program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}})]
{
    func main<ios18>(tensor<int32, [1, 16, 125]> audio_codes) {
            tensor<int32, [3]> var_401_begin_0 = const()[name = string("op_401_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> var_401_end_0 = const()[name = string("op_401_end_0"), val = tensor<int32, [3]>([1, 1, 125])];
            tensor<bool, [3]> var_401_end_mask_0 = const()[name = string("op_401_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<int32, [1, 1, 125]> var_401 = slice_by_index(begin = var_401_begin_0, end = var_401_end_0, end_mask = var_401_end_mask_0, x = audio_codes)[name = string("op_401")];
            tensor<int32, [3]> var_414_begin_0 = const()[name = string("op_414_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> var_414_end_0 = const()[name = string("op_414_end_0"), val = tensor<int32, [3]>([1, 1, 125])];
            tensor<bool, [3]> var_414_end_mask_0 = const()[name = string("op_414_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_414_squeeze_mask_0 = const()[name = string("op_414_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_414 = slice_by_index(begin = var_414_begin_0, end = var_414_end_0, end_mask = var_414_end_mask_0, squeeze_mask = var_414_squeeze_mask_0, x = var_401)[name = string("op_414")];
            int32 quantized_1_batch_dims_0 = const()[name = string("quantized_1_batch_dims_0"), val = int32(0)];
            bool quantized_1_validate_indices_0 = const()[name = string("quantized_1_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> first_vq_layers_0_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(524416))))[name = string("first_vq_layers_0_embedding_to_fp16_palettized")];
            string var_414_to_int16_dtype_0 = const()[name = string("op_414_to_int16_dtype_0"), val = string("int16")];
            string cast_49_dtype_0 = const()[name = string("cast_49_dtype_0"), val = string("int32")];
            int32 greater_equal_0_y_0 = const()[name = string("greater_equal_0_y_0"), val = int32(0)];
            tensor<int16, [1, 125]> var_414_to_int16 = cast(dtype = var_414_to_int16_dtype_0, x = var_414)[name = string("cast_19")];
            tensor<int32, [1, 125]> cast_49 = cast(dtype = cast_49_dtype_0, x = var_414_to_int16)[name = string("cast_18")];
            tensor<bool, [1, 125]> greater_equal_0 = greater_equal(x = cast_49, y = greater_equal_0_y_0)[name = string("greater_equal_0")];
            int32 slice_by_index_0 = const()[name = string("slice_by_index_0"), val = int32(2048)];
            tensor<int32, [1, 125]> add_0 = add(x = cast_49, y = slice_by_index_0)[name = string("add_0")];
            tensor<int32, [1, 125]> select_0 = select(a = cast_49, b = add_0, cond = greater_equal_0)[name = string("select_0")];
            string select_0_to_int16_dtype_0 = const()[name = string("select_0_to_int16_dtype_0"), val = string("int16")];
            string cast_0_dtype_0 = const()[name = string("cast_0_dtype_0"), val = string("int32")];
            int32 greater_equal_0_y_0_1 = const()[name = string("greater_equal_0_y_0_1"), val = int32(0)];
            tensor<int16, [1, 125]> select_0_to_int16 = cast(dtype = select_0_to_int16_dtype_0, x = select_0)[name = string("cast_17")];
            tensor<int32, [1, 125]> cast_0 = cast(dtype = cast_0_dtype_0, x = select_0_to_int16)[name = string("cast_16")];
            tensor<bool, [1, 125]> greater_equal_0_1 = greater_equal(x = cast_0, y = greater_equal_0_y_0_1)[name = string("greater_equal_0_1")];
            int32 slice_by_index_0_1 = const()[name = string("slice_by_index_0_1"), val = int32(2048)];
            tensor<int32, [1, 125]> add_0_1 = add(x = cast_0, y = slice_by_index_0_1)[name = string("add_0_1")];
            tensor<int32, [1, 125]> select_0_1 = select(a = cast_0, b = add_0_1, cond = greater_equal_0_1)[name = string("select_0_1")];
            int32 quantized_1_cast_fp16_cast_uint16_cast_uint16_axis_0 = const()[name = string("quantized_1_cast_fp16_cast_uint16_cast_uint16_axis_0"), val = int32(0)];
            tensor<fp16, [1, 125, 256]> quantized_1_cast_fp16_cast_uint16_cast_uint16 = gather(axis = quantized_1_cast_fp16_cast_uint16_cast_uint16_axis_0, batch_dims = quantized_1_batch_dims_0, indices = select_0_1, validate_indices = quantized_1_validate_indices_0, x = first_vq_layers_0_embedding_to_fp16_palettized)[name = string("quantized_1_cast_fp16_cast_uint16_cast_uint16")];
            tensor<int32, [3]> input_3_perm_0 = const()[name = string("input_3_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string quantized_first_pad_type_0 = const()[name = string("quantized_first_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> quantized_first_strides_0 = const()[name = string("quantized_first_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> quantized_first_pad_0 = const()[name = string("quantized_first_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> quantized_first_dilations_0 = const()[name = string("quantized_first_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 quantized_first_groups_0 = const()[name = string("quantized_first_groups_0"), val = int32(1)];
            tensor<fp16, [512, 256, 1]> first_output_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(524992))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(656128))))[name = string("first_output_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 256, 125]> input_3_cast_fp16 = transpose(perm = input_3_perm_0, x = quantized_1_cast_fp16_cast_uint16_cast_uint16)[name = string("transpose_39")];
            tensor<fp16, [1, 512, 125]> quantized_first_cast_fp16 = conv(dilations = quantized_first_dilations_0, groups = quantized_first_groups_0, pad = quantized_first_pad_0, pad_type = quantized_first_pad_type_0, strides = quantized_first_strides_0, weight = first_output_proj_weight_to_fp16_palettized, x = input_3_cast_fp16)[name = string("quantized_first_cast_fp16")];
            tensor<int32, [3]> var_447_begin_0 = const()[name = string("op_447_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_447_end_0 = const()[name = string("op_447_end_0"), val = tensor<int32, [3]>([1, 16, 125])];
            tensor<bool, [3]> var_447_end_mask_0 = const()[name = string("op_447_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<int32, [1, 15, 125]> var_447 = slice_by_index(begin = var_447_begin_0, end = var_447_end_0, end_mask = var_447_end_mask_0, x = audio_codes)[name = string("op_447")];
            tensor<int32, [3]> var_460_begin_0 = const()[name = string("op_460_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> var_460_end_0 = const()[name = string("op_460_end_0"), val = tensor<int32, [3]>([1, 1, 125])];
            tensor<bool, [3]> var_460_end_mask_0 = const()[name = string("op_460_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_460_squeeze_mask_0 = const()[name = string("op_460_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_460 = slice_by_index(begin = var_460_begin_0, end = var_460_end_0, end_mask = var_460_end_mask_0, squeeze_mask = var_460_squeeze_mask_0, x = var_447)[name = string("op_460")];
            int32 quantized_3_axis_0 = const()[name = string("quantized_3_axis_0"), val = int32(0)];
            int32 quantized_3_batch_dims_0 = const()[name = string("quantized_3_batch_dims_0"), val = int32(0)];
            bool quantized_3_validate_indices_0 = const()[name = string("quantized_3_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_0_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(656704))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1181056))))[name = string("rest_vq_layers_0_embedding_to_fp16_palettized")];
            string var_460_to_uint16_dtype_0 = const()[name = string("op_460_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_460_to_uint16 = cast(dtype = var_460_to_uint16_dtype_0, x = var_460)[name = string("cast_15")];
            tensor<fp16, [1, 125, 256]> quantized_3_cast_fp16_cast_uint16 = gather(axis = quantized_3_axis_0, batch_dims = quantized_3_batch_dims_0, indices = var_460_to_uint16, validate_indices = quantized_3_validate_indices_0, x = rest_vq_layers_0_embedding_to_fp16_palettized)[name = string("quantized_3_cast_fp16_cast_uint16")];
            tensor<int32, [3]> var_480_begin_0 = const()[name = string("op_480_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_480_end_0 = const()[name = string("op_480_end_0"), val = tensor<int32, [3]>([1, 2, 125])];
            tensor<bool, [3]> var_480_end_mask_0 = const()[name = string("op_480_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_480_squeeze_mask_0 = const()[name = string("op_480_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_480 = slice_by_index(begin = var_480_begin_0, end = var_480_end_0, end_mask = var_480_end_mask_0, squeeze_mask = var_480_squeeze_mask_0, x = var_447)[name = string("op_480")];
            int32 quantized_5_axis_0 = const()[name = string("quantized_5_axis_0"), val = int32(0)];
