110080))), name = tensor<string, []>("layers_5_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63110656)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_530_cast_fp16 = reshape(shape = concat_24x, x = linear_33_cast_fp16)[name = tensor<string, []>("op_530_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_533_transpose_x_0 = const()[name = tensor<string, []>("op_533_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_533_transpose_y_0 = const()[name = tensor<string, []>("op_533_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_82_perm_0 = const()[name = tensor<string, []>("transpose_82_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_83_perm_0 = const()[name = tensor<string, []>("transpose_83_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 14, 64, ?]> transpose_83 = transpose(perm = transpose_83_perm_0, x = var_524_cast_fp16)[name = tensor<string, []>("transpose_158")];
            tensor<fp16, [1, 14, ?, 64]> transpose_82 = transpose(perm = transpose_82_perm_0, x = var_518_cast_fp16)[name = tensor<string, []>("transpose_159")];
            tensor<fp16, [1, 14, ?, ?]> var_533_cast_fp16 = matmul(transpose_x = var_533_transpose_x_0, transpose_y = var_533_transpose_y_0, x = transpose_82, y = transpose_83)[name = tensor<string, []>("op_533_cast_fp16")];
            tensor<fp16, []> var_534_to_fp16 = const()[name = tensor<string, []>("op_534_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_71_cast_fp16 = mul(x = var_533_cast_fp16, y = var_534_to_fp16)[name = tensor<string, []>("input_71_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_11_cast_fp16 = softmax(axis = var_494, 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, 14, ?, 64]> v_11_cast_fp16 = transpose(perm = v_11_perm_0, x = var_530_cast_fp16)[name = tensor<string, []>("transpose_157")];
            tensor<fp16, [1, 14, ?, 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_538_perm_0 = const()[name = tensor<string, []>("op_538_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_538_cast_fp16 = transpose(perm = var_538_perm_0, x = out_11_cast_fp16)[name = tensor<string, []>("transpose_156")];
            tensor<fp16, [1, ?, 896]> input_73_cast_fp16 = reshape(shape = concat_25x, x = var_538_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")];
            tensor<fp16, [896, 896]> layers_5_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63112512))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63915392))), name = tensor<string, []>("layers_5_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63915968)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_5_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63917824)))];
            tensor<fp16, [896]> layers_5_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63919680)))];
            tensor<fp16, [1, ?, 896]> input_77_cast_fp16 = layer_norm(axes = input_77_axes_0, beta = layers_5_final_layer_norm_bias_to_fp16, epsilon = var_497_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, [3584, 896]> layers_5_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63921536))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67132864))), name = tensor<string, []>("layers_5_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_5_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67133440)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = linear_35_cast_fp16)[name = tensor<string, []>("input_79_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_5_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67140672))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70352000))), name = tensor<string, []>("layers_5_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_5_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70352576)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_564 = const()[name = tensor<string, []>("op_564"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70354432)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70356288)))];
            tensor<fp16, []> var_567_to_fp16 = const()[name = tensor<string, []>("op_567_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_41_cast_fp16 = layer_norm(axes = x_41_axes_0, beta = layers_6_self_attn_layer_norm_bias_to_fp16, epsilon = var_567_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, [896, 896]> layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70358144))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71161024))), name = tensor<string, []>("layers_6_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71161600)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_588_cast_fp16 = reshape(shape = concat_26x, x = linear_37_cast_fp16)[name = tensor<string, []>("op_588_cast_fp16")];
            tensor<fp16, [896, 896]> layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71163456))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71966336))), name = tensor<string, []>("layers_6_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71966912)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_594_cast_fp16 = reshape(shape = concat_27x, x = linear_38_cast_fp16)[name = tensor<string, []>("op_594_cast_fp16")];
            tensor<fp16, [896, 896]> layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71968768))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72771648))), name = tensor<string, []>("layers_6_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72772224)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_600_cast_fp16 = reshape(shape = concat_28x, x = linear_39_cast_fp16)[name = tensor<string, []>("op_600_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_603_transpose_x_0 = const()[name = tensor<string, []>("op_603_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_603_transpose_y_0 = const()[name = tensor<string, []>("op_603_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_84_perm_0 = const()[name = tensor<string, []>("transpose_84_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_85_perm_0 = const()[name = tensor<string, []>("transpose_85_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 14, 64, ?]> transpose_85 = transpose(perm = transpose_85_perm_0, x = var_594_cast_fp16)[name = tensor<string, []>("transpose_154")];
            tensor<fp16, [1, 14, ?, 64]> transpose_84 = transpose(perm = transpose_84_perm_0, x = var_588_cast_fp16)[name = tensor<string, []>("transpose_155")];
            tensor<fp16, [1, 14, ?, ?]> var_603_cast_fp16 = matmul(transpose_x = var_603_transpose_x_0, transpose_y = var_603_transpose_y_0, x = transpose_84, y = transpose_85)[name = tensor<string, []>("op_603_cast_fp16")];
            tensor<fp16, []> var_604_to_fp16 = const()[name = tensor<string, []>("op_604_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_83_cast_fp16 = mul(x = var_603_cast_fp16, y = var_604_to_fp16)[name = tensor<string, []>("input_83_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_13_cast_fp16 = softmax(axis = var_564, 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, 14, ?, 64]> v_13_cast_fp16 = transpose(perm = v_13_perm_0, x = var_600_cast_fp16)[name = tensor<string, []>("transpose_153")];
            tensor<fp16, [1, 14, ?, 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_608_perm_0 = const()[name = tensor<string, []>("op_608_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_608_cast_fp16 = transpose(perm = var_608_perm_0, x = out_13_cast_fp16)[name = tensor<string, []>("transpose_152")];
            tensor<fp16, [1, ?, 896]> input_85_cast_fp16 = reshape(shape = concat_29x, x = var_608_cast_fp16)[name = tensor<string, []>("input_85_cast_fp16")];
            tensor<fp16, [896, 896]> layers_6_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72774080))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73576960))), name = tensor<string, []>("layers_6_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73577536)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_6_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73579392)))];
            tensor<fp16, [896]> layers_6_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73581248)))];
            tensor<fp16, [1, ?, 896]> input_89_cast_fp16 = layer_norm(axes = input_89_axes_0, beta = layers_6_final_layer_norm_bias_to_fp16, epsilon = var_567_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, [3584, 896]> layers_6_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73583104))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76794432))), name = tensor<string, []>("layers_6_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_6_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76795008)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_91_cast_fp16 = gelu(mode = input_91_mode_0, x = linear_41_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_6_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76802240))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80013568))), name = tensor<string, []>("layers_6_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_6_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80014144)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_634 = const()[name = tensor<string, []>("op_634"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80016000)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80017856)))];
