diff --git "a/encoder.mlmodelc/model.mil" "b/encoder.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/encoder.mlmodelc/model.mil" @@ -0,0 +1,1106 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}})] +{ + 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_59_pad_type_0 = const()[name = tensor("op_59_pad_type_0"), val = tensor("custom")]; + tensor var_59_pad_0 = const()[name = tensor("op_59_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor var_59_strides_0 = const()[name = tensor("op_59_strides_0"), val = tensor([2, 2])]; + tensor var_59_dilations_0 = const()[name = tensor("op_59_dilations_0"), val = tensor([1, 1])]; + tensor var_59_groups_0 = const()[name = tensor("op_59_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(4480))), 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(5056)))]; + tensor var_59_cast_fp16 = conv(bias = conv2d1_bias_to_fp16, dilations = var_59_dilations_0, groups = var_59_groups_0, pad = var_59_pad_0, pad_type = var_59_pad_type_0, strides = var_59_strides_0, weight = conv2d1_weight_to_fp16_palettized, x = input_1_cast_fp16)[name = tensor("op_59_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_59_cast_fp16)[name = tensor("input_3_cast_fp16")]; + tensor var_73_pad_type_0 = const()[name = tensor("op_73_pad_type_0"), val = tensor("custom")]; + tensor var_73_pad_0 = const()[name = tensor("op_73_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor var_73_strides_0 = const()[name = tensor("op_73_strides_0"), val = tensor([2, 2])]; + tensor var_73_dilations_0 = const()[name = tensor("op_73_dilations_0"), val = tensor([1, 1])]; + tensor var_73_groups_0 = const()[name = tensor("op_73_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(6080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2079744))), 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(2080320)))]; + tensor var_73_cast_fp16 = conv(bias = conv2d2_bias_to_fp16, dilations = var_73_dilations_0, groups = var_73_groups_0, pad = var_73_pad_0, pad_type = var_73_pad_type_0, strides = var_73_strides_0, weight = conv2d2_weight_to_fp16_palettized, x = input_3_cast_fp16)[name = tensor("op_73_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_73_cast_fp16)[name = tensor("input_5_cast_fp16")]; + tensor var_87_pad_type_0 = const()[name = tensor("op_87_pad_type_0"), val = tensor("custom")]; + tensor var_87_pad_0 = const()[name = tensor("op_87_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor var_87_strides_0 = const()[name = tensor("op_87_strides_0"), val = tensor([2, 2])]; + tensor var_87_dilations_0 = const()[name = tensor("op_87_dilations_0"), val = tensor([1, 1])]; + tensor var_87_groups_0 = const()[name = tensor("op_87_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(2081344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4155008))), 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(4155584)))]; + tensor var_87_cast_fp16 = conv(bias = conv2d3_bias_to_fp16, dilations = var_87_dilations_0, groups = var_87_groups_0, pad = var_87_pad_0, pad_type = var_87_pad_type_0, strides = var_87_strides_0, weight = conv2d3_weight_to_fp16_palettized, x = input_5_cast_fp16)[name = tensor("op_87_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_87_cast_fp16)[name = tensor("x_1_cast_fp16")]; + tensor var_108 = const()[name = tensor("op_108"), val = tensor([0, 3, 1, 2])]; + tensor concat_0x = const()[name = tensor("concat_0x"), val = tensor([1, -1, 7680])]; + tensor var_109_cast_fp16 = transpose(perm = var_108, x = x_1_cast_fp16)[name = tensor("transpose_180")]; + tensor input_7_cast_fp16 = reshape(shape = concat_0x, x = var_109_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(4156608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11037952))), name = tensor("conv_out_weight_to_fp16_palettized"), shape = tensor([896, 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(11038528)))]; + 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_118_shape_cast_fp16 = shape(x = linear_0_cast_fp16)[name = tensor("op_118_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_118_shape_cast_fp16_to_uint16_dtype_0 = const()[name = tensor("op_118_shape_cast_fp16_to_uint16_dtype_0"), val = tensor("uint16")]; + tensor gather_4_indices_0_to_uint16 = const()[name = tensor("gather_4_indices_0_to_uint16"), val = tensor(1)]; + tensor var_118_shape_cast_fp16_to_uint16 = cast(dtype = var_118_shape_cast_fp16_to_uint16_dtype_0, x = var_118_shape_cast_fp16)[name = tensor("cast_1")]; + tensor gather_4_cast_uint16 = gather(axis = gather_4_axis_0, batch_dims = gather_4_batch_dims_0, indices = gather_4_indices_0_to_uint16, validate_indices = gather_4_validate_indices_0, x = var_118_shape_cast_fp16_to_uint16)[name = tensor("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(896)]; + 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_129_begin_0 = const()[name = tensor("op_129_begin_0"), val = tensor([0, 0, 0])]; + tensor var_129_end_mask_0 = const()[name = tensor("op_129_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(11040384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12384448))), name = tensor("pos_embed_to_fp16_palettized"), shape = tensor([1, 1500, 896])]; + tensor var_129_cast_fp16 = slice_by_index(begin = var_129_begin_0, end = concat_1, end_mask = var_129_end_mask_0, x = pos_embed_to_fp16_palettized)[name = tensor("op_129_cast_fp16")]; + tensor input_9_cast_fp16 = add(x = linear_0_cast_fp16, y = var_129_cast_fp16)[name = tensor("input_9_cast_fp16")]; + tensor var_144 = const()[name = tensor("op_144"), 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(12385024)))]; + 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(12386880)))]; + tensor var_147_to_fp16 = const()[name = tensor("op_147_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_147_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(12388736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13191616))), name = tensor("layers_0_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(13192192)))]; + 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, 14, 64])]; + tensor var_168_cast_fp16 = reshape(shape = concat_2x, x = linear_1_cast_fp16)[name = tensor("op_168_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(13194048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13996928))), name = tensor("layers_0_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(13997504)))]; + 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, 14, 64])]; + tensor var_174_cast_fp16 = reshape(shape = concat_3x, x = linear_2_cast_fp16)[name = tensor("op_174_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(13999360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14802240))), name = tensor("layers_0_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(14802816)))]; + 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, 14, 64])]; + tensor var_180_cast_fp16 = reshape(shape = concat_4x, x = linear_3_cast_fp16)[name = tensor("op_180_cast_fp16")]; + tensor v_1_perm_0 = const()[name = tensor("v_1_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_183_transpose_x_0 = const()[name = tensor("op_183_transpose_x_0"), val = tensor(false)]; + tensor var_183_transpose_y_0 = const()[name = tensor("op_183_transpose_y_0"), val = tensor(false)]; + tensor transpose_72_perm_0 = const()[name = tensor("transpose_72_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_73_perm_0 = const()[name = tensor("transpose_73_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_73 = transpose(perm = transpose_73_perm_0, x = var_174_cast_fp16)[name = tensor("transpose_178")]; + tensor transpose_72 = transpose(perm = transpose_72_perm_0, x = var_168_cast_fp16)[name = tensor("transpose_179")]; + tensor var_183_cast_fp16 = matmul(transpose_x = var_183_transpose_x_0, transpose_y = var_183_transpose_y_0, x = transpose_72, y = transpose_73)[name = tensor("op_183_cast_fp16")]; + tensor var_184_to_fp16 = const()[name = tensor("op_184_to_fp16"), val = tensor(0x1p-3)]; + tensor input_11_cast_fp16 = mul(x = var_183_cast_fp16, y = var_184_to_fp16)[name = tensor("input_11_cast_fp16")]; + tensor attn_1_cast_fp16 = softmax(axis = var_144, 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_180_cast_fp16)[name = tensor("transpose_177")]; + 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_188_perm_0 = const()[name = tensor("op_188_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_5x = const()[name = tensor("concat_5x"), val = tensor([1, -1, 896])]; + tensor var_188_cast_fp16 = transpose(perm = var_188_perm_0, x = out_1_cast_fp16)[name = tensor("transpose_176")]; + tensor input_13_cast_fp16 = reshape(shape = concat_5x, x = var_188_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(14804672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15607552))), name = tensor("layers_0_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(15608128)))]; + 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(15609984)))]; + 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(15611840)))]; + tensor input_17_cast_fp16 = layer_norm(axes = input_17_axes_0, beta = layers_0_final_layer_norm_bias_to_fp16, epsilon = var_147_to_fp16, gamma = layers_0_final_layer_norm_weight_to_fp16, x = input_15_cast_fp16)[name = tensor("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(15613696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18825024))), name = tensor("layers_0_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(18825600)))]; + 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(18832832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22044160))), name = tensor("layers_0_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(22044736)))]; + 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_214 = const()[name = tensor("op_214"), 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(22046592)))]; + 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(22048448)))]; + tensor var_217_to_fp16 = const()[name = tensor("op_217_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_217_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(22050304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22853184))), name = tensor("layers_1_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(22853760)))]; + 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, 14, 64])]; + tensor var_238_cast_fp16 = reshape(shape = concat_6x, x = linear_7_cast_fp16)[name = tensor("op_238_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(22855616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23658496))), name = tensor("layers_1_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(23659072)))]; + 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, 14, 64])]; + tensor var_244_cast_fp16 = reshape(shape = concat_7x, x = linear_8_cast_fp16)[name = tensor("op_244_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(23660928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24463808))), name = tensor("layers_1_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(24464384)))]; + 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, 14, 64])]; + tensor var_250_cast_fp16 = reshape(shape = concat_8x, x = linear_9_cast_fp16)[name = tensor("op_250_cast_fp16")]; + tensor v_3_perm_0 = const()[name = tensor("v_3_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_253_transpose_x_0 = const()[name = tensor("op_253_transpose_x_0"), val = tensor(false)]; + tensor var_253_transpose_y_0 = const()[name = tensor("op_253_transpose_y_0"), val = tensor(false)]; + tensor transpose_74_perm_0 = const()[name = tensor("transpose_74_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_75_perm_0 = const()[name = tensor("transpose_75_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_75 = transpose(perm = transpose_75_perm_0, x = var_244_cast_fp16)[name = tensor("transpose_174")]; + tensor transpose_74 = transpose(perm = transpose_74_perm_0, x = var_238_cast_fp16)[name = tensor("transpose_175")]; + tensor var_253_cast_fp16 = matmul(transpose_x = var_253_transpose_x_0, transpose_y = var_253_transpose_y_0, x = transpose_74, y = transpose_75)[name = tensor("op_253_cast_fp16")]; + tensor var_254_to_fp16 = const()[name = tensor("op_254_to_fp16"), val = tensor(0x1p-3)]; + tensor input_23_cast_fp16 = mul(x = var_253_cast_fp16, y = var_254_to_fp16)[name = tensor("input_23_cast_fp16")]; + tensor attn_3_cast_fp16 = softmax(axis = var_214, 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_250_cast_fp16)[name = tensor("transpose_173")]; + 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_258_perm_0 = const()[name = tensor("op_258_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_9x = const()[name = tensor("concat_9x"), val = tensor([1, -1, 896])]; + tensor var_258_cast_fp16 = transpose(perm = var_258_perm_0, x = out_3_cast_fp16)[name = tensor("transpose_172")]; + tensor input_25_cast_fp16 = reshape(shape = concat_9x, x = var_258_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(24466240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25269120))), name = tensor("layers_1_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(25269696)))]; + 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(25271552)))]; + 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(25273408)))]; + tensor input_29_cast_fp16 = layer_norm(axes = input_29_axes_0, beta = layers_1_final_layer_norm_bias_to_fp16, epsilon = var_217_to_fp16, gamma = layers_1_final_layer_norm_weight_to_fp16, x = input_27_cast_fp16)[name = tensor("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(25275264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28486592))), name = tensor("layers_1_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(28487168)))]; + 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(28494400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31705728))), name = tensor("layers_1_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(31706304)))]; + 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_284 = const()[name = tensor("op_284"), 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(31708160)))]; + 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(31710016)))]; + tensor var_287_to_fp16 = const()[name = tensor("op_287_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_287_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(31711872))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32514752))), name = tensor("layers_2_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(32515328)))]; + 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, 14, 64])]; + tensor var_308_cast_fp16 = reshape(shape = concat_10x, x = linear_13_cast_fp16)[name = tensor("op_308_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(32517184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33320064))), name = tensor("layers_2_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(33320640)))]; + 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, 14, 64])]; + tensor var_314_cast_fp16 = reshape(shape = concat_11x, x = linear_14_cast_fp16)[name = tensor("op_314_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(33322496))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34125376))), name = tensor("layers_2_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(34125952)))]; + 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, 14, 64])]; + tensor var_320_cast_fp16 = reshape(shape = concat_12x, x = linear_15_cast_fp16)[name = tensor("op_320_cast_fp16")]; + tensor v_5_perm_0 = const()[name = tensor("v_5_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_323_transpose_x_0 = const()[name = tensor("op_323_transpose_x_0"), val = tensor(false)]; + tensor var_323_transpose_y_0 = const()[name = tensor("op_323_transpose_y_0"), val = tensor(false)]; + tensor transpose_76_perm_0 = const()[name = tensor("transpose_76_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_77_perm_0 = const()[name = tensor("transpose_77_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_77 = transpose(perm = transpose_77_perm_0, x = var_314_cast_fp16)[name = tensor("transpose_170")]; + tensor transpose_76 = transpose(perm = transpose_76_perm_0, x = var_308_cast_fp16)[name = tensor("transpose_171")]; + tensor var_323_cast_fp16 = matmul(transpose_x = var_323_transpose_x_0, transpose_y = var_323_transpose_y_0, x = transpose_76, y = transpose_77)[name = tensor("op_323_cast_fp16")]; + tensor var_324_to_fp16 = const()[name = tensor("op_324_to_fp16"), val = tensor(0x1p-3)]; + tensor input_35_cast_fp16 = mul(x = var_323_cast_fp16, y = var_324_to_fp16)[name = tensor("input_35_cast_fp16")]; + tensor attn_5_cast_fp16 = softmax(axis = var_284, 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_320_cast_fp16)[name = tensor("transpose_169")]; + 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_328_perm_0 = const()[name = tensor("op_328_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_13x = const()[name = tensor("concat_13x"), val = tensor([1, -1, 896])]; + tensor var_328_cast_fp16 = transpose(perm = var_328_perm_0, x = out_5_cast_fp16)[name = tensor("transpose_168")]; + tensor input_37_cast_fp16 = reshape(shape = concat_13x, x = var_328_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(34127808))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34930688))), name = tensor("layers_2_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(34931264)))]; + 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(34933120)))]; + 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(34934976)))]; + tensor input_41_cast_fp16 = layer_norm(axes = input_41_axes_0, beta = layers_2_final_layer_norm_bias_to_fp16, epsilon = var_287_to_fp16, gamma = layers_2_final_layer_norm_weight_to_fp16, x = input_39_cast_fp16)[name = tensor("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(34936832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38148160))), name = tensor("layers_2_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(38148736)))]; + 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(38155968))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41367296))), name = tensor("layers_2_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(41367872)))]; + 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_354 = const()[name = tensor("op_354"), 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(41369728)))]; + 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(41371584)))]; + tensor var_357_to_fp16 = const()[name = tensor("op_357_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_357_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(41373440))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42176320))), name = tensor("layers_3_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(42176896)))]; + 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, 14, 64])]; + tensor var_378_cast_fp16 = reshape(shape = concat_14x, x = linear_19_cast_fp16)[name = tensor("op_378_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(42178752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42981632))), name = tensor("layers_3_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(42982208)))]; + 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, 14, 64])]; + tensor var_384_cast_fp16 = reshape(shape = concat_15x, x = linear_20_cast_fp16)[name = tensor("op_384_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(42984064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43786944))), name = tensor("layers_3_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(43787520)))]; + 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, 14, 64])]; + tensor var_390_cast_fp16 = reshape(shape = concat_16x, x = linear_21_cast_fp16)[name = tensor("op_390_cast_fp16")]; + tensor v_7_perm_0 = const()[name = tensor("v_7_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_393_transpose_x_0 = const()[name = tensor("op_393_transpose_x_0"), val = tensor(false)]; + tensor var_393_transpose_y_0 = const()[name = tensor("op_393_transpose_y_0"), val = tensor(false)]; + tensor transpose_78_perm_0 = const()[name = tensor("transpose_78_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_79_perm_0 = const()[name = tensor("transpose_79_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_79 = transpose(perm = transpose_79_perm_0, x = var_384_cast_fp16)[name = tensor("transpose_166")]; + tensor transpose_78 = transpose(perm = transpose_78_perm_0, x = var_378_cast_fp16)[name = tensor("transpose_167")]; + tensor var_393_cast_fp16 = matmul(transpose_x = var_393_transpose_x_0, transpose_y = var_393_transpose_y_0, x = transpose_78, y = transpose_79)[name = tensor("op_393_cast_fp16")]; + tensor var_394_to_fp16 = const()[name = tensor("op_394_to_fp16"), val = tensor(0x1p-3)]; + tensor