            int32 quantized_5_batch_dims_0 = const()[name = string("quantized_5_batch_dims_0"), val = int32(0)];
            bool quantized_5_validate_indices_0 = const()[name = string("quantized_5_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_1_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1181632))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1705984))))[name = string("rest_vq_layers_1_embedding_to_fp16_palettized")];
            string var_480_to_uint16_dtype_0 = const()[name = string("op_480_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_480_to_uint16 = cast(dtype = var_480_to_uint16_dtype_0, x = var_480)[name = string("cast_14")];
            tensor<fp16, [1, 125, 256]> quantized_5_cast_fp16_cast_uint16 = gather(axis = quantized_5_axis_0, batch_dims = quantized_5_batch_dims_0, indices = var_480_to_uint16, validate_indices = quantized_5_validate_indices_0, x = rest_vq_layers_1_embedding_to_fp16_palettized)[name = string("quantized_5_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 125, 256]> quantized_rest_sum_3_cast_fp16 = add(x = quantized_3_cast_fp16_cast_uint16, y = quantized_5_cast_fp16_cast_uint16)[name = string("quantized_rest_sum_3_cast_fp16")];
            tensor<int32, [3]> var_502_begin_0 = const()[name = string("op_502_begin_0"), val = tensor<int32, [3]>([0, 2, 0])];
            tensor<int32, [3]> var_502_end_0 = const()[name = string("op_502_end_0"), val = tensor<int32, [3]>([1, 3, 125])];
            tensor<bool, [3]> var_502_end_mask_0 = const()[name = string("op_502_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_502_squeeze_mask_0 = const()[name = string("op_502_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_502 = slice_by_index(begin = var_502_begin_0, end = var_502_end_0, end_mask = var_502_end_mask_0, squeeze_mask = var_502_squeeze_mask_0, x = var_447)[name = string("op_502")];
            int32 quantized_7_axis_0 = const()[name = string("quantized_7_axis_0"), val = int32(0)];
            int32 quantized_7_batch_dims_0 = const()[name = string("quantized_7_batch_dims_0"), val = int32(0)];
            bool quantized_7_validate_indices_0 = const()[name = string("quantized_7_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_2_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1706560))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2230912))))[name = string("rest_vq_layers_2_embedding_to_fp16_palettized")];
            string var_502_to_uint16_dtype_0 = const()[name = string("op_502_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_502_to_uint16 = cast(dtype = var_502_to_uint16_dtype_0, x = var_502)[name = string("cast_13")];
            tensor<fp16, [1, 125, 256]> quantized_7_cast_fp16_cast_uint16 = gather(axis = quantized_7_axis_0, batch_dims = quantized_7_batch_dims_0, indices = var_502_to_uint16, validate_indices = quantized_7_validate_indices_0, x = rest_vq_layers_2_embedding_to_fp16_palettized)[name = string("quantized_7_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 125, 256]> quantized_rest_sum_5_cast_fp16 = add(x = quantized_rest_sum_3_cast_fp16, y = quantized_7_cast_fp16_cast_uint16)[name = string("quantized_rest_sum_5_cast_fp16")];
            tensor<int32, [3]> var_524_begin_0 = const()[name = string("op_524_begin_0"), val = tensor<int32, [3]>([0, 3, 0])];
            tensor<int32, [3]> var_524_end_0 = const()[name = string("op_524_end_0"), val = tensor<int32, [3]>([1, 4, 125])];
            tensor<bool, [3]> var_524_end_mask_0 = const()[name = string("op_524_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_524_squeeze_mask_0 = const()[name = string("op_524_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_524 = slice_by_index(begin = var_524_begin_0, end = var_524_end_0, end_mask = var_524_end_mask_0, squeeze_mask = var_524_squeeze_mask_0, x = var_447)[name = string("op_524")];
            int32 quantized_9_axis_0 = const()[name = string("quantized_9_axis_0"), val = int32(0)];
            int32 quantized_9_batch_dims_0 = const()[name = string("quantized_9_batch_dims_0"), val = int32(0)];
            bool quantized_9_validate_indices_0 = const()[name = string("quantized_9_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_3_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2231488))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2755840))))[name = string("rest_vq_layers_3_embedding_to_fp16_palettized")];
            string var_524_to_uint16_dtype_0 = const()[name = string("op_524_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_524_to_uint16 = cast(dtype = var_524_to_uint16_dtype_0, x = var_524)[name = string("cast_12")];
            tensor<fp16, [1, 125, 256]> quantized_9_cast_fp16_cast_uint16 = gather(axis = quantized_9_axis_0, batch_dims = quantized_9_batch_dims_0, indices = var_524_to_uint16, validate_indices = quantized_9_validate_indices_0, x = rest_vq_layers_3_embedding_to_fp16_palettized)[name = string("quantized_9_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 125, 256]> quantized_rest_sum_7_cast_fp16 = add(x = quantized_rest_sum_5_cast_fp16, y = quantized_9_cast_fp16_cast_uint16)[name = string("quantized_rest_sum_7_cast_fp16")];
            tensor<int32, [3]> var_546_begin_0 = const()[name = string("op_546_begin_0"), val = tensor<int32, [3]>([0, 4, 0])];
            tensor<int32, [3]> var_546_end_0 = const()[name = string("op_546_end_0"), val = tensor<int32, [3]>([1, 5, 125])];
            tensor<bool, [3]> var_546_end_mask_0 = const()[name = string("op_546_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_546_squeeze_mask_0 = const()[name = string("op_546_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_546 = slice_by_index(begin = var_546_begin_0, end = var_546_end_0, end_mask = var_546_end_mask_0, squeeze_mask = var_546_squeeze_mask_0, x = var_447)[name = string("op_546")];
            int32 quantized_11_axis_0 = const()[name = string("quantized_11_axis_0"), val = int32(0)];
            int32 quantized_11_batch_dims_0 = const()[name = string("quantized_11_batch_dims_0"), val = int32(0)];
            bool quantized_11_validate_indices_0 = const()[name = string("quantized_11_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_4_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2756416))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3280768))))[name = string("rest_vq_layers_4_embedding_to_fp16_palettized")];
            string var_546_to_uint16_dtype_0 = const()[name = string("op_546_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_546_to_uint16 = cast(dtype = var_546_to_uint16_dtype_0, x = var_546)[name = string("cast_11")];
            tensor<fp16, [1, 125, 256]> quantized_11_cast_fp16_cast_uint16 = gather(axis = quantized_11_axis_0, batch_dims = quantized_11_batch_dims_0, indices = var_546_to_uint16, validate_indices = quantized_11_validate_indices_0, x = rest_vq_layers_4_embedding_to_fp16_palettized)[name = string("quantized_11_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 125, 256]> quantized_rest_sum_9_cast_fp16 = add(x = quantized_rest_sum_7_cast_fp16, y = quantized_11_cast_fp16_cast_uint16)[name = string("quantized_rest_sum_9_cast_fp16")];
            tensor<int32, [3]> var_568_begin_0 = const()[name = string("op_568_begin_0"), val = tensor<int32, [3]>([0, 5, 0])];
            tensor<int32, [3]> var_568_end_0 = const()[name = string("op_568_end_0"), val = tensor<int32, [3]>([1, 6, 125])];
            tensor<bool, [3]> var_568_end_mask_0 = const()[name = string("op_568_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_568_squeeze_mask_0 = const()[name = string("op_568_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_568 = slice_by_index(begin = var_568_begin_0, end = var_568_end_0, end_mask = var_568_end_mask_0, squeeze_mask = var_568_squeeze_mask_0, x = var_447)[name = string("op_568")];
            int32 quantized_13_axis_0 = const()[name = string("quantized_13_axis_0"), val = int32(0)];
            int32 quantized_13_batch_dims_0 = const()[name = string("quantized_13_batch_dims_0"), val = int32(0)];
            bool quantized_13_validate_indices_0 = const()[name = string("quantized_13_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_5_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3281344))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3805696))))[name = string("rest_vq_layers_5_embedding_to_fp16_palettized")];
            string var_568_to_uint16_dtype_0 = const()[name = string("op_568_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_568_to_uint16 = cast(dtype = var_568_to_uint16_dtype_0, x = var_568)[name = string("cast_10")];
            tensor<fp16, [1, 125, 256]> quantized_13_cast_fp16_cast_uint16 = gather(axis = quantized_13_axis_0, batch_dims = quantized_13_batch_dims_0, indices = var_568_to_uint16, validate_indices = quantized_13_validate_indices_0, x = rest_vq_layers_5_embedding_to_fp16_palettized)[name = string("quantized_13_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 125, 256]> quantized_rest_sum_11_cast_fp16 = add(x = quantized_rest_sum_9_cast_fp16, y = quantized_13_cast_fp16_cast_uint16)[name = string("quantized_rest_sum_11_cast_fp16")];