            tensor<fp16, []> var_637_to_fp16 = const()[name = tensor<string, []>("op_637_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_47_cast_fp16 = layer_norm(axes = x_47_axes_0, beta = layers_7_self_attn_layer_norm_bias_to_fp16, epsilon = var_637_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, [896, 896]> layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80019712))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80822592))), name = tensor<string, []>("layers_7_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80823168)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_658_cast_fp16 = reshape(shape = concat_30x, x = linear_43_cast_fp16)[name = tensor<string, []>("op_658_cast_fp16")];
            tensor<fp16, [896, 896]> layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80825024))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81627904))), name = tensor<string, []>("layers_7_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81628480)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_664_cast_fp16 = reshape(shape = concat_31x, x = linear_44_cast_fp16)[name = tensor<string, []>("op_664_cast_fp16")];
            tensor<fp16, [896, 896]> layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81630336))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82433216))), name = tensor<string, []>("layers_7_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82433792)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_670_cast_fp16 = reshape(shape = concat_32x, x = linear_45_cast_fp16)[name = tensor<string, []>("op_670_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_673_transpose_x_0 = const()[name = tensor<string, []>("op_673_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_673_transpose_y_0 = const()[name = tensor<string, []>("op_673_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_86_perm_0 = const()[name = tensor<string, []>("transpose_86_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_87_perm_0 = const()[name = tensor<string, []>("transpose_87_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 14, 64, ?]> transpose_87 = transpose(perm = transpose_87_perm_0, x = var_664_cast_fp16)[name = tensor<string, []>("transpose_150")];
            tensor<fp16, [1, 14, ?, 64]> transpose_86 = transpose(perm = transpose_86_perm_0, x = var_658_cast_fp16)[name = tensor<string, []>("transpose_151")];
            tensor<fp16, [1, 14, ?, ?]> var_673_cast_fp16 = matmul(transpose_x = var_673_transpose_x_0, transpose_y = var_673_transpose_y_0, x = transpose_86, y = transpose_87)[name = tensor<string, []>("op_673_cast_fp16")];
            tensor<fp16, []> var_674_to_fp16 = const()[name = tensor<string, []>("op_674_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_95_cast_fp16 = mul(x = var_673_cast_fp16, y = var_674_to_fp16)[name = tensor<string, []>("input_95_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_15_cast_fp16 = softmax(axis = var_634, 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, 14, ?, 64]> v_15_cast_fp16 = transpose(perm = v_15_perm_0, x = var_670_cast_fp16)[name = tensor<string, []>("transpose_149")];
            tensor<fp16, [1, 14, ?, 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_678_perm_0 = const()[name = tensor<string, []>("op_678_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_678_cast_fp16 = transpose(perm = var_678_perm_0, x = out_15_cast_fp16)[name = tensor<string, []>("transpose_148")];
            tensor<fp16, [1, ?, 896]> input_97_cast_fp16 = reshape(shape = concat_33x, x = var_678_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")];
            tensor<fp16, [896, 896]> layers_7_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82435648))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83238528))), name = tensor<string, []>("layers_7_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83239104)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_7_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83240960)))];
            tensor<fp16, [896]> layers_7_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83242816)))];
            tensor<fp16, [1, ?, 896]> input_101_cast_fp16 = layer_norm(axes = input_101_axes_0, beta = layers_7_final_layer_norm_bias_to_fp16, epsilon = var_637_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, [3584, 896]> layers_7_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83244672))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86456000))), name = tensor<string, []>("layers_7_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_7_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86456576)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_103_cast_fp16 = gelu(mode = input_103_mode_0, x = linear_47_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_7_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86463808))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89675136))), name = tensor<string, []>("layers_7_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_7_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89675712)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_704 = const()[name = tensor<string, []>("op_704"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89677568)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89679424)))];
            tensor<fp16, []> var_707_to_fp16 = const()[name = tensor<string, []>("op_707_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_53_cast_fp16 = layer_norm(axes = x_53_axes_0, beta = layers_8_self_attn_layer_norm_bias_to_fp16, epsilon = var_707_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, [896, 896]> layers_8_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89681280))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90484160))), name = tensor<string, []>("layers_8_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90484736)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_728_cast_fp16 = reshape(shape = concat_34x, x = linear_49_cast_fp16)[name = tensor<string, []>("op_728_cast_fp16")];
            tensor<fp16, [896, 896]> layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90486592))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91289472))), name = tensor<string, []>("layers_8_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91290048)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_734_cast_fp16 = reshape(shape = concat_35x, x = linear_50_cast_fp16)[name = tensor<string, []>("op_734_cast_fp16")];
            tensor<fp16, [896, 896]> layers_8_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91291904))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92094784))), name = tensor<string, []>("layers_8_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92095360)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_740_cast_fp16 = reshape(shape = concat_36x, x = linear_51_cast_fp16)[name = tensor<string, []>("op_740_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_743_transpose_x_0 = const()[name = tensor<string, []>("op_743_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_743_transpose_y_0 = const()[name = tensor<string, []>("op_743_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_88_perm_0 = const()[name = tensor<string, []>("transpose_88_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_89_perm_0 = const()[name = tensor<string, []>("transpose_89_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 14, 64, ?]> transpose_89 = transpose(perm = transpose_89_perm_0, x = var_734_cast_fp16)[name = tensor<string, []>("transpose_146")];
            tensor<fp16, [1, 14, ?, 64]> transpose_88 = transpose(perm = transpose_88_perm_0, x = var_728_cast_fp16)[name = tensor<string, []>("transpose_147")];
            tensor<fp16, [1, 14, ?, ?]> var_743_cast_fp16 = matmul(transpose_x = var_743_transpose_x_0, transpose_y = var_743_transpose_y_0, x = transpose_88, y = transpose_89)[name = tensor<string, []>("op_743_cast_fp16")];
            tensor<fp16, []> var_744_to_fp16 = const()[name = tensor<string, []>("op_744_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_107_cast_fp16 = mul(x = var_743_cast_fp16, y = var_744_to_fp16)[name = tensor<string, []>("input_107_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_17_cast_fp16 = softmax(axis = var_704, 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, 14, ?, 64]> v_17_cast_fp16 = transpose(perm = v_17_perm_0, x = var_740_cast_fp16)[name = tensor<string, []>("transpose_145")];
