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