            tensor<int32, [3]> var_590_begin_0 = const()[name = string("op_590_begin_0"), val = tensor<int32, [3]>([0, 6, 0])];
            tensor<int32, [3]> var_590_end_0 = const()[name = string("op_590_end_0"), val = tensor<int32, [3]>([1, 7, 125])];
            tensor<bool, [3]> var_590_end_mask_0 = const()[name = string("op_590_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_590_squeeze_mask_0 = const()[name = string("op_590_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_590 = slice_by_index(begin = var_590_begin_0, end = var_590_end_0, end_mask = var_590_end_mask_0, squeeze_mask = var_590_squeeze_mask_0, x = var_447)[name = string("op_590")];
            int32 quantized_15_axis_0 = const()[name = string("quantized_15_axis_0"), val = int32(0)];
            int32 quantized_15_batch_dims_0 = const()[name = string("quantized_15_batch_dims_0"), val = int32(0)];
            bool quantized_15_validate_indices_0 = const()[name = string("quantized_15_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_6_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3806272))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4330624))))[name = string("rest_vq_layers_6_embedding_to_fp16_palettized")];
            string var_590_to_uint16_dtype_0 = const()[name = string("op_590_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_590_to_uint16 = cast(dtype = var_590_to_uint16_dtype_0, x = var_590)[name = string("cast_9")];
            tensor<fp16, [1, 125, 256]> quantized_15_cast_fp16_cast_uint16 = gather(axis = quantized_15_axis_0, batch_dims = quantized_15_batch_dims_0, indices = var_590_to_uint16, validate_indices = quantized_15_validate_indices_0, x = rest_vq_layers_6_embedding_to_fp16_palettized)[name = string("quantized_15_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 125, 256]> quantized_rest_sum_13_cast_fp16 = add(x = quantized_rest_sum_11_cast_fp16, y = quantized_15_cast_fp16_cast_uint16)[name = string("quantized_rest_sum_13_cast_fp16")];
            tensor<int32, [3]> var_612_begin_0 = const()[name = string("op_612_begin_0"), val = tensor<int32, [3]>([0, 7, 0])];
            tensor<int32, [3]> var_612_end_0 = const()[name = string("op_612_end_0"), val = tensor<int32, [3]>([1, 8, 125])];
            tensor<bool, [3]> var_612_end_mask_0 = const()[name = string("op_612_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_612_squeeze_mask_0 = const()[name = string("op_612_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_612 = slice_by_index(begin = var_612_begin_0, end = var_612_end_0, end_mask = var_612_end_mask_0, squeeze_mask = var_612_squeeze_mask_0, x = var_447)[name = string("op_612")];
            int32 quantized_17_axis_0 = const()[name = string("quantized_17_axis_0"), val = int32(0)];
            int32 quantized_17_batch_dims_0 = const()[name = string("quantized_17_batch_dims_0"), val = int32(0)];
            bool quantized_17_validate_indices_0 = const()[name = string("quantized_17_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_7_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4331200))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4855552))))[name = string("rest_vq_layers_7_embedding_to_fp16_palettized")];
            string var_612_to_uint16_dtype_0 = const()[name = string("op_612_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_612_to_uint16 = cast(dtype = var_612_to_uint16_dtype_0, x = var_612)[name = string("cast_8")];
            tensor<fp16, [1, 125, 256]> quantized_17_cast_fp16_cast_uint16 = gather(axis = quantized_17_axis_0, batch_dims = quantized_17_batch_dims_0, indices = var_612_to_uint16, validate_indices = quantized_17_validate_indices_0, x = rest_vq_layers_7_embedding_to_fp16_palettized)[name = string("quantized_17_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 125, 256]> quantized_rest_sum_15_cast_fp16 = add(x = quantized_rest_sum_13_cast_fp16, y = quantized_17_cast_fp16_cast_uint16)[name = string("quantized_rest_sum_15_cast_fp16")];
            tensor<int32, [3]> var_634_begin_0 = const()[name = string("op_634_begin_0"), val = tensor<int32, [3]>([0, 8, 0])];
            tensor<int32, [3]> var_634_end_0 = const()[name = string("op_634_end_0"), val = tensor<int32, [3]>([1, 9, 125])];
            tensor<bool, [3]> var_634_end_mask_0 = const()[name = string("op_634_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_634_squeeze_mask_0 = const()[name = string("op_634_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_634 = slice_by_index(begin = var_634_begin_0, end = var_634_end_0, end_mask = var_634_end_mask_0, squeeze_mask = var_634_squeeze_mask_0, x = var_447)[name = string("op_634")];
            int32 quantized_19_axis_0 = const()[name = string("quantized_19_axis_0"), val = int32(0)];
            int32 quantized_19_batch_dims_0 = const()[name = string("quantized_19_batch_dims_0"), val = int32(0)];
            bool quantized_19_validate_indices_0 = const()[name = string("quantized_19_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_8_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4856128))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5380480))))[name = string("rest_vq_layers_8_embedding_to_fp16_palettized")];
            string var_634_to_uint16_dtype_0 = const()[name = string("op_634_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_634_to_uint16 = cast(dtype = var_634_to_uint16_dtype_0, x = var_634)[name = string("cast_7")];
            tensor<fp16, [1, 125, 256]> quantized_19_cast_fp16_cast_uint16 = gather(axis = quantized_19_axis_0, batch_dims = quantized_19_batch_dims_0, indices = var_634_to_uint16, validate_indices = quantized_19_validate_indices_0, x = rest_vq_layers_8_embedding_to_fp16_palettized)[name = string("quantized_19_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 125, 256]> quantized_rest_sum_17_cast_fp16 = add(x = quantized_rest_sum_15_cast_fp16, y = quantized_19_cast_fp16_cast_uint16)[name = string("quantized_rest_sum_17_cast_fp16")];
            tensor<int32, [3]> var_656_begin_0 = const()[name = string("op_656_begin_0"), val = tensor<int32, [3]>([0, 9, 0])];
            tensor<int32, [3]> var_656_end_0 = const()[name = string("op_656_end_0"), val = tensor<int32, [3]>([1, 10, 125])];
            tensor<bool, [3]> var_656_end_mask_0 = const()[name = string("op_656_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_656_squeeze_mask_0 = const()[name = string("op_656_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_656 = slice_by_index(begin = var_656_begin_0, end = var_656_end_0, end_mask = var_656_end_mask_0, squeeze_mask = var_656_squeeze_mask_0, x = var_447)[name = string("op_656")];
            int32 quantized_21_axis_0 = const()[name = string("quantized_21_axis_0"), val = int32(0)];
            int32 quantized_21_batch_dims_0 = const()[name = string("quantized_21_batch_dims_0"), val = int32(0)];
            bool quantized_21_validate_indices_0 = const()[name = string("quantized_21_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_9_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5381056))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5905408))))[name = string("rest_vq_layers_9_embedding_to_fp16_palettized")];
            string var_656_to_uint16_dtype_0 = const()[name = string("op_656_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_656_to_uint16 = cast(dtype = var_656_to_uint16_dtype_0, x = var_656)[name = string("cast_6")];
            tensor<fp16, [1, 125, 256]> quantized_21_cast_fp16_cast_uint16 = gather(axis = quantized_21_axis_0, batch_dims = quantized_21_batch_dims_0, indices = var_656_to_uint16, validate_indices = quantized_21_validate_indices_0, x = rest_vq_layers_9_embedding_to_fp16_palettized)[name = string("quantized_21_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 125, 256]> quantized_rest_sum_19_cast_fp16 = add(x = quantized_rest_sum_17_cast_fp16, y = quantized_21_cast_fp16_cast_uint16)[name = string("quantized_rest_sum_19_cast_fp16")];
            tensor<int32, [3]> var_678_begin_0 = const()[name = string("op_678_begin_0"), val = tensor<int32, [3]>([0, 10, 0])];
            tensor<int32, [3]> var_678_end_0 = const()[name = string("op_678_end_0"), val = tensor<int32, [3]>([1, 11, 125])];
            tensor<bool, [3]> var_678_end_mask_0 = const()[name = string("op_678_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_678_squeeze_mask_0 = const()[name = string("op_678_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_678 = slice_by_index(begin = var_678_begin_0, end = var_678_end_0, end_mask = var_678_end_mask_0, squeeze_mask = var_678_squeeze_mask_0, x = var_447)[name = string("op_678")];