            tensor<fp16, [1, 14, ?, 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_748_perm_0 = const()[name = tensor<string, []>("op_748_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_748_cast_fp16 = transpose(perm = var_748_perm_0, x = out_17_cast_fp16)[name = tensor<string, []>("transpose_144")];
            tensor<fp16, [1, ?, 896]> input_109_cast_fp16 = reshape(shape = concat_37x, x = var_748_cast_fp16)[name = tensor<string, []>("input_109_cast_fp16")];
            tensor<fp16, [896, 896]> layers_8_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92097216))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92900096))), name = tensor<string, []>("layers_8_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92900672)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_8_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92902528)))];
            tensor<fp16, [896]> layers_8_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92904384)))];
            tensor<fp16, [1, ?, 896]> input_113_cast_fp16 = layer_norm(axes = input_113_axes_0, beta = layers_8_final_layer_norm_bias_to_fp16, epsilon = var_707_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, [3584, 896]> layers_8_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92906240))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96117568))), name = tensor<string, []>("layers_8_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_8_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96118144)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_115_cast_fp16 = gelu(mode = input_115_mode_0, x = linear_53_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_8_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96125376))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99336704))), name = tensor<string, []>("layers_8_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_8_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99337280)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_774 = const()[name = tensor<string, []>("op_774"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99339136)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99340992)))];
            tensor<fp16, []> var_777_to_fp16 = const()[name = tensor<string, []>("op_777_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_59_cast_fp16 = layer_norm(axes = x_59_axes_0, beta = layers_9_self_attn_layer_norm_bias_to_fp16, epsilon = var_777_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, [896, 896]> layers_9_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99342848))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100145728))), name = tensor<string, []>("layers_9_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100146304)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_798_cast_fp16 = reshape(shape = concat_38x, x = linear_55_cast_fp16)[name = tensor<string, []>("op_798_cast_fp16")];
            tensor<fp16, [896, 896]> layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100148160))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100951040))), name = tensor<string, []>("layers_9_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100951616)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_804_cast_fp16 = reshape(shape = concat_39x, x = linear_56_cast_fp16)[name = tensor<string, []>("op_804_cast_fp16")];
            tensor<fp16, [896, 896]> layers_9_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100953472))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101756352))), name = tensor<string, []>("layers_9_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101756928)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_810_cast_fp16 = reshape(shape = concat_40x, x = linear_57_cast_fp16)[name = tensor<string, []>("op_810_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_813_transpose_x_0 = const()[name = tensor<string, []>("op_813_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_813_transpose_y_0 = const()[name = tensor<string, []>("op_813_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_90_perm_0 = const()[name = tensor<string, []>("transpose_90_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_91_perm_0 = const()[name = tensor<string, []>("transpose_91_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 14, 64, ?]> transpose_91 = transpose(perm = transpose_91_perm_0, x = var_804_cast_fp16)[name = tensor<string, []>("transpose_142")];
            tensor<fp16, [1, 14, ?, 64]> transpose_90 = transpose(perm = transpose_90_perm_0, x = var_798_cast_fp16)[name = tensor<string, []>("transpose_143")];
            tensor<fp16, [1, 14, ?, ?]> var_813_cast_fp16 = matmul(transpose_x = var_813_transpose_x_0, transpose_y = var_813_transpose_y_0, x = transpose_90, y = transpose_91)[name = tensor<string, []>("op_813_cast_fp16")];
            tensor<fp16, []> var_814_to_fp16 = const()[name = tensor<string, []>("op_814_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_119_cast_fp16 = mul(x = var_813_cast_fp16, y = var_814_to_fp16)[name = tensor<string, []>("input_119_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_19_cast_fp16 = softmax(axis = var_774, 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, 14, ?, 64]> v_19_cast_fp16 = transpose(perm = v_19_perm_0, x = var_810_cast_fp16)[name = tensor<string, []>("transpose_141")];
            tensor<fp16, [1, 14, ?, 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_818_perm_0 = const()[name = tensor<string, []>("op_818_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_818_cast_fp16 = transpose(perm = var_818_perm_0, x = out_19_cast_fp16)[name = tensor<string, []>("transpose_140")];
            tensor<fp16, [1, ?, 896]> input_121_cast_fp16 = reshape(shape = concat_41x, x = var_818_cast_fp16)[name = tensor<string, []>("input_121_cast_fp16")];
            tensor<fp16, [896, 896]> layers_9_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101758784))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102561664))), name = tensor<string, []>("layers_9_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102562240)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_9_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102564096)))];
            tensor<fp16, [896]> layers_9_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102565952)))];
            tensor<fp16, [1, ?, 896]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = layers_9_final_layer_norm_bias_to_fp16, epsilon = var_777_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, [3584, 896]> layers_9_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102567808))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105779136))), name = tensor<string, []>("layers_9_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_9_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105779712)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_127_cast_fp16 = gelu(mode = input_127_mode_0, x = linear_59_cast_fp16)[name = tensor<string, []>("input_127_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_9_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105786944))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108998272))), name = tensor<string, []>("layers_9_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_9_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108998848)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_844 = const()[name = tensor<string, []>("op_844"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109000704)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109002560)))];
            tensor<fp16, []> var_847_to_fp16 = const()[name = tensor<string, []>("op_847_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_65_cast_fp16 = layer_norm(axes = x_65_axes_0, beta = layers_10_self_attn_layer_norm_bias_to_fp16, epsilon = var_847_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, [896, 896]> layers_10_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109004416))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109807296))), name = tensor<string, []>("layers_10_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109807872)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_868_cast_fp16 = reshape(shape = concat_42x, x = linear_61_cast_fp16)[name = tensor<string, []>("op_868_cast_fp16")];
            tensor<fp16, [896, 896]> layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109809728))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110612608))), name = tensor<string, []>("layers_10_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110613184)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_874_cast_fp16 = reshape(shape = concat_43x, x = linear_62_cast_fp16)[name = tensor<string, []>("op_874_cast_fp16")];