input_47_cast_fp16 = mul(x = var_393_cast_fp16, y = var_394_to_fp16)[name = tensor("input_47_cast_fp16")]; + tensor attn_7_cast_fp16 = softmax(axis = var_354, 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_390_cast_fp16)[name = tensor("transpose_165")]; + 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_398_perm_0 = const()[name = tensor("op_398_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_17x = const()[name = tensor("concat_17x"), val = tensor([1, -1, 896])]; + tensor var_398_cast_fp16 = transpose(perm = var_398_perm_0, x = out_7_cast_fp16)[name = tensor("transpose_164")]; + tensor input_49_cast_fp16 = reshape(shape = concat_17x, x = var_398_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(43789376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44592256))), name = tensor("layers_3_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(44592832)))]; + 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(44594688)))]; + 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(44596544)))]; + tensor input_53_cast_fp16 = layer_norm(axes = input_53_axes_0, beta = layers_3_final_layer_norm_bias_to_fp16, epsilon = var_357_to_fp16, gamma = layers_3_final_layer_norm_weight_to_fp16, x = input_51_cast_fp16)[name = tensor("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(44598400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47809728))), name = tensor("layers_3_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(47810304)))]; + 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(47817536))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51028864))), name = tensor("layers_3_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(51029440)))]; + 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_424 = const()[name = tensor("op_424"), 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(51031296)))]; + 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(51033152)))]; + tensor var_427_to_fp16 = const()[name = tensor("op_427_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_427_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(51035008))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51837888))), name = tensor("layers_4_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(51838464)))]; + 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, 14, 64])]; + tensor var_448_cast_fp16 = reshape(shape = concat_18x, x = linear_25_cast_fp16)[name = tensor("op_448_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(51840320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(52643200))), name = tensor("layers_4_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(52643776)))]; + 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, 14, 64])]; + tensor var_454_cast_fp16 = reshape(shape = concat_19x, x = linear_26_cast_fp16)[name = tensor("op_454_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(52645632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53448512))), name = tensor("layers_4_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(53449088)))]; + 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, 14, 64])]; + tensor var_460_cast_fp16 = reshape(shape = concat_20x, x = linear_27_cast_fp16)[name = tensor("op_460_cast_fp16")]; + tensor v_9_perm_0 = const()[name = tensor("v_9_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_463_transpose_x_0 = const()[name = tensor("op_463_transpose_x_0"), val = tensor(false)]; + tensor var_463_transpose_y_0 = const()[name = tensor("op_463_transpose_y_0"), val = tensor(false)]; + tensor transpose_80_perm_0 = const()[name = tensor("transpose_80_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_81_perm_0 = const()[name = tensor("transpose_81_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_81 = transpose(perm = transpose_81_perm_0, x = var_454_cast_fp16)[name = tensor("transpose_162")]; + tensor transpose_80 = transpose(perm = transpose_80_perm_0, x = var_448_cast_fp16)[name = tensor("transpose_163")]; + tensor var_463_cast_fp16 = matmul(transpose_x = var_463_transpose_x_0, transpose_y = var_463_transpose_y_0, x = transpose_80, y = transpose_81)[name = tensor("op_463_cast_fp16")]; + tensor var_464_to_fp16 = const()[name = tensor("op_464_to_fp16"), val = tensor(0x1p-3)]; + tensor input_59_cast_fp16 = mul(x = var_463_cast_fp16, y = var_464_to_fp16)[name = tensor("input_59_cast_fp16")]; + tensor attn_9_cast_fp16 = softmax(axis = var_424, 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_460_cast_fp16)[name = tensor("transpose_161")]; + 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_468_perm_0 = const()[name = tensor("op_468_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_21x = const()[name = tensor("concat_21x"), val = tensor([1, -1, 896])]; + tensor var_468_cast_fp16 = transpose(perm = var_468_perm_0, x = out_9_cast_fp16)[name = tensor("transpose_160")]; + tensor input_61_cast_fp16 = reshape(shape = concat_21x, x = var_468_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(53450944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54253824))), name = tensor("layers_4_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(54254400)))]; + 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(54256256)))]; + 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(54258112)))]; + tensor input_65_cast_fp16 = layer_norm(axes = input_65_axes_0, beta = layers_4_final_layer_norm_bias_to_fp16, epsilon = var_427_to_fp16, gamma = layers_4_final_layer_norm_weight_to_fp16, x = input_63_cast_fp16)[name = tensor("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(54259968))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57471296))), name = tensor("layers_4_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(57471872)))]; + 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(57479104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60690432))), name = tensor("layers_4_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(60691008)))]; + 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_494 = const()[name = tensor("op_494"), 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(60692864)))]; + 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(60694720)))]; + tensor var_497_to_fp16 = const()[name = tensor("op_497_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_497_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(60696576))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61499456))), name = tensor("layers_5_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(61500032)))]; + 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, 14, 64])]; + tensor var_518_cast_fp16 = reshape(shape = concat_22x, x = linear_31_cast_fp16)[name = tensor("op_518_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(61501888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62304768))), name = tensor("layers_5_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(62305344)))]; + 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, 14, 64])]; + tensor var_524_cast_fp16 = reshape(shape = concat_23x, x = linear_32_cast_fp16)[name = tensor("op_524_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(62307200))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63110080))), name = tensor("layers_5_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(63110656)))]; + 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, 14, 64])]; + tensor var_530_cast_fp16 = reshape(shape = concat_24x, x = linear_33_cast_fp16)[name = tensor("op_530_cast_fp16")]; + tensor v_11_perm_0 = const()[name = tensor("v_11_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_533_transpose_x_0 = const()[name = tensor("op_533_transpose_x_0"), val = tensor(false)]; + tensor var_533_transpose_y_0 = const()[name = tensor("op_533_transpose_y_0"), val = tensor(false)]; + tensor transpose_82_perm_0 = const()[name = tensor("transpose_82_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_83_perm_0 = const()[name = tensor("transpose_83_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_83 = transpose(perm = transpose_83_perm_0, x = var_524_cast_fp16)[name = tensor("transpose_158")]; + tensor transpose_82 = transpose(perm = transpose_82_perm_0, x = var_518_cast_fp16)[name = tensor("transpose_159")]; + tensor var_533_cast_fp16 = matmul(transpose_x = var_533_transpose_x_0, transpose_y = var_533_transpose_y_0, x = transpose_82, y = transpose_83)[name = tensor("op_533_cast_fp16")]; + tensor var_534_to_fp16 = const()[name = tensor("op_534_to_fp16"), val = tensor(0x1p-3)]; + tensor input_71_cast_fp16 = mul(x = var_533_cast_fp16, y = var_534_to_fp16)[name = tensor("input_71_cast_fp16")]; + tensor attn_11_cast_fp16 = softmax(axis = var_494, 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_530_cast_fp16)[name = tensor("transpose_157")]; + 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_538_perm_0 = const()[name = tensor("op_538_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_25x = const()[name = tensor("concat_25x"), val = tensor([1, -1, 896])]; + tensor var_538_cast_fp16 = transpose(perm = var_538_perm_0, x = out_11_cast_fp16)[name = tensor("transpose_156")]; + tensor input_73_cast_fp16 = reshape(shape = concat_25x, x = var_538_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(63112512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63915392))), name = tensor("layers_5_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(63915968)))]; + 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(63917824)))]; + 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(63919680)))]; + tensor input_77_cast_fp16 = layer_norm(axes = input_77_axes_0, beta = layers_5_final_layer_norm_bias_to_fp16, epsilon = var_497_to_fp16, gamma = layers_5_final_layer_norm_weight_to_fp16, x = input_75_cast_fp16)[name = tensor("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(63921536))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67132864))), name = tensor("layers_5_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(67133440)))]; + 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(67140672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70352000))), name = tensor("layers_5_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(70352576)))]; + 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_564 = const()[name = tensor("op_564"), 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(70354432)))]; + 