            int32 quantized_23_axis_0 = const()[name = string("quantized_23_axis_0"), val = int32(0)];
            int32 quantized_23_batch_dims_0 = const()[name = string("quantized_23_batch_dims_0"), val = int32(0)];
            bool quantized_23_validate_indices_0 = const()[name = string("quantized_23_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_10_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5905984))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6430336))))[name = string("rest_vq_layers_10_embedding_to_fp16_palettized")];
            string var_678_to_uint16_dtype_0 = const()[name = string("op_678_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_678_to_uint16 = cast(dtype = var_678_to_uint16_dtype_0, x = var_678)[name = string("cast_5")];
            tensor<fp16, [1, 125, 256]> quantized_23_cast_fp16_cast_uint16 = gather(axis = quantized_23_axis_0, batch_dims = quantized_23_batch_dims_0, indices = var_678_to_uint16, validate_indices = quantized_23_validate_indices_0, x = rest_vq_layers_10_embedding_to_fp16_palettized)[name = string("quantized_23_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 125, 256]> quantized_rest_sum_21_cast_fp16 = add(x = quantized_rest_sum_19_cast_fp16, y = quantized_23_cast_fp16_cast_uint16)[name = string("quantized_rest_sum_21_cast_fp16")];
            tensor<int32, [3]> var_700_begin_0 = const()[name = string("op_700_begin_0"), val = tensor<int32, [3]>([0, 11, 0])];
            tensor<int32, [3]> var_700_end_0 = const()[name = string("op_700_end_0"), val = tensor<int32, [3]>([1, 12, 125])];
            tensor<bool, [3]> var_700_end_mask_0 = const()[name = string("op_700_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_700_squeeze_mask_0 = const()[name = string("op_700_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_700 = slice_by_index(begin = var_700_begin_0, end = var_700_end_0, end_mask = var_700_end_mask_0, squeeze_mask = var_700_squeeze_mask_0, x = var_447)[name = string("op_700")];
            int32 quantized_25_axis_0 = const()[name = string("quantized_25_axis_0"), val = int32(0)];
            int32 quantized_25_batch_dims_0 = const()[name = string("quantized_25_batch_dims_0"), val = int32(0)];
            bool quantized_25_validate_indices_0 = const()[name = string("quantized_25_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_11_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6430912))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6955264))))[name = string("rest_vq_layers_11_embedding_to_fp16_palettized")];
            string var_700_to_uint16_dtype_0 = const()[name = string("op_700_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_700_to_uint16 = cast(dtype = var_700_to_uint16_dtype_0, x = var_700)[name = string("cast_4")];
            tensor<fp16, [1, 125, 256]> quantized_25_cast_fp16_cast_uint16 = gather(axis = quantized_25_axis_0, batch_dims = quantized_25_batch_dims_0, indices = var_700_to_uint16, validate_indices = quantized_25_validate_indices_0, x = rest_vq_layers_11_embedding_to_fp16_palettized)[name = string("quantized_25_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 125, 256]> quantized_rest_sum_23_cast_fp16 = add(x = quantized_rest_sum_21_cast_fp16, y = quantized_25_cast_fp16_cast_uint16)[name = string("quantized_rest_sum_23_cast_fp16")];
            tensor<int32, [3]> var_722_begin_0 = const()[name = string("op_722_begin_0"), val = tensor<int32, [3]>([0, 12, 0])];
            tensor<int32, [3]> var_722_end_0 = const()[name = string("op_722_end_0"), val = tensor<int32, [3]>([1, 13, 125])];
            tensor<bool, [3]> var_722_end_mask_0 = const()[name = string("op_722_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_722_squeeze_mask_0 = const()[name = string("op_722_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_722 = slice_by_index(begin = var_722_begin_0, end = var_722_end_0, end_mask = var_722_end_mask_0, squeeze_mask = var_722_squeeze_mask_0, x = var_447)[name = string("op_722")];
            int32 quantized_27_axis_0 = const()[name = string("quantized_27_axis_0"), val = int32(0)];
            int32 quantized_27_batch_dims_0 = const()[name = string("quantized_27_batch_dims_0"), val = int32(0)];
            bool quantized_27_validate_indices_0 = const()[name = string("quantized_27_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_12_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6955840))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7480192))))[name = string("rest_vq_layers_12_embedding_to_fp16_palettized")];
            string var_722_to_uint16_dtype_0 = const()[name = string("op_722_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_722_to_uint16 = cast(dtype = var_722_to_uint16_dtype_0, x = var_722)[name = string("cast_3")];
            tensor<fp16, [1, 125, 256]> quantized_27_cast_fp16_cast_uint16 = gather(axis = quantized_27_axis_0, batch_dims = quantized_27_batch_dims_0, indices = var_722_to_uint16, validate_indices = quantized_27_validate_indices_0, x = rest_vq_layers_12_embedding_to_fp16_palettized)[name = string("quantized_27_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 125, 256]> quantized_rest_sum_25_cast_fp16 = add(x = quantized_rest_sum_23_cast_fp16, y = quantized_27_cast_fp16_cast_uint16)[name = string("quantized_rest_sum_25_cast_fp16")];
            tensor<int32, [3]> var_744_begin_0 = const()[name = string("op_744_begin_0"), val = tensor<int32, [3]>([0, 13, 0])];
            tensor<int32, [3]> var_744_end_0 = const()[name = string("op_744_end_0"), val = tensor<int32, [3]>([1, 14, 125])];
            tensor<bool, [3]> var_744_end_mask_0 = const()[name = string("op_744_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_744_squeeze_mask_0 = const()[name = string("op_744_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_744 = slice_by_index(begin = var_744_begin_0, end = var_744_end_0, end_mask = var_744_end_mask_0, squeeze_mask = var_744_squeeze_mask_0, x = var_447)[name = string("op_744")];
            int32 quantized_29_axis_0 = const()[name = string("quantized_29_axis_0"), val = int32(0)];
            int32 quantized_29_batch_dims_0 = const()[name = string("quantized_29_batch_dims_0"), val = int32(0)];
            bool quantized_29_validate_indices_0 = const()[name = string("quantized_29_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_13_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7480768))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8005120))))[name = string("rest_vq_layers_13_embedding_to_fp16_palettized")];
            string var_744_to_uint16_dtype_0 = const()[name = string("op_744_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_744_to_uint16 = cast(dtype = var_744_to_uint16_dtype_0, x = var_744)[name = string("cast_2")];
            tensor<fp16, [1, 125, 256]> quantized_29_cast_fp16_cast_uint16 = gather(axis = quantized_29_axis_0, batch_dims = quantized_29_batch_dims_0, indices = var_744_to_uint16, validate_indices = quantized_29_validate_indices_0, x = rest_vq_layers_13_embedding_to_fp16_palettized)[name = string("quantized_29_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 125, 256]> quantized_rest_sum_cast_fp16 = add(x = quantized_rest_sum_25_cast_fp16, y = quantized_29_cast_fp16_cast_uint16)[name = string("quantized_rest_sum_cast_fp16")];
            tensor<int32, [3]> var_766_begin_0 = const()[name = string("op_766_begin_0"), val = tensor<int32, [3]>([0, 14, 0])];
            tensor<int32, [3]> var_766_end_0 = const()[name = string("op_766_end_0"), val = tensor<int32, [3]>([1, 15, 125])];
            tensor<bool, [3]> var_766_end_mask_0 = const()[name = string("op_766_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<bool, [3]> var_766_squeeze_mask_0 = const()[name = string("op_766_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<int32, [1, 125]> var_766 = slice_by_index(begin = var_766_begin_0, end = var_766_end_0, end_mask = var_766_end_mask_0, squeeze_mask = var_766_squeeze_mask_0, x = var_447)[name = string("op_766")];
            int32 quantized_axis_0 = const()[name = string("quantized_axis_0"), val = int32(0)];
            int32 quantized_batch_dims_0 = const()[name = string("quantized_batch_dims_0"), val = int32(0)];
            bool quantized_validate_indices_0 = const()[name = string("quantized_validate_indices_0"), val = bool(false)];
            tensor<fp16, [2048, 256]> rest_vq_layers_14_embedding_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [2048, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8005696))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8530048))))[name = string("rest_vq_layers_14_embedding_to_fp16_palettized")];
            string var_766_to_uint16_dtype_0 = const()[name = string("op_766_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1, 125]> var_766_to_uint16 = cast(dtype = var_766_to_uint16_dtype_0, x = var_766)[name = string("cast_1")];