            tensor<fp16, [896, 896]> layers_10_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110615040))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111417920))), name = tensor<string, []>("layers_10_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111418496)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_880_cast_fp16 = reshape(shape = concat_44x, x = linear_63_cast_fp16)[name = tensor<string, []>("op_880_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_883_transpose_x_0 = const()[name = tensor<string, []>("op_883_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_883_transpose_y_0 = const()[name = tensor<string, []>("op_883_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_92_perm_0 = const()[name = tensor<string, []>("transpose_92_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_93_perm_0 = const()[name = tensor<string, []>("transpose_93_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 14, 64, ?]> transpose_93 = transpose(perm = transpose_93_perm_0, x = var_874_cast_fp16)[name = tensor<string, []>("transpose_138")];
            tensor<fp16, [1, 14, ?, 64]> transpose_92 = transpose(perm = transpose_92_perm_0, x = var_868_cast_fp16)[name = tensor<string, []>("transpose_139")];
            tensor<fp16, [1, 14, ?, ?]> var_883_cast_fp16 = matmul(transpose_x = var_883_transpose_x_0, transpose_y = var_883_transpose_y_0, x = transpose_92, y = transpose_93)[name = tensor<string, []>("op_883_cast_fp16")];
            tensor<fp16, []> var_884_to_fp16 = const()[name = tensor<string, []>("op_884_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_131_cast_fp16 = mul(x = var_883_cast_fp16, y = var_884_to_fp16)[name = tensor<string, []>("input_131_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_21_cast_fp16 = softmax(axis = var_844, 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, 14, ?, 64]> v_21_cast_fp16 = transpose(perm = v_21_perm_0, x = var_880_cast_fp16)[name = tensor<string, []>("transpose_137")];
            tensor<fp16, [1, 14, ?, 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_888_perm_0 = const()[name = tensor<string, []>("op_888_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_888_cast_fp16 = transpose(perm = var_888_perm_0, x = out_21_cast_fp16)[name = tensor<string, []>("transpose_136")];
            tensor<fp16, [1, ?, 896]> input_133_cast_fp16 = reshape(shape = concat_45x, x = var_888_cast_fp16)[name = tensor<string, []>("input_133_cast_fp16")];
            tensor<fp16, [896, 896]> layers_10_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111420352))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112223232))), name = tensor<string, []>("layers_10_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112223808)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_10_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112225664)))];
            tensor<fp16, [896]> layers_10_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112227520)))];
            tensor<fp16, [1, ?, 896]> input_137_cast_fp16 = layer_norm(axes = input_137_axes_0, beta = layers_10_final_layer_norm_bias_to_fp16, epsilon = var_847_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, [3584, 896]> layers_10_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112229376))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115440704))), name = tensor<string, []>("layers_10_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_10_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115441280)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_139_cast_fp16 = gelu(mode = input_139_mode_0, x = linear_65_cast_fp16)[name = tensor<string, []>("input_139_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_10_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115448512))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118659840))), name = tensor<string, []>("layers_10_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_10_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118660416)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_914 = const()[name = tensor<string, []>("op_914"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118662272)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118664128)))];
            tensor<fp16, []> var_917_to_fp16 = const()[name = tensor<string, []>("op_917_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_71_cast_fp16 = layer_norm(axes = x_71_axes_0, beta = layers_11_self_attn_layer_norm_bias_to_fp16, epsilon = var_917_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, [896, 896]> layers_11_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118665984))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119468864))), name = tensor<string, []>("layers_11_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119469440)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_938_cast_fp16 = reshape(shape = concat_46x, x = linear_67_cast_fp16)[name = tensor<string, []>("op_938_cast_fp16")];
            tensor<fp16, [896, 896]> layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119471296))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120274176))), name = tensor<string, []>("layers_11_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120274752)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_944_cast_fp16 = reshape(shape = concat_47x, x = linear_68_cast_fp16)[name = tensor<string, []>("op_944_cast_fp16")];
            tensor<fp16, [896, 896]> layers_11_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120276608))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121079488))), name = tensor<string, []>("layers_11_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121080064)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_950_cast_fp16 = reshape(shape = concat_48x, x = linear_69_cast_fp16)[name = tensor<string, []>("op_950_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_953_transpose_x_0 = const()[name = tensor<string, []>("op_953_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_953_transpose_y_0 = const()[name = tensor<string, []>("op_953_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_94_perm_0 = const()[name = tensor<string, []>("transpose_94_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_95_perm_0 = const()[name = tensor<string, []>("transpose_95_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 14, 64, ?]> transpose_95 = transpose(perm = transpose_95_perm_0, x = var_944_cast_fp16)[name = tensor<string, []>("transpose_134")];
            tensor<fp16, [1, 14, ?, 64]> transpose_94 = transpose(perm = transpose_94_perm_0, x = var_938_cast_fp16)[name = tensor<string, []>("transpose_135")];
            tensor<fp16, [1, 14, ?, ?]> var_953_cast_fp16 = matmul(transpose_x = var_953_transpose_x_0, transpose_y = var_953_transpose_y_0, x = transpose_94, y = transpose_95)[name = tensor<string, []>("op_953_cast_fp16")];
            tensor<fp16, []> var_954_to_fp16 = const()[name = tensor<string, []>("op_954_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_143_cast_fp16 = mul(x = var_953_cast_fp16, y = var_954_to_fp16)[name = tensor<string, []>("input_143_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_23_cast_fp16 = softmax(axis = var_914, 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, 14, ?, 64]> v_23_cast_fp16 = transpose(perm = v_23_perm_0, x = var_950_cast_fp16)[name = tensor<string, []>("transpose_133")];
            tensor<fp16, [1, 14, ?, 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_958_perm_0 = const()[name = tensor<string, []>("op_958_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_958_cast_fp16 = transpose(perm = var_958_perm_0, x = out_23_cast_fp16)[name = tensor<string, []>("transpose_132")];
            tensor<fp16, [1, ?, 896]> input_145_cast_fp16 = reshape(shape = concat_49x, x = var_958_cast_fp16)[name = tensor<string, []>("input_145_cast_fp16")];
            tensor<fp16, [896, 896]> layers_11_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121081920))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121884800))), name = tensor<string, []>("layers_11_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121885376)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_11_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121887232)))];
            tensor<fp16, [896]> layers_11_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121889088)))];
            tensor<fp16, [1, ?, 896]> input_149_cast_fp16 = layer_norm(axes = input_149_axes_0, beta = layers_11_final_layer_norm_bias_to_fp16, epsilon = var_917_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, [3584, 896]> layers_11_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121890944))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125102272))), name = tensor<string, []>("layers_11_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_11_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125102848)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_151_cast_fp16 = gelu(mode = input_151_mode_0, x = linear_71_cast_fp16)[name = tensor<string, []>("input_151_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_11_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125110080))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128321408))), name = tensor<string, []>("layers_11_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_11_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128321984)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_984 = const()[name = tensor<string, []>("op_984"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128323840)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128325696)))];