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(70356288)))]; + tensor var_567_to_fp16 = const()[name = tensor("op_567_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_567_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(70358144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(71161024))), name = tensor("layers_6_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(71161600)))]; + 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, 14, 64])]; + tensor var_588_cast_fp16 = reshape(shape = concat_26x, x = linear_37_cast_fp16)[name = tensor("op_588_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(71163456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(71966336))), name = tensor("layers_6_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(71966912)))]; + 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, 14, 64])]; + tensor var_594_cast_fp16 = reshape(shape = concat_27x, x = linear_38_cast_fp16)[name = tensor("op_594_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(71968768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72771648))), name = tensor("layers_6_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(72772224)))]; + 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, 14, 64])]; + tensor var_600_cast_fp16 = reshape(shape = concat_28x, x = linear_39_cast_fp16)[name = tensor("op_600_cast_fp16")]; + tensor v_13_perm_0 = const()[name = tensor("v_13_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_603_transpose_x_0 = const()[name = tensor("op_603_transpose_x_0"), val = tensor(false)]; + tensor var_603_transpose_y_0 = const()[name = tensor("op_603_transpose_y_0"), val = tensor(false)]; + tensor transpose_84_perm_0 = const()[name = tensor("transpose_84_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_85_perm_0 = const()[name = tensor("transpose_85_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_85 = transpose(perm = transpose_85_perm_0, x = var_594_cast_fp16)[name = tensor("transpose_154")]; + tensor transpose_84 = transpose(perm = transpose_84_perm_0, x = var_588_cast_fp16)[name = tensor("transpose_155")]; + tensor var_603_cast_fp16 = matmul(transpose_x = var_603_transpose_x_0, transpose_y = var_603_transpose_y_0, x = transpose_84, y = transpose_85)[name = tensor("op_603_cast_fp16")]; + tensor var_604_to_fp16 = const()[name = tensor("op_604_to_fp16"), val = tensor(0x1p-3)]; + tensor input_83_cast_fp16 = mul(x = var_603_cast_fp16, y = var_604_to_fp16)[name = tensor("input_83_cast_fp16")]; + tensor attn_13_cast_fp16 = softmax(axis = var_564, 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_600_cast_fp16)[name = tensor("transpose_153")]; + 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_608_perm_0 = const()[name = tensor("op_608_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_29x = const()[name = tensor("concat_29x"), val = tensor([1, -1, 896])]; + tensor var_608_cast_fp16 = transpose(perm = var_608_perm_0, x = out_13_cast_fp16)[name = tensor("transpose_152")]; + tensor input_85_cast_fp16 = reshape(shape = concat_29x, x = var_608_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(72774080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73576960))), name = tensor("layers_6_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(73577536)))]; + 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(73579392)))]; + 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(73581248)))]; + tensor input_89_cast_fp16 = layer_norm(axes = input_89_axes_0, beta = layers_6_final_layer_norm_bias_to_fp16, epsilon = var_567_to_fp16, gamma = layers_6_final_layer_norm_weight_to_fp16, x = input_87_cast_fp16)[name = tensor("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(73583104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76794432))), name = tensor("layers_6_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(76795008)))]; + 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(76802240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80013568))), name = tensor("layers_6_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(80014144)))]; + 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_634 = const()[name = tensor("op_634"), 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(80016000)))]; + 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(80017856)))]; + tensor var_637_to_fp16 = const()[name = tensor("op_637_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_637_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(80019712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80822592))), name = tensor("layers_7_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(80823168)))]; + 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, 14, 64])]; + tensor var_658_cast_fp16 = reshape(shape = concat_30x, x = linear_43_cast_fp16)[name = tensor("op_658_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(80825024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81627904))), name = tensor("layers_7_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(81628480)))]; + 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, 14, 64])]; + tensor var_664_cast_fp16 = reshape(shape = concat_31x, x = linear_44_cast_fp16)[name = tensor("op_664_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(81630336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82433216))), name = tensor("layers_7_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(82433792)))]; + 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, 14, 64])]; + tensor var_670_cast_fp16 = reshape(shape = concat_32x, x = linear_45_cast_fp16)[name = tensor("op_670_cast_fp16")]; + tensor v_15_perm_0 = const()[name = tensor("v_15_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_673_transpose_x_0 = const()[name = tensor("op_673_transpose_x_0"), val = tensor(false)]; + tensor var_673_transpose_y_0 = const()[name = tensor("op_673_transpose_y_0"), val = tensor(false)]; + tensor transpose_86_perm_0 = const()[name = tensor("transpose_86_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_87_perm_0 = const()[name = tensor("transpose_87_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_87 = transpose(perm = transpose_87_perm_0, x = var_664_cast_fp16)[name = tensor("transpose_150")]; + tensor transpose_86 = transpose(perm = transpose_86_perm_0, x = var_658_cast_fp16)[name = tensor("transpose_151")]; + tensor var_673_cast_fp16 = matmul(transpose_x = var_673_transpose_x_0, transpose_y = var_673_transpose_y_0, x = transpose_86, y = transpose_87)[name = tensor("op_673_cast_fp16")]; + tensor var_674_to_fp16 = const()[name = tensor("op_674_to_fp16"), val = tensor(0x1p-3)]; + tensor input_95_cast_fp16 = mul(x = var_673_cast_fp16, y = var_674_to_fp16)[name = tensor("input_95_cast_fp16")]; + tensor attn_15_cast_fp16 = softmax(axis = var_634, 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_670_cast_fp16)[name = tensor("transpose_149")]; + 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_678_perm_0 = const()[name = tensor("op_678_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_33x = const()[name = tensor("concat_33x"), val = tensor([1, -1, 896])]; + tensor var_678_cast_fp16 = transpose(perm = var_678_perm_0, x = out_15_cast_fp16)[name = tensor("transpose_148")]; + tensor input_97_cast_fp16 = reshape(shape = concat_33x, x = var_678_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(82435648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83238528))), name = tensor("layers_7_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(83239104)))]; + 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(83240960)))]; + 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(83242816)))]; + tensor input_101_cast_fp16 = layer_norm(axes = input_101_axes_0, beta = layers_7_final_layer_norm_bias_to_fp16, epsilon = var_637_to_fp16, gamma = layers_7_final_layer_norm_weight_to_fp16, x = input_99_cast_fp16)[name = tensor("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(83244672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86456000))), name = tensor("layers_7_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(86456576)))]; + 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(86463808))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89675136))), name = tensor("layers_7_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(89675712)))]; + 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_704 = const()[name = tensor("op_704"), 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(89677568)))]; + 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(89679424)))]; + tensor var_707_to_fp16 = const()[name = tensor("op_707_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_707_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(89681280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90484160))), name = tensor("layers_8_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(90484736)))]; + 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, 14, 64])]; + tensor var_728_cast_fp16 = reshape(shape = concat_34x, x = linear_49_cast_fp16)[name = tensor("op_728_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(90486592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91289472))), name = tensor("layers_8_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(91290048)))]; + 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, 14, 64])]; + tensor var_734_cast_fp16 = reshape(shape = concat_35x, x = linear_50_cast_fp16)[name = tensor("op_734_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(91291904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92094784))), name = tensor("layers_8_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(92095360)))]; + 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, 14, 64])]; + tensor var_740_cast_fp16 = reshape(shape = concat_36x, x = linear_51_cast_fp16)[name = tensor("op_740_cast_fp16")]; + tensor v_17_perm_0 = const()[name = tensor("v_17_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_743_transpose_x_0 = const()[name = tensor("op_743_transpose_x_0"), val = tensor(false)]; + tensor var_743_transpose_y_0 = const()[name = tensor("op_743_transpose_y_0"), val = tensor(false)]; + tensor transpose_88_perm_0 = const()[name = tensor("transpose_88_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_89_perm_0 = const()[name = tensor("transpose_89_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_89 = transpose(perm = transpose_89_perm_0, x = var_734_cast_fp16)[name = tensor("transpose_146")]; + tensor transpose_88 = transpose(perm = transpose_88_perm_0, x = var_728_cast_fp16)[name = tensor("transpose_147")]; + tensor var_743_cast_fp16 = matmul(transpose_x = var_743_transpose_x_0, transpose_y = var_743_transpose_y_0, x = transpose_88, y = transpose_89)[name = tensor("op_743_cast_fp16")]; + tensor var_744_to_fp16 = const()[name = tensor("op_744_to_fp16"), val = tensor(0x1p-3)]; + tensor input_107_cast_fp16 = mul(x = var_743_cast_fp16, y = var_744_to_fp16)[name = tensor("input_107_cast_fp16")]; + tensor attn_17_cast_fp16 = softmax(axis = var_704, 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_740_cast_fp16)[name = tensor("transpose_145")]; + 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_748_perm_0 = const()[name = tensor("op_748_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_37x = const()[name = tensor("concat_37x"), val = tensor([1, -1, 896])]; + tensor var_748_cast_fp16 = transpose(perm = var_748_perm_0, x = out_17_cast_fp16)[name = tensor("transpose_144")]; + tensor input_109_cast_fp16 = reshape(shape = concat_37x, x = var_748_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(92097216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92900096))), name = tensor("layers_8_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(92900672)))]; + 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(92902528)))]; + 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(92904384)))]; + tensor input_113_cast_fp16 = layer_norm(axes = input_113_axes_0, beta = layers_8_final_layer_norm_bias_to_fp16, epsilon = var_707_to_fp16, gamma = layers_8_final_layer_norm_weight_to_fp16, x = input_111_cast_fp16)[name = tensor("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(92906240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96117568))), name = tensor("layers_8_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(96118144)))]; + 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(96125376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99336704))), name = tensor("layers_8_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(99337280)))]; + 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_774 = const()[name = tensor("op_774"), 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(99339136)))]; + 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(99340992)))]; + tensor var_777_to_fp16 = const()[name = tensor("op_777_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_777_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(99342848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100145728))), name = tensor("layers_9_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(100146304)))]; + 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, 14, 64])]; + tensor var_798_cast_fp16 = reshape(shape = concat_38x, x = linear_55_cast_fp16)[name = tensor("op_798_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(100148160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100951040))), name = tensor("layers_9_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(100951616)))]; + 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, 14, 64])]; + tensor var_804_cast_fp16 = reshape(shape = concat_39x, x = linear_56_cast_fp16)[name = tensor("op_804_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(100953472))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(101756352))), name = tensor("layers_9_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(101756928)))]; + 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, 14, 64])]; + tensor var_810_cast_fp16 = reshape(shape = concat_40x, x = linear_57_cast_fp16)[name = tensor("op_810_cast_fp16")]; + tensor v_19_perm_0 = const()[name = tensor("v_19_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_813_transpose_x_0 = const()[name = tensor("op_813_transpose_x_0"), val = tensor(false)]; + tensor var_813_transpose_y_0 = const()[name = tensor("op_813_transpose_y_0"), val = tensor(false)]; + tensor transpose_90_perm_0 = const()[name = tensor("transpose_90_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_91_perm_0 = const()[name = tensor("transpose_91_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_91 = transpose(perm = transpose_91_perm_0, x = var_804_cast_fp16)[name = tensor("transpose_142")]; + tensor transpose_90 = transpose(perm = transpose_90_perm_0, x = var_798_cast_fp16)[name = tensor("transpose_143")]; + tensor var_813_cast_fp16 = matmul(transpose_x = var_813_transpose_x_0, transpose_y = var_813_transpose_y_0, x = transpose_90, y = transpose_91)[name = tensor("op_813_cast_fp16")]; + tensor var_814_to_fp16 = const()[name = tensor("op_814_to_fp16"), val = tensor(0x1p-3)]; + tensor input_119_cast_fp16 = mul(x = var_813_cast_fp16, y = var_814_to_fp16)[name = tensor("input_119_cast_fp16")]; + tensor attn_19_cast_fp16 = softmax(axis = var_774, 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_810_cast_fp16)[name = tensor("transpose_141")]; + 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_818_perm_0 = const()[name = tensor("op_818_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_41x = const()[name = tensor("concat_41x"), val = tensor([1, -1, 896])]; + tensor var_818_cast_fp16 = transpose(perm = var_818_perm_0, x = out_19_cast_fp16)[name = tensor("transpose_140")]; + tensor input_121_cast_fp16 = reshape(shape = concat_41x, x = var_818_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(101758784))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102561664))), name = tensor("layers_9_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(102562240)))]; + 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(102564096)))]; + 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(102565952)))]; + tensor input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = layers_9_final_layer_norm_bias_to_fp16, epsilon = var_777_to_fp16, gamma = layers_9_final_layer_norm_weight_to_fp16, x = input_123_cast_fp16)[name = tensor("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(102567808))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105779136))), name = tensor("layers_9_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(105779712)))]; + 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(105786944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108998272))), name = tensor("layers_9_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(108998848)))]; + 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_844 = const()[name = tensor("op_844"), 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(109000704)))]; + 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(109002560)))]; + tensor var_847_to_fp16 = const()[name = tensor("op_847_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_847_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(109004416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109807296))), name = tensor("layers_10_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(109807872)))]; + 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, 14, 64])]; + tensor var_868_cast_fp16 = reshape(shape = concat_42x, x = linear_61_cast_fp16)[name = tensor("op_868_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(109809728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110612608))), name = tensor("layers_10_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(110613184)))]; + 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, 14, 64])]; + tensor var_874_cast_fp16 = reshape(shape = concat_43x, x = linear_62_cast_fp16)[name = tensor("op_874_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(110615040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111417920))), name = tensor("layers_10_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(111418496)))]; + 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, 14, 64])]; + tensor var_880_cast_fp16 = reshape(shape = concat_44x, x = linear_63_cast_fp16)[name = tensor("op_880_cast_fp16")]; + tensor v_21_perm_0 = const()[name = tensor("v_21_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_883_transpose_x_0 = const()[name = tensor("op_883_transpose_x_0"), val = tensor(false)]; + tensor var_883_transpose_y_0 = const()[name = tensor("op_883_transpose_y_0"), val = tensor(false)]; + tensor transpose_92_perm_0 = const()[name = tensor("transpose_92_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_93_perm_0 = const()[name = tensor("transpose_93_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_93 = transpose(perm = transpose_93_perm_0, x = var_874_cast_fp16)[name = tensor("transpose_138")]; + tensor transpose_92 = transpose(perm = transpose_92_perm_0, x = var_868_cast_fp16)[name = tensor("transpose_139")]; + tensor var_883_cast_fp16 = matmul(transpose_x = var_883_transpose_x_0, transpose_y = var_883_transpose_y_0, x = transpose_92, y = transpose_93)[name = tensor("op_883_cast_fp16")]; + tensor var_884_to_fp16 = const()[name = tensor("op_884_to_fp16"), val = tensor(0x1p-3)]; + tensor input_131_cast_fp16 = mul(x = var_883_cast_fp16, y = var_884_to_fp16)[name = tensor("input_131_cast_fp16")]; + tensor attn_21_cast_fp16 = softmax(axis = var_844, 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_880_cast_fp16)[name = tensor("transpose_137")]; + 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_888_perm_0 = const()[name = tensor("op_888_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_45x = const()[name = tensor("concat_45x"), val = tensor([1, -1, 896])]; + tensor var_888_cast_fp16 = transpose(perm = var_888_perm_0, x = out_21_cast_fp16)[name = tensor("transpose_136")]; + tensor input_133_cast_fp16 = reshape(shape = concat_45x, x = var_888_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(111420352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112223232))), name = tensor("layers_10_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(112223808)))]; + 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(112225664)))]; + 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(112227520)))]; + tensor input_137_cast_fp16 = layer_norm(axes = input_137_axes_0, beta = layers_10_final_layer_norm_bias_to_fp16, epsilon = var_847_to_fp16, gamma = layers_10_final_layer_norm_weight_to_fp16, x = input_135_cast_fp16)[name = tensor("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(112229376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115440704))), name = tensor("layers_10_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(115441280)))]; + 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(115448512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118659840))), name = tensor("layers_10_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(118660416)))]; + 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_914 = const()[name = tensor("op_914"), 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(118662272)))]; + 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(118664128)))]; + tensor var_917_to_fp16 = const()[name = tensor("op_917_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_917_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(118665984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119468864))), name = tensor("layers_11_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(119469440)))]; + 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, 14, 64])]; + tensor var_938_cast_fp16 = reshape(shape = concat_46x, x = linear_67_cast_fp16)[name = tensor("op_938_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(119471296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120274176))), name = tensor("layers_11_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(120274752)))]; + 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, 14, 64])]; + tensor var_944_cast_fp16 = reshape(shape = concat_47x, x = linear_68_cast_fp16)[name = tensor("op_944_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(120276608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121079488))), name = tensor("layers_11_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(121080064)))]; + 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, 14, 64])]; + tensor var_950_cast_fp16 = reshape(shape = concat_48x, x = linear_69_cast_fp16)[name = tensor("op_950_cast_fp16")]; + tensor v_23_perm_0 = const()[name = tensor("v_23_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_953_transpose_x_0 = const()[name = tensor("op_953_transpose_x_0"), val = tensor(false)]; + tensor var_953_transpose_y_0 = const()[name = tensor("op_953_transpose_y_0"), val = tensor(false)]; + tensor transpose_94_perm_0 = const()[name = tensor("transpose_94_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_95_perm_0 = const()[name = tensor("transpose_95_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_95 = transpose(perm = transpose_95_perm_0, x = var_944_cast_fp16)[name = tensor("transpose_134")]; + tensor transpose_94 = transpose(perm = transpose_94_perm_0, x = var_938_cast_fp16)[name = tensor("transpose_135")]; + tensor var_953_cast_fp16 = matmul(transpose_x = var_953_transpose_x_0, transpose_y = var_953_transpose_y_0, x = transpose_94, y = transpose_95)[name = tensor("op_953_cast_fp16")]; + tensor var_954_to_fp16 = const()[name = tensor("op_954_to_fp16"), val = tensor(0x1p-3)]; + tensor input_143_cast_fp16 = mul(x = var_953_cast_fp16, y = var_954_to_fp16)[name = tensor("input_143_cast_fp16")]; + tensor attn_23_cast_fp16 = softmax(axis = var_914, 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_950_cast_fp16)[name = tensor("transpose_133")]; + 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_958_perm_0 = const()[name = tensor("op_958_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_49x = const()[name = tensor("concat_49x"), val = tensor([1, -1, 896])]; + tensor var_958_cast_fp16 = transpose(perm = var_958_perm_0, x = out_23_cast_fp16)[name = tensor("transpose_132")]; + tensor input_145_cast_fp16 = reshape(shape = concat_49x, x = var_958_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(121081920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121884800))), name = tensor("layers_11_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(121885376)))]; + 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(121887232)))]; + 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(121889088)))]; + tensor input_149_cast_fp16 = layer_norm(axes = input_149_axes_0, beta = layers_11_final_layer_norm_bias_to_fp16, epsilon = var_917_to_fp16, gamma = layers_11_final_layer_norm_weight_to_fp16, x = input_147_cast_fp16)[name = tensor("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(121890944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125102272))), name = tensor("layers_11_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(125102848)))]; + 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(125110080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(128321408))), name = tensor("layers_11_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(128321984)))]; + 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_984 = const()[name = tensor("op_984"), 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(128323840)))]; + 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(128325696)))]; + tensor var_987_to_fp16 = const()[name = tensor("op_987_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_987_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(128327552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129130432))), name = tensor("layers_12_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(129131008)))]; + 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, 14, 64])]; + tensor var_1008_cast_fp16 = reshape(shape = concat_50x, x = linear_73_cast_fp16)[name = tensor("op_1008_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(129132864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129935744))), name = tensor("layers_12_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(129936320)))]; + 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, 14, 64])]; + tensor var_1014_cast_fp16 = reshape(shape = concat_51x, x = linear_74_cast_fp16)[name = tensor("op_1014_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(129938176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130741056))), name = tensor("layers_12_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(130741632)))]; + 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, 14, 64])]; + tensor var_1020_cast_fp16 = reshape(shape = concat_52x, x = linear_75_cast_fp16)[name = tensor("op_1020_cast_fp16")]; + tensor v_25_perm_0 = const()[name = tensor("v_25_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_1023_transpose_x_0 = const()[name = tensor("op_1023_transpose_x_0"), val = tensor(false)]; + tensor var_1023_transpose_y_0 = const()[name = tensor("op_1023_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_1014_cast_fp16)[name = tensor("transpose_130")]; + tensor transpose_96 = transpose(perm = transpose_96_perm_0, x = var_1008_cast_fp16)[name = tensor("transpose_131")]; + tensor var_1023_cast_fp16 = matmul(transpose_x = var_1023_transpose_x_0, transpose_y = var_1023_transpose_y_0, x = transpose_96, y = transpose_97)[name = tensor("op_1023_cast_fp16")]; + tensor var_1024_to_fp16 = const()[name = tensor("op_1024_to_fp16"), val = tensor(0x1p-3)]; + tensor input_155_cast_fp16 = mul(x = var_1023_cast_fp16, y = var_1024_to_fp16)[name = tensor("input_155_cast_fp16")]; + tensor attn_25_cast_fp16 = softmax(axis = var_984, 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_1020_cast_fp16)[name = tensor("transpose_129")]; + 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_1028_perm_0 = const()[name = tensor("op_1028_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_53x = const()[name = tensor("concat_53x"), val = tensor([1, -1, 896])]; + tensor var_1028_cast_fp16 = transpose(perm = var_1028_perm_0, x = out_25_cast_fp16)[name = tensor("transpose_128")]; + tensor input_157_cast_fp16 = reshape(shape = concat_53x, x = var_1028_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(130743488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131546368))), name = tensor("layers_12_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(131546944)))]; + 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(131548800)))]; + 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(131550656)))]; + tensor input_161_cast_fp16 = layer_norm(axes = input_161_axes_0, beta = layers_12_final_layer_norm_bias_to_fp16, epsilon = var_987_to_fp16, gamma = layers_12_final_layer_norm_weight_to_fp16, x = input_159_cast_fp16)[name = tensor("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(131552512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134763840))), name = tensor("layers_12_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(134764416)))]; + 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(134771648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137982976))), name = tensor("layers_12_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(137983552)))]; + 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_1054 = const()[name = tensor("op_1054"), 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(137985408)))]; + 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(137987264)))]; + tensor var_1057_to_fp16 = const()[name = tensor("op_1057_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_1057_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(137989120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138792000))), name = tensor("layers_13_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(138792576)))]; + 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, 14, 64])]; + tensor var_1078_cast_fp16 = reshape(shape = concat_54x, x = linear_79_cast_fp16)[name = tensor("op_1078_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(138794432))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139597312))), name = tensor("layers_13_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(139597888)))]; + 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, 14, 64])]; + tensor var_1084_cast_fp16 = reshape(shape = concat_55x, x = linear_80_cast_fp16)[name = tensor("op_1084_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(139599744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(140402624))), name = tensor("layers_13_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(140403200)))]; + 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, 14, 64])]; + tensor var_1090_cast_fp16 = reshape(shape = concat_56x, x = linear_81_cast_fp16)[name = tensor("op_1090_cast_fp16")]; + tensor v_27_perm_0 = const()[name = tensor("v_27_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_1093_transpose_x_0 = const()[name = tensor("op_1093_transpose_x_0"), val = tensor(false)]; + tensor var_1093_transpose_y_0 = const()[name = tensor("op_1093_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_1084_cast_fp16)[name = tensor("transpose_126")]; + tensor transpose_98 = transpose(perm = transpose_98_perm_0, x = var_1078_cast_fp16)[name = tensor("transpose_127")]; + tensor var_1093_cast_fp16 = matmul(transpose_x = var_1093_transpose_x_0, transpose_y = var_1093_transpose_y_0, x = transpose_98, y = transpose_99)[name = tensor("op_1093_cast_fp16")]; + tensor var_1094_to_fp16 = const()[name = tensor("op_1094_to_fp16"), val = tensor(0x1p-3)]; + tensor input_167_cast_fp16 = mul(x = var_1093_cast_fp16, y = var_1094_to_fp16)[name = tensor("input_167_cast_fp16")]; + tensor attn_27_cast_fp16 = softmax(axis = var_1054, 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_1090_cast_fp16)[name = tensor("transpose_125")]; + 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_1098_perm_0 = const()[name = tensor("op_1098_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_57x = const()[name = tensor("concat_57x"), val = tensor([1, -1, 896])]; + tensor var_1098_cast_fp16 = transpose(perm = var_1098_perm_0, x = out_27_cast_fp16)[name = tensor("transpose_124")]; + tensor input_169_cast_fp16 = reshape(shape = concat_57x, x = var_1098_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(140405056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(141207936))), name = tensor("layers_13_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(141208512)))]; + 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(141210368)))]; + 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(141212224)))]; + tensor input_173_cast_fp16 = layer_norm(axes = input_173_axes_0, beta = layers_13_final_layer_norm_bias_to_fp16, epsilon = var_1057_to_fp16, gamma = layers_13_final_layer_norm_weight_to_fp16, x = input_171_cast_fp16)[name = tensor("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(141214080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144425408))), name = tensor("layers_13_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(144425984)))]; + 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(144433216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147644544))), name = tensor("layers_13_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(147645120)))]; + 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_1124 = const()[name = tensor("op_1124"), 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(147646976)))]; + 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(147648832)))]; + tensor var_1127_to_fp16 = const()[name = tensor("op_1127_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_1127_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(147650688))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148453568))), name = tensor("layers_14_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(148454144)))]; + 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, 14, 64])]; + tensor var_1148_cast_fp16 = reshape(shape = concat_58x, x = linear_85_cast_fp16)[name = tensor("op_1148_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(148456000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149258880))), name = tensor("layers_14_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(149259456)))]; + 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, 14, 64])]; + tensor var_1154_cast_fp16 = reshape(shape = concat_59x, x = linear_86_cast_fp16)[name = tensor("op_1154_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(149261312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150064192))), name = tensor("layers_14_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(150064768)))]; + 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, 14, 64])]; + tensor var_1160_cast_fp16 = reshape(shape = concat_60x, x = linear_87_cast_fp16)[name = tensor("op_1160_cast_fp16")]; + tensor v_29_perm_0 = const()[name = tensor("v_29_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_1163_transpose_x_0 = const()[name = tensor("op_1163_transpose_x_0"), val = tensor(false)]; + tensor var_1163_transpose_y_0 = const()[name = tensor("op_1163_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_1154_cast_fp16)[name = tensor("transpose_122")]; + tensor transpose_100 = transpose(perm = transpose_100_perm_0, x = var_1148_cast_fp16)[name = tensor("transpose_123")]; + tensor var_1163_cast_fp16 = matmul(transpose_x = var_1163_transpose_x_0, transpose_y = var_1163_transpose_y_0, x = transpose_100, y = transpose_101)[name = tensor("op_1163_cast_fp16")]; + tensor var_1164_to_fp16 = const()[name = tensor("op_1164_to_fp16"), val = tensor(0x1p-3)]; + tensor input_179_cast_fp16 = mul(x = var_1163_cast_fp16, y = var_1164_to_fp16)[name = tensor("input_179_cast_fp16")]; + tensor attn_29_cast_fp16 = softmax(axis = var_1124, 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_1160_cast_fp16)[name = tensor("transpose_121")]; + 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_1168_perm_0 = const()[name = tensor("op_1168_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_61x = const()[name = tensor("concat_61x"), val = tensor([1, -1, 896])]; + tensor var_1168_cast_fp16 = transpose(perm = var_1168_perm_0, x = out_29_cast_fp16)[name = tensor("transpose_120")]; + tensor input_181_cast_fp16 = reshape(shape = concat_61x, x = var_1168_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(150066624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150869504))), name = tensor("layers_14_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(150870080)))]; + 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(150871936)))]; + 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(150873792)))]; + tensor input_185_cast_fp16 = layer_norm(axes = input_185_axes_0, beta = layers_14_final_layer_norm_bias_to_fp16, epsilon = var_1127_to_fp16, gamma = layers_14_final_layer_norm_weight_to_fp16, x = input_183_cast_fp16)[name = tensor("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(150875648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154086976))), name = tensor("layers_14_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(154087552)))]; + 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(154094784))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157306112))), name = tensor("layers_14_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(157306688)))]; + 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_1194 = const()[name = tensor("op_1194"), 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(157308544)))]; + 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(157310400)))]; + tensor var_1197_to_fp16 = const()[name = tensor("op_1197_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_1197_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(157312256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158115136))), name = tensor("layers_15_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(158115712)))]; + 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, 14, 64])]; + tensor var_1218_cast_fp16 = reshape(shape = concat_62x, x = linear_91_cast_fp16)[name = tensor("op_1218_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(158117568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158920448))), name = tensor("layers_15_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(158921024)))]; + 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, 14, 64])]; + tensor var_1224_cast_fp16 = reshape(shape = concat_63x, x = linear_92_cast_fp16)[name = tensor("op_1224_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(158922880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159725760))), name = tensor("layers_15_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(159726336)))]; + 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, 14, 64])]; + tensor var_1230_cast_fp16 = reshape(shape = concat_64x, x = linear_93_cast_fp16)[name = tensor("op_1230_cast_fp16")]; + tensor v_31_perm_0 = const()[name = tensor("v_31_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_1233_transpose_x_0 = const()[name = tensor("op_1233_transpose_x_0"), val = tensor(false)]; + tensor var_1233_transpose_y_0 = const()[name = tensor("op_1233_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_1224_cast_fp16)[name = tensor("transpose_118")]; + tensor transpose_102 = transpose(perm = transpose_102_perm_0, x = var_1218_cast_fp16)[name = tensor("transpose_119")]; + tensor var_1233_cast_fp16 = matmul(transpose_x = var_1233_transpose_x_0, transpose_y = var_1233_transpose_y_0, x = transpose_102, y = transpose_103)[name = tensor("op_1233_cast_fp16")]; + tensor var_1234_to_fp16 = const()[name = tensor("op_1234_to_fp16"), val = tensor(0x1p-3)]; + tensor input_191_cast_fp16 = mul(x = var_1233_cast_fp16, y = var_1234_to_fp16)[name = tensor("input_191_cast_fp16")]; + tensor attn_31_cast_fp16 = softmax(axis = var_1194, 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_1230_cast_fp16)[name = tensor("transpose_117")]; + 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_1238_perm_0 = const()[name = tensor("op_1238_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_65x = const()[name = tensor("concat_65x"), val = tensor([1, -1, 896])]; + tensor var_1238_cast_fp16 = transpose(perm = var_1238_perm_0, x = out_31_cast_fp16)[name = tensor("transpose_116")]; + tensor input_193_cast_fp16 = reshape(shape = concat_65x, x = var_1238_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(159728192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(160531072))), name = tensor("layers_15_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(160531648)))]; + 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(160533504)))]; + 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(160535360)))]; + tensor input_197_cast_fp16 = layer_norm(axes = input_197_axes_0, beta = layers_15_final_layer_norm_bias_to_fp16, epsilon = var_1197_to_fp16, gamma = layers_15_final_layer_norm_weight_to_fp16, x = input_195_cast_fp16)[name = tensor("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(160537216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163748544))), name = tensor("layers_15_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(163749120)))]; + 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(163756352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166967680))), name = tensor("layers_15_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(166968256)))]; + 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_1264 = const()[name = tensor("op_1264"), 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(166970112)))]; + 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(166971968)))]; + tensor var_1267_to_fp16 = const()[name = tensor("op_1267_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_1267_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(166973824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(167776704))), name = tensor("layers_16_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(167777280)))]; + 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, 14, 64])]; + tensor var_1288_cast_fp16 = reshape(shape = concat_66x, x = linear_97_cast_fp16)[name = tensor("op_1288_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(167779136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(168582016))), name = tensor("layers_16_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(168582592)))]; + 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, 14, 64])]; + tensor var_1294_cast_fp16 = reshape(shape = concat_67x, x = linear_98_cast_fp16)[name = tensor("op_1294_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(168584448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169387328))), name = tensor("layers_16_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(169387904)))]; + 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, 14, 64])]; + tensor var_1300_cast_fp16 = reshape(shape = concat_68x, x = linear_99_cast_fp16)[name = tensor("op_1300_cast_fp16")]; + tensor v_33_perm_0 = const()[name = tensor("v_33_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_1303_transpose_x_0 = const()[name = tensor("op_1303_transpose_x_0"), val = tensor(false)]; + tensor var_1303_transpose_y_0 = const()[name = tensor("op_1303_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_1294_cast_fp16)[name = tensor("transpose_114")]; + tensor transpose_104 = transpose(perm = transpose_104_perm_0, x = var_1288_cast_fp16)[name = tensor("transpose_115")]; + tensor var_1303_cast_fp16 = matmul(transpose_x = var_1303_transpose_x_0, transpose_y = var_1303_transpose_y_0, x = transpose_104, y = transpose_105)[name = tensor("op_1303_cast_fp16")]; + tensor var_1304_to_fp16 = const()[name = tensor("op_1304_to_fp16"), val = tensor(0x1p-3)]; + tensor input_203_cast_fp16 = mul(x = var_1303_cast_fp16, y = var_1304_to_fp16)[name = tensor("input_203_cast_fp16")]; + tensor attn_33_cast_fp16 = softmax(axis = var_1264, 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_1300_cast_fp16)[name = tensor("transpose_113")]; + 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_1308_perm_0 = const()[name = tensor("op_1308_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_69x = const()[name = tensor("concat_69x"), val = tensor([1, -1, 896])]; + tensor var_1308_cast_fp16 = transpose(perm = var_1308_perm_0, x = out_33_cast_fp16)[name = tensor("transpose_112")]; + tensor input_205_cast_fp16 = reshape(shape = concat_69x, x = var_1308_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(169389760))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(170192640))), name = tensor("layers_16_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(170193216)))]; + 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(170195072)))]; + 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(170196928)))]; + tensor input_209_cast_fp16 = layer_norm(axes = input_209_axes_0, beta = layers_16_final_layer_norm_bias_to_fp16, epsilon = var_1267_to_fp16, gamma = layers_16_final_layer_norm_weight_to_fp16, x = input_207_cast_fp16)[name = tensor("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(170198784))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173410112))), name = tensor("layers_16_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(173410688)))]; + 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(173417920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176629248))), name = tensor("layers_16_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(176629824)))]; + 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_1334 = const()[name = tensor("op_1334"), 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(176631680)))]; + 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(176633536)))]; + tensor var_1337_to_fp16 = const()[name = tensor("op_1337_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_1337_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(176635392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177438272))), name = tensor("layers_17_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(177438848)))]; + 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, 14, 64])]; + tensor var_1358_cast_fp16 = reshape(shape = concat_70x, x = linear_103_cast_fp16)[name = tensor("op_1358_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(177440704))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178243584))), name = tensor("layers_17_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(178244160)))]; + 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, 14, 64])]; + tensor var_1364_cast_fp16 = reshape(shape = concat_71x, x = linear_104_cast_fp16)[name = tensor("op_1364_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(178246016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(179048896))), name = tensor("layers_17_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(179049472)))]; + 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, 14, 64])]; + tensor var_1370_cast_fp16 = reshape(shape = concat_72x, x = linear_105_cast_fp16)[name = tensor("op_1370_cast_fp16")]; + tensor v_perm_0 = const()[name = tensor("v_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor var_1373_transpose_x_0 = const()[name = tensor("op_1373_transpose_x_0"), val = tensor(false)]; + tensor var_1373_transpose_y_0 = const()[name = tensor("op_1373_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_1364_cast_fp16)[name = tensor("transpose_110")]; + tensor transpose_106 = transpose(perm = transpose_106_perm_0, x = var_1358_cast_fp16)[name = tensor("transpose_111")]; + tensor var_1373_cast_fp16 = matmul(transpose_x = var_1373_transpose_x_0, transpose_y = var_1373_transpose_y_0, x = transpose_106, y = transpose_107)[name = tensor("op_1373_cast_fp16")]; + tensor var_1374_to_fp16 = const()[name = tensor("op_1374_to_fp16"), val = tensor(0x1p-3)]; + tensor input_215_cast_fp16 = mul(x = var_1373_cast_fp16, y = var_1374_to_fp16)[name = tensor("input_215_cast_fp16")]; + tensor attn_cast_fp16 = softmax(axis = var_1334, x = input_215_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_1370_cast_fp16)[name = tensor("transpose_109")]; + 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_1378_perm_0 = const()[name = tensor("op_1378_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_73x = const()[name = tensor("concat_73x"), val = tensor([1, -1, 896])]; + tensor var_1378_cast_fp16 = transpose(perm = var_1378_perm_0, x = out_cast_fp16)[name = tensor("transpose_108")]; + tensor input_217_cast_fp16 = reshape(shape = concat_73x, x = var_1378_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(179051328))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(179854208))), name = tensor("layers_17_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + 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(179854784)))]; + 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(179856640)))]; + 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(179858496)))]; + tensor input_221_cast_fp16 = layer_norm(axes = input_221_axes_0, beta = layers_17_final_layer_norm_bias_to_fp16, epsilon = var_1337_to_fp16, gamma = layers_17_final_layer_norm_weight_to_fp16, x = input_219_cast_fp16)[name = tensor("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(179860352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(183071680))), name = tensor("layers_17_fc1_weight_to_fp16_palettized"), shape = tensor([3584, 896])]; + 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(183072256)))]; + 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(183079488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186290816))), name = tensor("layers_17_fc2_weight_to_fp16_palettized"), shape = tensor([896, 3584])]; + 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(186291392)))]; + 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 input_227_axes_0 = const()[name = tensor("input_227_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(186293248)))]; + 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(186295104)))]; + tensor var_1398_to_fp16 = const()[name = tensor("op_1398_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_227_cast_fp16 = layer_norm(axes = input_227_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_1398_to_fp16, gamma = ln_post_weight_to_fp16, x = input_225_cast_fp16)[name = tensor("input_227_cast_fp16")]; + tensor proj1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186296960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187099840))), name = tensor("proj1_weight_to_fp16_palettized"), shape = tensor([896, 896])]; + tensor proj1_bias_to_fp16 = const()[name = tensor("proj1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187100416)))]; + tensor linear_109_cast_fp16 = linear(bias = proj1_bias_to_fp16, weight = proj1_weight_to_fp16_palettized, x = input_227_cast_fp16)[name = tensor("linear_109_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_109_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(187102272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188019840))), name = tensor("proj2_weight_to_fp16_palettized"), shape = tensor([1024, 896])]; + tensor proj2_bias_to_fp16 = const()[name = tensor("proj2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188020416)))]; + tensor audio_embeddings = linear(bias = proj2_bias_to_fp16, weight = proj2_weight_to_fp16_palettized, x = input_cast_fp16)[name = tensor("linear_110_cast_fp16")]; + } -> (audio_embeddings); +} \ No newline at end of file