            tensor<fp16, [1, 125, 256]> quantized_cast_fp16_cast_uint16 = gather(axis = quantized_axis_0, batch_dims = quantized_batch_dims_0, indices = var_766_to_uint16, validate_indices = quantized_validate_indices_0, x = rest_vq_layers_14_embedding_to_fp16_palettized)[name = string("quantized_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 125, 256]> input_35_cast_fp16 = add(x = quantized_rest_sum_cast_fp16, y = quantized_cast_fp16_cast_uint16)[name = string("input_35_cast_fp16")];
            string quantized_rest_pad_type_0 = const()[name = string("quantized_rest_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> quantized_rest_strides_0 = const()[name = string("quantized_rest_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> quantized_rest_pad_0 = const()[name = string("quantized_rest_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> quantized_rest_dilations_0 = const()[name = string("quantized_rest_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 quantized_rest_groups_0 = const()[name = string("quantized_rest_groups_0"), val = int32(1)];
            tensor<int32, [3]> transpose_0_perm_0 = const()[name = string("transpose_0_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 256, 1]> rest_output_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8530624))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8661760))))[name = string("rest_output_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 256, 125]> transpose_0_cast_fp16 = transpose(perm = transpose_0_perm_0, x = input_35_cast_fp16)[name = string("transpose_38")];
            tensor<fp16, [1, 512, 125]> quantized_rest_cast_fp16 = conv(dilations = quantized_rest_dilations_0, groups = quantized_rest_groups_0, pad = quantized_rest_pad_0, pad_type = quantized_rest_pad_type_0, strides = quantized_rest_strides_0, weight = rest_output_proj_weight_to_fp16_palettized, x = transpose_0_cast_fp16)[name = string("quantized_rest_cast_fp16")];
            tensor<fp16, [1, 512, 125]> hidden_state_1_cast_fp16 = add(x = quantized_first_cast_fp16, y = quantized_rest_cast_fp16)[name = string("hidden_state_1_cast_fp16")];
            tensor<int32, [6]> input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string input_37_mode_0 = const()[name = string("input_37_mode_0"), val = string("constant")];
            fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 127]> input_37_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_37_mode_0, pad = input_37_pad_0, x = hidden_state_1_cast_fp16)[name = string("input_37_cast_fp16")];
            string var_815_pad_type_0 = const()[name = string("op_815_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_815_strides_0 = const()[name = string("op_815_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_815_pad_0 = const()[name = string("op_815_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_815_dilations_0 = const()[name = string("op_815_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_815_groups_0 = const()[name = string("op_815_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 512, 3]> pre_conv_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8662336))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10235264))))[name = string("pre_conv_conv_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> pre_conv_conv_bias_to_fp16 = const()[name = string("pre_conv_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10235840)))];
            tensor<fp16, [1, 1024, 125]> var_815_cast_fp16 = conv(bias = pre_conv_conv_bias_to_fp16, dilations = var_815_dilations_0, groups = var_815_groups_0, pad = var_815_pad_0, pad_type = var_815_pad_type_0, strides = var_815_strides_0, weight = pre_conv_conv_weight_to_fp16_palettized, x = input_37_cast_fp16)[name = string("op_815_cast_fp16")];
            tensor<int32, [3]> input_39_perm_0 = const()[name = string("input_39_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 1024]> pre_transformer_input_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10237952))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10762304))))[name = string("pre_transformer_input_proj_weight_to_fp16_palettized")];
            tensor<fp16, [512]> pre_transformer_input_proj_bias_to_fp16 = const()[name = string("pre_transformer_input_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10762880)))];
            tensor<fp16, [1, 125, 1024]> input_39_cast_fp16 = transpose(perm = input_39_perm_0, x = var_815_cast_fp16)[name = string("transpose_37")];
            tensor<fp16, [1, 125, 512]> linear_0_cast_fp16 = linear(bias = pre_transformer_input_proj_bias_to_fp16, weight = pre_transformer_input_proj_weight_to_fp16_palettized, x = input_39_cast_fp16)[name = string("linear_0_cast_fp16")];
            fp16 var_926_promoted_to_fp16 = const()[name = string("op_926_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_932_cast_fp16 = pow(x = linear_0_cast_fp16, y = var_926_promoted_to_fp16)[name = string("op_932_cast_fp16")];
            tensor<int32, [1]> variance_1_axes_0 = const()[name = string("variance_1_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_1_keep_dims_0 = const()[name = string("variance_1_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_1_cast_fp16 = reduce_mean(axes = variance_1_axes_0, keep_dims = variance_1_keep_dims_0, x = var_932_cast_fp16)[name = string("variance_1_cast_fp16")];
            fp16 var_935_to_fp16 = const()[name = string("op_935_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_936_cast_fp16 = add(x = variance_1_cast_fp16, y = var_935_to_fp16)[name = string("op_936_cast_fp16")];
            fp32 var_937_epsilon_0 = const()[name = string("op_937_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_937_cast_fp16 = rsqrt(epsilon = var_937_epsilon_0, x = var_936_cast_fp16)[name = string("op_937_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_5_cast_fp16 = mul(x = linear_0_cast_fp16, y = var_937_cast_fp16)[name = string("hidden_states_5_cast_fp16")];
            tensor<fp16, [512]> const_12_to_fp16 = const()[name = string("const_12_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10763968)))];
            tensor<fp16, [1, 125, 512]> h_1_cast_fp16 = mul(x = const_12_to_fp16, y = hidden_states_5_cast_fp16)[name = string("h_1_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10765056))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11289408))))[name = string("pre_transformer_layers_0_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11289984)))];
            tensor<fp16, [1, 125, 1024]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = h_1_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<int32, [4]> var_963 = const()[name = string("op_963"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_964_cast_fp16 = reshape(shape = var_963, x = linear_1_cast_fp16)[name = string("op_964_cast_fp16")];
            tensor<int32, [4]> query_states_1_perm_0 = const()[name = string("query_states_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11292096))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11816448))))[name = string("pre_transformer_layers_0_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_2_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = h_1_cast_fp16)[name = string("linear_2_cast_fp16")];
            tensor<int32, [4]> var_973 = const()[name = string("op_973"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_974_cast_fp16 = reshape(shape = var_973, x = linear_2_cast_fp16)[name = string("op_974_cast_fp16")];
            tensor<int32, [4]> key_states_1_perm_0 = const()[name = string("key_states_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11817024))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12341376))))[name = string("pre_transformer_layers_0_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_3_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = h_1_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<int32, [4]> var_983 = const()[name = string("op_983"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_984_cast_fp16 = reshape(shape = var_983, x = linear_3_cast_fp16)[name = string("op_984_cast_fp16")];
            tensor<int32, [4]> value_states_1_perm_0 = const()[name = string("value_states_1_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 1, 125, 64]> op_989_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 1, 125, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12341952))), lut = tensor<fp16, [1, 1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12350016))))[name = string("op_989_to_fp16_palettized")];
            tensor<fp16, [1, 16, 125, 64]> query_states_1_cast_fp16 = transpose(perm = query_states_1_perm_0, x = var_964_cast_fp16)[name = string("transpose_36")];
            tensor<fp16, [1, 16, 125, 64]> var_990_cast_fp16 = mul(x = query_states_1_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_990_cast_fp16")];