            tensor<fp16, []> var_987_to_fp16 = const()[name = tensor<string, []>("op_987_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_77_cast_fp16 = layer_norm(axes = x_77_axes_0, beta = layers_12_self_attn_layer_norm_bias_to_fp16, epsilon = var_987_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, [896, 896]> layers_12_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128327552))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129130432))), name = tensor<string, []>("layers_12_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129131008)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1008_cast_fp16 = reshape(shape = concat_50x, x = linear_73_cast_fp16)[name = tensor<string, []>("op_1008_cast_fp16")];
            tensor<fp16, [896, 896]> layers_12_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129132864))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129935744))), name = tensor<string, []>("layers_12_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129936320)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1014_cast_fp16 = reshape(shape = concat_51x, x = linear_74_cast_fp16)[name = tensor<string, []>("op_1014_cast_fp16")];
            tensor<fp16, [896, 896]> layers_12_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129938176))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130741056))), name = tensor<string, []>("layers_12_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130741632)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1020_cast_fp16 = reshape(shape = concat_52x, x = linear_75_cast_fp16)[name = tensor<string, []>("op_1020_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_1023_transpose_x_0 = const()[name = tensor<string, []>("op_1023_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1023_transpose_y_0 = const()[name = tensor<string, []>("op_1023_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, 14, 64, ?]> transpose_97 = transpose(perm = transpose_97_perm_0, x = var_1014_cast_fp16)[name = tensor<string, []>("transpose_130")];
            tensor<fp16, [1, 14, ?, 64]> transpose_96 = transpose(perm = transpose_96_perm_0, x = var_1008_cast_fp16)[name = tensor<string, []>("transpose_131")];
            tensor<fp16, [1, 14, ?, ?]> var_1023_cast_fp16 = matmul(transpose_x = var_1023_transpose_x_0, transpose_y = var_1023_transpose_y_0, x = transpose_96, y = transpose_97)[name = tensor<string, []>("op_1023_cast_fp16")];
            tensor<fp16, []> var_1024_to_fp16 = const()[name = tensor<string, []>("op_1024_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_155_cast_fp16 = mul(x = var_1023_cast_fp16, y = var_1024_to_fp16)[name = tensor<string, []>("input_155_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_25_cast_fp16 = softmax(axis = var_984, 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, 14, ?, 64]> v_25_cast_fp16 = transpose(perm = v_25_perm_0, x = var_1020_cast_fp16)[name = tensor<string, []>("transpose_129")];
            tensor<fp16, [1, 14, ?, 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_1028_perm_0 = const()[name = tensor<string, []>("op_1028_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_1028_cast_fp16 = transpose(perm = var_1028_perm_0, x = out_25_cast_fp16)[name = tensor<string, []>("transpose_128")];
            tensor<fp16, [1, ?, 896]> input_157_cast_fp16 = reshape(shape = concat_53x, x = var_1028_cast_fp16)[name = tensor<string, []>("input_157_cast_fp16")];
            tensor<fp16, [896, 896]> layers_12_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130743488))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131546368))), name = tensor<string, []>("layers_12_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131546944)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_12_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_12_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131548800)))];
            tensor<fp16, [896]> layers_12_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131550656)))];
            tensor<fp16, [1, ?, 896]> input_161_cast_fp16 = layer_norm(axes = input_161_axes_0, beta = layers_12_final_layer_norm_bias_to_fp16, epsilon = var_987_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, [3584, 896]> layers_12_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131552512))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134763840))), name = tensor<string, []>("layers_12_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_12_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134764416)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_163_cast_fp16 = gelu(mode = input_163_mode_0, x = linear_77_cast_fp16)[name = tensor<string, []>("input_163_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_12_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134771648))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137982976))), name = tensor<string, []>("layers_12_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_12_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_12_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137983552)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_1054 = const()[name = tensor<string, []>("op_1054"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137985408)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137987264)))];
            tensor<fp16, []> var_1057_to_fp16 = const()[name = tensor<string, []>("op_1057_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_83_cast_fp16 = layer_norm(axes = x_83_axes_0, beta = layers_13_self_attn_layer_norm_bias_to_fp16, epsilon = var_1057_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, [896, 896]> layers_13_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137989120))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138792000))), name = tensor<string, []>("layers_13_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138792576)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1078_cast_fp16 = reshape(shape = concat_54x, x = linear_79_cast_fp16)[name = tensor<string, []>("op_1078_cast_fp16")];
            tensor<fp16, [896, 896]> layers_13_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138794432))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139597312))), name = tensor<string, []>("layers_13_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139597888)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1084_cast_fp16 = reshape(shape = concat_55x, x = linear_80_cast_fp16)[name = tensor<string, []>("op_1084_cast_fp16")];
            tensor<fp16, [896, 896]> layers_13_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139599744))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140402624))), name = tensor<string, []>("layers_13_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140403200)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1090_cast_fp16 = reshape(shape = concat_56x, x = linear_81_cast_fp16)[name = tensor<string, []>("op_1090_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_1093_transpose_x_0 = const()[name = tensor<string, []>("op_1093_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1093_transpose_y_0 = const()[name = tensor<string, []>("op_1093_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, 14, 64, ?]> transpose_99 = transpose(perm = transpose_99_perm_0, x = var_1084_cast_fp16)[name = tensor<string, []>("transpose_126")];
            tensor<fp16, [1, 14, ?, 64]> transpose_98 = transpose(perm = transpose_98_perm_0, x = var_1078_cast_fp16)[name = tensor<string, []>("transpose_127")];
            tensor<fp16, [1, 14, ?, ?]> var_1093_cast_fp16 = matmul(transpose_x = var_1093_transpose_x_0, transpose_y = var_1093_transpose_y_0, x = transpose_98, y = transpose_99)[name = tensor<string, []>("op_1093_cast_fp16")];
            tensor<fp16, []> var_1094_to_fp16 = const()[name = tensor<string, []>("op_1094_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_167_cast_fp16 = mul(x = var_1093_cast_fp16, y = var_1094_to_fp16)[name = tensor<string, []>("input_167_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_27_cast_fp16 = softmax(axis = var_1054, 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, 14, ?, 64]> v_27_cast_fp16 = transpose(perm = v_27_perm_0, x = var_1090_cast_fp16)[name = tensor<string, []>("transpose_125")];
            tensor<fp16, [1, 14, ?, 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_1098_perm_0 = const()[name = tensor<string, []>("op_1098_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_1098_cast_fp16 = transpose(perm = var_1098_perm_0, x = out_27_cast_fp16)[name = tensor<string, []>("transpose_124")];