            tensor<int32, [4]> x1_1_begin_0 = const()[name = string("x1_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_1_end_0 = const()[name = string("x1_1_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_1_end_mask_0 = const()[name = string("x1_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_1_cast_fp16 = slice_by_index(begin = x1_1_begin_0, end = x1_1_end_0, end_mask = x1_1_end_mask_0, x = query_states_1_cast_fp16)[name = string("x1_1_cast_fp16")];
            tensor<int32, [4]> x2_1_begin_0 = const()[name = string("x2_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_1_end_0 = const()[name = string("x2_1_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_1_end_mask_0 = const()[name = string("x2_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_1_cast_fp16 = slice_by_index(begin = x2_1_begin_0, end = x2_1_end_0, end_mask = x2_1_end_mask_0, x = query_states_1_cast_fp16)[name = string("x2_1_cast_fp16")];
            fp16 const_17_promoted_to_fp16 = const()[name = string("const_17_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_1011_cast_fp16 = mul(x = x2_1_cast_fp16, y = const_17_promoted_to_fp16)[name = string("op_1011_cast_fp16")];
            int32 var_1013 = const()[name = string("op_1013"), val = int32(-1)];
            bool var_1014_interleave_0 = const()[name = string("op_1014_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_1014_cast_fp16 = concat(axis = var_1013, interleave = var_1014_interleave_0, values = (var_1011_cast_fp16, x1_1_cast_fp16))[name = string("op_1014_cast_fp16")];
            tensor<fp16, [1, 1, 125, 64]> op_1016_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 1, 125, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12350592))), lut = tensor<fp16, [1, 1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12358656))))[name = string("op_1016_to_fp16_palettized")];
            tensor<fp16, [1, 16, 125, 64]> var_1017_cast_fp16 = mul(x = var_1014_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_1017_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> q_embed_1_cast_fp16 = add(x = var_990_cast_fp16, y = var_1017_cast_fp16)[name = string("q_embed_1_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> key_states_1_cast_fp16 = transpose(perm = key_states_1_perm_0, x = var_974_cast_fp16)[name = string("transpose_35")];
            tensor<fp16, [1, 16, 125, 64]> var_1022_cast_fp16 = mul(x = key_states_1_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_1022_cast_fp16")];
            tensor<int32, [4]> x1_3_begin_0 = const()[name = string("x1_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_3_end_0 = const()[name = string("x1_3_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_3_end_mask_0 = const()[name = string("x1_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_3_cast_fp16 = slice_by_index(begin = x1_3_begin_0, end = x1_3_end_0, end_mask = x1_3_end_mask_0, x = key_states_1_cast_fp16)[name = string("x1_3_cast_fp16")];
            tensor<int32, [4]> x2_3_begin_0 = const()[name = string("x2_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_3_end_0 = const()[name = string("x2_3_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_3_end_mask_0 = const()[name = string("x2_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_3_cast_fp16 = slice_by_index(begin = x2_3_begin_0, end = x2_3_end_0, end_mask = x2_3_end_mask_0, x = key_states_1_cast_fp16)[name = string("x2_3_cast_fp16")];
            fp16 const_20_promoted_to_fp16 = const()[name = string("const_20_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_1043_cast_fp16 = mul(x = x2_3_cast_fp16, y = const_20_promoted_to_fp16)[name = string("op_1043_cast_fp16")];
            int32 var_1045 = const()[name = string("op_1045"), val = int32(-1)];
            bool var_1046_interleave_0 = const()[name = string("op_1046_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_1046_cast_fp16 = concat(axis = var_1045, interleave = var_1046_interleave_0, values = (var_1043_cast_fp16, x1_3_cast_fp16))[name = string("op_1046_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_1049_cast_fp16 = mul(x = var_1046_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_1049_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> k_embed_1_cast_fp16 = add(x = var_1022_cast_fp16, y = var_1049_cast_fp16)[name = string("k_embed_1_cast_fp16")];
            bool var_1055_transpose_x_1 = const()[name = string("op_1055_transpose_x_1"), val = bool(false)];
            bool var_1055_transpose_y_1 = const()[name = string("op_1055_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 125, 125]> var_1055_cast_fp16 = matmul(transpose_x = var_1055_transpose_x_1, transpose_y = var_1055_transpose_y_1, x = q_embed_1_cast_fp16, y = k_embed_1_cast_fp16)[name = string("op_1055_cast_fp16")];
            fp16 var_1056_to_fp16 = const()[name = string("op_1056_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_1_cast_fp16 = mul(x = var_1055_cast_fp16, y = var_1056_to_fp16)[name = string("attn_weights_1_cast_fp16")];
            tensor<fp16, [1, 1, 125, 125]> op_1070_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 1, 125, 125]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12359232))), lut = tensor<fp16, [1, 1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12374976))))[name = string("op_1070_to_fp16_palettized")];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_3_cast_fp16 = add(x = attn_weights_1_cast_fp16, y = op_1070_to_fp16_palettized)[name = string("attn_weights_3_cast_fp16")];
            int32 var_1073 = const()[name = string("op_1073"), val = int32(-1)];
            tensor<fp16, [1, 16, 125, 125]> var_1075_cast_fp16 = softmax(axis = var_1073, x = attn_weights_3_cast_fp16)[name = string("op_1075_cast_fp16")];
            bool attn_output_1_transpose_x_0 = const()[name = string("attn_output_1_transpose_x_0"), val = bool(false)];
            bool attn_output_1_transpose_y_0 = const()[name = string("attn_output_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> value_states_1_cast_fp16 = transpose(perm = value_states_1_perm_0, x = var_984_cast_fp16)[name = string("transpose_34")];
            tensor<fp16, [1, 16, 125, 64]> attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = var_1075_cast_fp16, y = value_states_1_cast_fp16)[name = string("attn_output_1_cast_fp16")];
            tensor<int32, [4]> var_1084_perm_0 = const()[name = string("op_1084_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1086 = const()[name = string("op_1086"), val = tensor<int32, [3]>([1, 125, -1])];
            tensor<fp16, [1, 125, 16, 64]> var_1084_cast_fp16 = transpose(perm = var_1084_perm_0, x = attn_output_1_cast_fp16)[name = string("transpose_33")];
            tensor<fp16, [1, 125, 1024]> var_1087_cast_fp16 = reshape(shape = var_1086, x = var_1084_cast_fp16)[name = string("op_1087_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_0_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12375552))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12899904))))[name = string("pre_transformer_layers_0_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [512]> linear_4_bias_0_to_fp16 = const()[name = string("linear_4_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12900480)))];
            tensor<fp16, [1, 125, 512]> linear_4_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_0_self_attn_o_proj_weight_to_fp16_palettized, x = var_1087_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_0_self_attn_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_0_self_attn_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12901568)))];
            tensor<fp16, [1, 125, 512]> var_1094_cast_fp16 = mul(x = pre_transformer_layers_0_self_attn_layer_scale_scale_to_fp16, y = linear_4_cast_fp16)[name = string("op_1094_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_7_cast_fp16 = add(x = linear_0_cast_fp16, y = var_1094_cast_fp16)[name = string("hidden_states_7_cast_fp16")];
            fp16 var_1100_promoted_to_fp16 = const()[name = string("op_1100_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_1106_cast_fp16 = pow(x = hidden_states_7_cast_fp16, y = var_1100_promoted_to_fp16)[name = string("op_1106_cast_fp16")];
            tensor<int32, [1]> variance_3_axes_0 = const()[name = string("variance_3_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_3_keep_dims_0 = const()[name = string("variance_3_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_3_cast_fp16 = reduce_mean(axes = variance_3_axes_0, keep_dims = variance_3_keep_dims_0, x = var_1106_cast_fp16)[name = string("variance_3_cast_fp16")];
            fp16 var_1109_to_fp16 = const()[name = string("op_1109_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_1110_cast_fp16 = add(x = variance_3_cast_fp16, y = var_1109_to_fp16)[name = string("op_1110_cast_fp16")];