            tensor<fp16, [1, ?, 896]> input_169_cast_fp16 = reshape(shape = concat_57x, x = var_1098_cast_fp16)[name = tensor<string, []>("input_169_cast_fp16")];
            tensor<fp16, [896, 896]> layers_13_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140405056))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141207936))), name = tensor<string, []>("layers_13_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141208512)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_13_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_13_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141210368)))];
            tensor<fp16, [896]> layers_13_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141212224)))];
            tensor<fp16, [1, ?, 896]> input_173_cast_fp16 = layer_norm(axes = input_173_axes_0, beta = layers_13_final_layer_norm_bias_to_fp16, epsilon = var_1057_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, [3584, 896]> layers_13_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141214080))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144425408))), name = tensor<string, []>("layers_13_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_13_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144425984)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_175_cast_fp16 = gelu(mode = input_175_mode_0, x = linear_83_cast_fp16)[name = tensor<string, []>("input_175_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_13_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144433216))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147644544))), name = tensor<string, []>("layers_13_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_13_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_13_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147645120)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_1124 = const()[name = tensor<string, []>("op_1124"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147646976)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147648832)))];
            tensor<fp16, []> var_1127_to_fp16 = const()[name = tensor<string, []>("op_1127_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_89_cast_fp16 = layer_norm(axes = x_89_axes_0, beta = layers_14_self_attn_layer_norm_bias_to_fp16, epsilon = var_1127_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, [896, 896]> layers_14_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147650688))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148453568))), name = tensor<string, []>("layers_14_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148454144)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1148_cast_fp16 = reshape(shape = concat_58x, x = linear_85_cast_fp16)[name = tensor<string, []>("op_1148_cast_fp16")];
            tensor<fp16, [896, 896]> layers_14_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148456000))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149258880))), name = tensor<string, []>("layers_14_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149259456)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1154_cast_fp16 = reshape(shape = concat_59x, x = linear_86_cast_fp16)[name = tensor<string, []>("op_1154_cast_fp16")];
            tensor<fp16, [896, 896]> layers_14_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149261312))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150064192))), name = tensor<string, []>("layers_14_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150064768)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1160_cast_fp16 = reshape(shape = concat_60x, x = linear_87_cast_fp16)[name = tensor<string, []>("op_1160_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_1163_transpose_x_0 = const()[name = tensor<string, []>("op_1163_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1163_transpose_y_0 = const()[name = tensor<string, []>("op_1163_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, 14, 64, ?]> transpose_101 = transpose(perm = transpose_101_perm_0, x = var_1154_cast_fp16)[name = tensor<string, []>("transpose_122")];
            tensor<fp16, [1, 14, ?, 64]> transpose_100 = transpose(perm = transpose_100_perm_0, x = var_1148_cast_fp16)[name = tensor<string, []>("transpose_123")];
            tensor<fp16, [1, 14, ?, ?]> var_1163_cast_fp16 = matmul(transpose_x = var_1163_transpose_x_0, transpose_y = var_1163_transpose_y_0, x = transpose_100, y = transpose_101)[name = tensor<string, []>("op_1163_cast_fp16")];
            tensor<fp16, []> var_1164_to_fp16 = const()[name = tensor<string, []>("op_1164_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_179_cast_fp16 = mul(x = var_1163_cast_fp16, y = var_1164_to_fp16)[name = tensor<string, []>("input_179_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_29_cast_fp16 = softmax(axis = var_1124, 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, 14, ?, 64]> v_29_cast_fp16 = transpose(perm = v_29_perm_0, x = var_1160_cast_fp16)[name = tensor<string, []>("transpose_121")];
            tensor<fp16, [1, 14, ?, 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_1168_perm_0 = const()[name = tensor<string, []>("op_1168_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_1168_cast_fp16 = transpose(perm = var_1168_perm_0, x = out_29_cast_fp16)[name = tensor<string, []>("transpose_120")];
            tensor<fp16, [1, ?, 896]> input_181_cast_fp16 = reshape(shape = concat_61x, x = var_1168_cast_fp16)[name = tensor<string, []>("input_181_cast_fp16")];
            tensor<fp16, [896, 896]> layers_14_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150066624))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150869504))), name = tensor<string, []>("layers_14_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150870080)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_14_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_14_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150871936)))];
            tensor<fp16, [896]> layers_14_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150873792)))];
            tensor<fp16, [1, ?, 896]> input_185_cast_fp16 = layer_norm(axes = input_185_axes_0, beta = layers_14_final_layer_norm_bias_to_fp16, epsilon = var_1127_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, [3584, 896]> layers_14_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150875648))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154086976))), name = tensor<string, []>("layers_14_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_14_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154087552)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_187_cast_fp16 = gelu(mode = input_187_mode_0, x = linear_89_cast_fp16)[name = tensor<string, []>("input_187_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_14_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154094784))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(157306112))), name = tensor<string, []>("layers_14_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_14_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_14_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(157306688)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_1194 = const()[name = tensor<string, []>("op_1194"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(157308544)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(157310400)))];
            tensor<fp16, []> var_1197_to_fp16 = const()[name = tensor<string, []>("op_1197_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_95_cast_fp16 = layer_norm(axes = x_95_axes_0, beta = layers_15_self_attn_layer_norm_bias_to_fp16, epsilon = var_1197_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, [896, 896]> layers_15_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(157312256))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158115136))), name = tensor<string, []>("layers_15_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158115712)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1218_cast_fp16 = reshape(shape = concat_62x, x = linear_91_cast_fp16)[name = tensor<string, []>("op_1218_cast_fp16")];
            tensor<fp16, [896, 896]> layers_15_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158117568))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158920448))), name = tensor<string, []>("layers_15_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158921024)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1224_cast_fp16 = reshape(shape = concat_63x, x = linear_92_cast_fp16)[name = tensor<string, []>("op_1224_cast_fp16")];
            tensor<fp16, [896, 896]> layers_15_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158922880))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159725760))), name = tensor<string, []>("layers_15_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159726336)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1230_cast_fp16 = reshape(shape = concat_64x, x = linear_93_cast_fp16)[name = tensor<string, []>("op_1230_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_1233_transpose_x_0 = const()[name = tensor<string, []>("op_1233_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1233_transpose_y_0 = const()[name = tensor<string, []>("op_1233_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, 14, 64, ?]> transpose_103 = transpose(perm = transpose_103_perm_0, x = var_1224_cast_fp16)[name = tensor<string, []>("transpose_118")];
            tensor<fp16, [1, 14, ?, 64]> transpose_102 = transpose(perm = transpose_102_perm_0, x = var_1218_cast_fp16)[name = tensor<string, []>("transpose_119")];
            tensor<fp16, [1, 14, ?, ?]> var_1233_cast_fp16 = matmul(transpose_x = var_1233_transpose_x_0, transpose_y = var_1233_transpose_y_0, x = transpose_102, y = transpose_103)[name = tensor<string, []>("op_1233_cast_fp16")];
            tensor<fp16, []> var_1234_to_fp16 = const()[name = tensor<string, []>("op_1234_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_191_cast_fp16 = mul(x = var_1233_cast_fp16, y = var_1234_to_fp16)[name = tensor<string, []>("input_191_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_31_cast_fp16 = softmax(axis = var_1194, 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, 14, ?, 64]> v_31_cast_fp16 = transpose(perm = v_31_perm_0, x = var_1230_cast_fp16)[name = tensor<string, []>("transpose_117")];
            tensor<fp16, [1, 14, ?, 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_1238_perm_0 = const()[name = tensor<string, []>("op_1238_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_1238_cast_fp16 = transpose(perm = var_1238_perm_0, x = out_31_cast_fp16)[name = tensor<string, []>("transpose_116")];
            tensor<fp16, [1, ?, 896]> input_193_cast_fp16 = reshape(shape = concat_65x, x = var_1238_cast_fp16)[name = tensor<string, []>("input_193_cast_fp16")];
            tensor<fp16, [896, 896]> layers_15_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159728192))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160531072))), name = tensor<string, []>("layers_15_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160531648)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_15_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_15_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160533504)))];
            tensor<fp16, [896]> layers_15_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160535360)))];
            tensor<fp16, [1, ?, 896]> input_197_cast_fp16 = layer_norm(axes = input_197_axes_0, beta = layers_15_final_layer_norm_bias_to_fp16, epsilon = var_1197_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, [3584, 896]> layers_15_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160537216))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163748544))), name = tensor<string, []>("layers_15_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_15_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163749120)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_199_cast_fp16 = gelu(mode = input_199_mode_0, x = linear_95_cast_fp16)[name = tensor<string, []>("input_199_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_15_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163756352))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166967680))), name = tensor<string, []>("layers_15_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_15_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_15_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166968256)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_1264 = const()[name = tensor<string, []>("op_1264"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166970112)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166971968)))];
            tensor<fp16, []> var_1267_to_fp16 = const()[name = tensor<string, []>("op_1267_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_101_cast_fp16 = layer_norm(axes = x_101_axes_0, beta = layers_16_self_attn_layer_norm_bias_to_fp16, epsilon = var_1267_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, [896, 896]> layers_16_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166973824))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167776704))), name = tensor<string, []>("layers_16_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167777280)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1288_cast_fp16 = reshape(shape = concat_66x, x = linear_97_cast_fp16)[name = tensor<string, []>("op_1288_cast_fp16")];
            tensor<fp16, [896, 896]> layers_16_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167779136))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(168582016))), name = tensor<string, []>("layers_16_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(168582592)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1294_cast_fp16 = reshape(shape = concat_67x, x = linear_98_cast_fp16)[name = tensor<string, []>("op_1294_cast_fp16")];
            tensor<fp16, [896, 896]> layers_16_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(168584448))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169387328))), name = tensor<string, []>("layers_16_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169387904)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1300_cast_fp16 = reshape(shape = concat_68x, x = linear_99_cast_fp16)[name = tensor<string, []>("op_1300_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_1303_transpose_x_0 = const()[name = tensor<string, []>("op_1303_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1303_transpose_y_0 = const()[name = tensor<string, []>("op_1303_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, 14, 64, ?]> transpose_105 = transpose(perm = transpose_105_perm_0, x = var_1294_cast_fp16)[name = tensor<string, []>("transpose_114")];
            tensor<fp16, [1, 14, ?, 64]> transpose_104 = transpose(perm = transpose_104_perm_0, x = var_1288_cast_fp16)[name = tensor<string, []>("transpose_115")];
            tensor<fp16, [1, 14, ?, ?]> var_1303_cast_fp16 = matmul(transpose_x = var_1303_transpose_x_0, transpose_y = var_1303_transpose_y_0, x = transpose_104, y = transpose_105)[name = tensor<string, []>("op_1303_cast_fp16")];
            tensor<fp16, []> var_1304_to_fp16 = const()[name = tensor<string, []>("op_1304_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_203_cast_fp16 = mul(x = var_1303_cast_fp16, y = var_1304_to_fp16)[name = tensor<string, []>("input_203_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_33_cast_fp16 = softmax(axis = var_1264, 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, 14, ?, 64]> v_33_cast_fp16 = transpose(perm = v_33_perm_0, x = var_1300_cast_fp16)[name = tensor<string, []>("transpose_113")];
            tensor<fp16, [1, 14, ?, 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_1308_perm_0 = const()[name = tensor<string, []>("op_1308_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_1308_cast_fp16 = transpose(perm = var_1308_perm_0, x = out_33_cast_fp16)[name = tensor<string, []>("transpose_112")];
            tensor<fp16, [1, ?, 896]> input_205_cast_fp16 = reshape(shape = concat_69x, x = var_1308_cast_fp16)[name = tensor<string, []>("input_205_cast_fp16")];
            tensor<fp16, [896, 896]> layers_16_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169389760))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170192640))), name = tensor<string, []>("layers_16_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170193216)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_16_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_16_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170195072)))];
            tensor<fp16, [896]> layers_16_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170196928)))];
            tensor<fp16, [1, ?, 896]> input_209_cast_fp16 = layer_norm(axes = input_209_axes_0, beta = layers_16_final_layer_norm_bias_to_fp16, epsilon = var_1267_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, [3584, 896]> layers_16_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170198784))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173410112))), name = tensor<string, []>("layers_16_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_16_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173410688)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_211_cast_fp16 = gelu(mode = input_211_mode_0, x = linear_101_cast_fp16)[name = tensor<string, []>("input_211_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_16_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173417920))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176629248))), name = tensor<string, []>("layers_16_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_16_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_16_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176629824)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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_1334 = const()[name = tensor<string, []>("op_1334"), 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, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176631680)))];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176633536)))];