            fp32 var_1111_epsilon_0 = const()[name = string("op_1111_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_1111_cast_fp16 = rsqrt(epsilon = var_1111_epsilon_0, x = var_1110_cast_fp16)[name = string("op_1111_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_11_cast_fp16 = mul(x = hidden_states_7_cast_fp16, y = var_1111_cast_fp16)[name = string("hidden_states_11_cast_fp16")];
            tensor<fp16, [512]> const_21_to_fp16 = const()[name = string("const_21_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12902656)))];
            tensor<fp16, [1, 125, 512]> input_43_cast_fp16 = mul(x = const_21_to_fp16, y = hidden_states_11_cast_fp16)[name = string("input_43_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_0_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12903744))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13428096))))[name = string("pre_transformer_layers_0_mlp_gate_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_5_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_0_mlp_gate_proj_weight_to_fp16_palettized, x = input_43_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> var_1124_cast_fp16 = silu(x = linear_5_cast_fp16)[name = string("op_1124_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_0_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13428672))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13953024))))[name = string("pre_transformer_layers_0_mlp_up_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_6_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_0_mlp_up_proj_weight_to_fp16_palettized, x = input_43_cast_fp16)[name = string("linear_6_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> input_47_cast_fp16 = mul(x = var_1124_cast_fp16, y = linear_6_cast_fp16)[name = string("input_47_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_0_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13953600))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14477952))))[name = string("pre_transformer_layers_0_mlp_down_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_7_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_0_mlp_down_proj_weight_to_fp16_palettized, x = input_47_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_0_mlp_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_0_mlp_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14478528)))];
            tensor<fp16, [1, 125, 512]> var_1131_cast_fp16 = mul(x = pre_transformer_layers_0_mlp_layer_scale_scale_to_fp16, y = linear_7_cast_fp16)[name = string("op_1131_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_13_cast_fp16 = add(x = hidden_states_7_cast_fp16, y = var_1131_cast_fp16)[name = string("hidden_states_13_cast_fp16")];
            fp16 var_1137_promoted_to_fp16 = const()[name = string("op_1137_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_1143_cast_fp16 = pow(x = hidden_states_13_cast_fp16, y = var_1137_promoted_to_fp16)[name = string("op_1143_cast_fp16")];
            tensor<int32, [1]> variance_5_axes_0 = const()[name = string("variance_5_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_5_keep_dims_0 = const()[name = string("variance_5_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_5_cast_fp16 = reduce_mean(axes = variance_5_axes_0, keep_dims = variance_5_keep_dims_0, x = var_1143_cast_fp16)[name = string("variance_5_cast_fp16")];
            fp16 var_1146_to_fp16 = const()[name = string("op_1146_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_1147_cast_fp16 = add(x = variance_5_cast_fp16, y = var_1146_to_fp16)[name = string("op_1147_cast_fp16")];
            fp32 var_1148_epsilon_0 = const()[name = string("op_1148_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_1148_cast_fp16 = rsqrt(epsilon = var_1148_epsilon_0, x = var_1147_cast_fp16)[name = string("op_1148_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_17_cast_fp16 = mul(x = hidden_states_13_cast_fp16, y = var_1148_cast_fp16)[name = string("hidden_states_17_cast_fp16")];
            tensor<fp16, [512]> const_22_to_fp16 = const()[name = string("const_22_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14479616)))];
            tensor<fp16, [1, 125, 512]> h_3_cast_fp16 = mul(x = const_22_to_fp16, y = hidden_states_17_cast_fp16)[name = string("h_3_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14480704))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15005056))))[name = string("pre_transformer_layers_1_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_8_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = h_3_cast_fp16)[name = string("linear_8_cast_fp16")];
            tensor<int32, [4]> var_1174 = const()[name = string("op_1174"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_1175_cast_fp16 = reshape(shape = var_1174, x = linear_8_cast_fp16)[name = string("op_1175_cast_fp16")];
            tensor<int32, [4]> query_states_3_perm_0 = const()[name = string("query_states_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15005632))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15529984))))[name = string("pre_transformer_layers_1_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_9_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = h_3_cast_fp16)[name = string("linear_9_cast_fp16")];
            tensor<int32, [4]> var_1184 = const()[name = string("op_1184"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_1185_cast_fp16 = reshape(shape = var_1184, x = linear_9_cast_fp16)[name = string("op_1185_cast_fp16")];
            tensor<int32, [4]> key_states_3_perm_0 = const()[name = string("key_states_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15530560))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16054912))))[name = string("pre_transformer_layers_1_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_10_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = h_3_cast_fp16)[name = string("linear_10_cast_fp16")];
            tensor<int32, [4]> var_1194 = const()[name = string("op_1194"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_1195_cast_fp16 = reshape(shape = var_1194, x = linear_10_cast_fp16)[name = string("op_1195_cast_fp16")];
            tensor<int32, [4]> value_states_3_perm_0 = const()[name = string("value_states_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 125, 64]> query_states_3_cast_fp16 = transpose(perm = query_states_3_perm_0, x = var_1175_cast_fp16)[name = string("transpose_32")];
            tensor<fp16, [1, 16, 125, 64]> var_1201_cast_fp16 = mul(x = query_states_3_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_1201_cast_fp16")];
            tensor<int32, [4]> x1_5_begin_0 = const()[name = string("x1_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_5_end_0 = const()[name = string("x1_5_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_5_end_mask_0 = const()[name = string("x1_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_5_cast_fp16 = slice_by_index(begin = x1_5_begin_0, end = x1_5_end_0, end_mask = x1_5_end_mask_0, x = query_states_3_cast_fp16)[name = string("x1_5_cast_fp16")];
            tensor<int32, [4]> x2_5_begin_0 = const()[name = string("x2_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_5_end_0 = const()[name = string("x2_5_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_5_end_mask_0 = const()[name = string("x2_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_5_cast_fp16 = slice_by_index(begin = x2_5_begin_0, end = x2_5_end_0, end_mask = x2_5_end_mask_0, x = query_states_3_cast_fp16)[name = string("x2_5_cast_fp16")];
            fp16 const_27_promoted_to_fp16 = const()[name = string("const_27_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_1222_cast_fp16 = mul(x = x2_5_cast_fp16, y = const_27_promoted_to_fp16)[name = string("op_1222_cast_fp16")];
            int32 var_1224 = const()[name = string("op_1224"), val = int32(-1)];
            bool var_1225_interleave_0 = const()[name = string("op_1225_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_1225_cast_fp16 = concat(axis = var_1224, interleave = var_1225_interleave_0, values = (var_1222_cast_fp16, x1_5_cast_fp16))[name = string("op_1225_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_1228_cast_fp16 = mul(x = var_1225_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_1228_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> q_embed_3_cast_fp16 = add(x = var_1201_cast_fp16, y = var_1228_cast_fp16)[name = string("q_embed_3_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> key_states_3_cast_fp16 = transpose(perm = key_states_3_perm_0, x = var_1185_cast_fp16)[name = string("transpose_31")];
            tensor<fp16, [1, 16, 125, 64]> var_1233_cast_fp16 = mul(x = key_states_3_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_1233_cast_fp16")];
            tensor<int32, [4]> x1_7_begin_0 = const()[name = string("x1_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_7_end_0 = const()[name = string("x1_7_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_7_end_mask_0 = const()[name = string("x1_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_7_cast_fp16 = slice_by_index(begin = x1_7_begin_0, end = x1_7_end_0, end_mask = x1_7_end_mask_0, x = key_states_3_cast_fp16)[name = string("x1_7_cast_fp16")];