            tensor<fp16, []> var_1337_to_fp16 = const()[name = tensor<string, []>("op_1337_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> x_107_cast_fp16 = layer_norm(axes = x_107_axes_0, beta = layers_17_self_attn_layer_norm_bias_to_fp16, epsilon = var_1337_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, [896, 896]> layers_17_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176635392))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177438272))), name = tensor<string, []>("layers_17_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177438848)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1358_cast_fp16 = reshape(shape = concat_70x, x = linear_103_cast_fp16)[name = tensor<string, []>("op_1358_cast_fp16")];
            tensor<fp16, [896, 896]> layers_17_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177440704))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178243584))), name = tensor<string, []>("layers_17_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178244160)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1364_cast_fp16 = reshape(shape = concat_71x, x = linear_104_cast_fp16)[name = tensor<string, []>("op_1364_cast_fp16")];
            tensor<fp16, [896, 896]> layers_17_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178246016))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179048896))), name = tensor<string, []>("layers_17_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179049472)))];
            tensor<fp16, [1, ?, 896]> 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, 14, 64])];
            tensor<fp16, [1, ?, 14, 64]> var_1370_cast_fp16 = reshape(shape = concat_72x, x = linear_105_cast_fp16)[name = tensor<string, []>("op_1370_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_1373_transpose_x_0 = const()[name = tensor<string, []>("op_1373_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1373_transpose_y_0 = const()[name = tensor<string, []>("op_1373_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, 14, 64, ?]> transpose_107 = transpose(perm = transpose_107_perm_0, x = var_1364_cast_fp16)[name = tensor<string, []>("transpose_110")];
            tensor<fp16, [1, 14, ?, 64]> transpose_106 = transpose(perm = transpose_106_perm_0, x = var_1358_cast_fp16)[name = tensor<string, []>("transpose_111")];
            tensor<fp16, [1, 14, ?, ?]> var_1373_cast_fp16 = matmul(transpose_x = var_1373_transpose_x_0, transpose_y = var_1373_transpose_y_0, x = transpose_106, y = transpose_107)[name = tensor<string, []>("op_1373_cast_fp16")];
            tensor<fp16, []> var_1374_to_fp16 = const()[name = tensor<string, []>("op_1374_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 14, ?, ?]> input_215_cast_fp16 = mul(x = var_1373_cast_fp16, y = var_1374_to_fp16)[name = tensor<string, []>("input_215_cast_fp16")];
            tensor<fp16, [1, 14, ?, ?]> attn_cast_fp16 = softmax(axis = var_1334, x = input_215_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, 14, ?, 64]> v_cast_fp16 = transpose(perm = v_perm_0, x = var_1370_cast_fp16)[name = tensor<string, []>("transpose_109")];
            tensor<fp16, [1, 14, ?, 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_1378_perm_0 = const()[name = tensor<string, []>("op_1378_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, 896])];
            tensor<fp16, [1, ?, 14, 64]> var_1378_cast_fp16 = transpose(perm = var_1378_perm_0, x = out_cast_fp16)[name = tensor<string, []>("transpose_108")];
            tensor<fp16, [1, ?, 896]> input_217_cast_fp16 = reshape(shape = concat_73x, x = var_1378_cast_fp16)[name = tensor<string, []>("input_217_cast_fp16")];
            tensor<fp16, [896, 896]> layers_17_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179051328))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179854208))), name = tensor<string, []>("layers_17_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> 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, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179854784)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> 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, [896]> layers_17_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("layers_17_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179856640)))];
            tensor<fp16, [896]> layers_17_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179858496)))];
            tensor<fp16, [1, ?, 896]> input_221_cast_fp16 = layer_norm(axes = input_221_axes_0, beta = layers_17_final_layer_norm_bias_to_fp16, epsilon = var_1337_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, [3584, 896]> layers_17_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179860352))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183071680))), name = tensor<string, []>("layers_17_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3584, 896])];
            tensor<fp16, [3584]> layers_17_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183072256)))];
            tensor<fp16, [1, ?, 3584]> 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, ?, 3584]> input_223_cast_fp16 = gelu(mode = input_223_mode_0, x = linear_107_cast_fp16)[name = tensor<string, []>("input_223_cast_fp16")];
            tensor<fp16, [896, 3584]> layers_17_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3211264]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183079488))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186290816))), name = tensor<string, []>("layers_17_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 3584])];
            tensor<fp16, [896]> layers_17_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_17_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186291392)))];
            tensor<fp16, [1, ?, 896]> 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, ?, 896]> input_225_cast_fp16 = add(x = input_219_cast_fp16, y = linear_108_cast_fp16)[name = tensor<string, []>("input_225_cast_fp16")];
            tensor<int32, [1]> input_227_axes_0 = const()[name = tensor<string, []>("input_227_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [896]> ln_post_weight_to_fp16 = const()[name = tensor<string, []>("ln_post_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186293248)))];
            tensor<fp16, [896]> ln_post_bias_to_fp16 = const()[name = tensor<string, []>("ln_post_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186295104)))];
            tensor<fp16, []> var_1398_to_fp16 = const()[name = tensor<string, []>("op_1398_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, ?, 896]> input_227_cast_fp16 = layer_norm(axes = input_227_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_1398_to_fp16, gamma = ln_post_weight_to_fp16, x = input_225_cast_fp16)[name = tensor<string, []>("input_227_cast_fp16")];
            tensor<fp16, [896, 896]> proj1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [802816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186296960))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187099840))), name = tensor<string, []>("proj1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([896, 896])];
            tensor<fp16, [896]> proj1_bias_to_fp16 = const()[name = tensor<string, []>("proj1_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187100416)))];
            tensor<fp16, [1, ?, 896]> linear_109_cast_fp16 = linear(bias = proj1_bias_to_fp16, weight = proj1_weight_to_fp16_palettized, x = input_227_cast_fp16)[name = tensor<string, []>("linear_109_cast_fp16")];
            tensor<string, []> input_mode_0 = const()[name = tensor<string, []>("input_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, ?, 896]> input_cast_fp16 = gelu(mode = input_mode_0, x = linear_109_cast_fp16)[name = tensor<string, []>("input_cast_fp16")];
            tensor<fp16, [1024, 896]> proj2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [917504]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187102272))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188019840))), name = tensor<string, []>("proj2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 896])];
            tensor<fp16, [1024]> proj2_bias_to_fp16 = const()[name = tensor<string, []>("proj2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188020416)))];
            tensor<fp16, [1, ?, 1024]> audio_embeddings = linear(bias = proj2_bias_to_fp16, weight = proj2_weight_to_fp16_palettized, x = input_cast_fp16)[name = tensor<string, []>("linear_110_cast_fp16")];
        } -> (audio_embeddings);
}