            tensor<int32, [4]> x2_7_begin_0 = const()[name = string("x2_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_7_end_0 = const()[name = string("x2_7_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_7_end_mask_0 = const()[name = string("x2_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_7_cast_fp16 = slice_by_index(begin = x2_7_begin_0, end = x2_7_end_0, end_mask = x2_7_end_mask_0, x = key_states_3_cast_fp16)[name = string("x2_7_cast_fp16")];
            fp16 const_30_promoted_to_fp16 = const()[name = string("const_30_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_1254_cast_fp16 = mul(x = x2_7_cast_fp16, y = const_30_promoted_to_fp16)[name = string("op_1254_cast_fp16")];
            int32 var_1256 = const()[name = string("op_1256"), val = int32(-1)];
            bool var_1257_interleave_0 = const()[name = string("op_1257_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_1257_cast_fp16 = concat(axis = var_1256, interleave = var_1257_interleave_0, values = (var_1254_cast_fp16, x1_7_cast_fp16))[name = string("op_1257_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_1260_cast_fp16 = mul(x = var_1257_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_1260_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> k_embed_3_cast_fp16 = add(x = var_1233_cast_fp16, y = var_1260_cast_fp16)[name = string("k_embed_3_cast_fp16")];
            bool var_1266_transpose_x_1 = const()[name = string("op_1266_transpose_x_1"), val = bool(false)];
            bool var_1266_transpose_y_1 = const()[name = string("op_1266_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 125, 125]> var_1266_cast_fp16 = matmul(transpose_x = var_1266_transpose_x_1, transpose_y = var_1266_transpose_y_1, x = q_embed_3_cast_fp16, y = k_embed_3_cast_fp16)[name = string("op_1266_cast_fp16")];
            fp16 var_1267_to_fp16 = const()[name = string("op_1267_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_7_cast_fp16 = mul(x = var_1266_cast_fp16, y = var_1267_to_fp16)[name = string("attn_weights_7_cast_fp16")];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_9_cast_fp16 = add(x = attn_weights_7_cast_fp16, y = op_1070_to_fp16_palettized)[name = string("attn_weights_9_cast_fp16")];
            int32 var_1284 = const()[name = string("op_1284"), val = int32(-1)];
            tensor<fp16, [1, 16, 125, 125]> var_1286_cast_fp16 = softmax(axis = var_1284, x = attn_weights_9_cast_fp16)[name = string("op_1286_cast_fp16")];
            bool attn_output_3_transpose_x_0 = const()[name = string("attn_output_3_transpose_x_0"), val = bool(false)];
            bool attn_output_3_transpose_y_0 = const()[name = string("attn_output_3_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> value_states_3_cast_fp16 = transpose(perm = value_states_3_perm_0, x = var_1195_cast_fp16)[name = string("transpose_30")];
            tensor<fp16, [1, 16, 125, 64]> attn_output_3_cast_fp16 = matmul(transpose_x = attn_output_3_transpose_x_0, transpose_y = attn_output_3_transpose_y_0, x = var_1286_cast_fp16, y = value_states_3_cast_fp16)[name = string("attn_output_3_cast_fp16")];
            tensor<int32, [4]> var_1295_perm_0 = const()[name = string("op_1295_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1297 = const()[name = string("op_1297"), val = tensor<int32, [3]>([1, 125, -1])];
            tensor<fp16, [1, 125, 16, 64]> var_1295_cast_fp16 = transpose(perm = var_1295_perm_0, x = attn_output_3_cast_fp16)[name = string("transpose_29")];
            tensor<fp16, [1, 125, 1024]> var_1298_cast_fp16 = reshape(shape = var_1297, x = var_1295_cast_fp16)[name = string("op_1298_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_1_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16055488))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16579840))))[name = string("pre_transformer_layers_1_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_11_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_1_self_attn_o_proj_weight_to_fp16_palettized, x = var_1298_cast_fp16)[name = string("linear_11_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_1_self_attn_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_1_self_attn_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16580416)))];
            tensor<fp16, [1, 125, 512]> var_1305_cast_fp16 = mul(x = pre_transformer_layers_1_self_attn_layer_scale_scale_to_fp16, y = linear_11_cast_fp16)[name = string("op_1305_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_19_cast_fp16 = add(x = hidden_states_13_cast_fp16, y = var_1305_cast_fp16)[name = string("hidden_states_19_cast_fp16")];
            fp16 var_1311_promoted_to_fp16 = const()[name = string("op_1311_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_1317_cast_fp16 = pow(x = hidden_states_19_cast_fp16, y = var_1311_promoted_to_fp16)[name = string("op_1317_cast_fp16")];
            tensor<int32, [1]> variance_7_axes_0 = const()[name = string("variance_7_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_7_keep_dims_0 = const()[name = string("variance_7_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_7_cast_fp16 = reduce_mean(axes = variance_7_axes_0, keep_dims = variance_7_keep_dims_0, x = var_1317_cast_fp16)[name = string("variance_7_cast_fp16")];
            fp16 var_1320_to_fp16 = const()[name = string("op_1320_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_1321_cast_fp16 = add(x = variance_7_cast_fp16, y = var_1320_to_fp16)[name = string("op_1321_cast_fp16")];
            fp32 var_1322_epsilon_0 = const()[name = string("op_1322_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_1322_cast_fp16 = rsqrt(epsilon = var_1322_epsilon_0, x = var_1321_cast_fp16)[name = string("op_1322_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_23_cast_fp16 = mul(x = hidden_states_19_cast_fp16, y = var_1322_cast_fp16)[name = string("hidden_states_23_cast_fp16")];
            tensor<fp16, [512]> const_31_to_fp16 = const()[name = string("const_31_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16581504)))];
            tensor<fp16, [1, 125, 512]> input_51_cast_fp16 = mul(x = const_31_to_fp16, y = hidden_states_23_cast_fp16)[name = string("input_51_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_1_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16582592))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17106944))))[name = string("pre_transformer_layers_1_mlp_gate_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_12_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_1_mlp_gate_proj_weight_to_fp16_palettized, x = input_51_cast_fp16)[name = string("linear_12_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> var_1335_cast_fp16 = silu(x = linear_12_cast_fp16)[name = string("op_1335_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_1_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17107520))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17631872))))[name = string("pre_transformer_layers_1_mlp_up_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_1_mlp_up_proj_weight_to_fp16_palettized, x = input_51_cast_fp16)[name = string("linear_13_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> input_55_cast_fp16 = mul(x = var_1335_cast_fp16, y = linear_13_cast_fp16)[name = string("input_55_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_1_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17632448))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18156800))))[name = string("pre_transformer_layers_1_mlp_down_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_14_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_1_mlp_down_proj_weight_to_fp16_palettized, x = input_55_cast_fp16)[name = string("linear_14_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_1_mlp_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_1_mlp_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18157376)))];
            tensor<fp16, [1, 125, 512]> var_1342_cast_fp16 = mul(x = pre_transformer_layers_1_mlp_layer_scale_scale_to_fp16, y = linear_14_cast_fp16)[name = string("op_1342_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_25_cast_fp16 = add(x = hidden_states_19_cast_fp16, y = var_1342_cast_fp16)[name = string("hidden_states_25_cast_fp16")];
            fp16 var_1348_promoted_to_fp16 = const()[name = string("op_1348_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_1354_cast_fp16 = pow(x = hidden_states_25_cast_fp16, y = var_1348_promoted_to_fp16)[name = string("op_1354_cast_fp16")];
            tensor<int32, [1]> variance_9_axes_0 = const()[name = string("variance_9_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_9_keep_dims_0 = const()[name = string("variance_9_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_9_cast_fp16 = reduce_mean(axes = variance_9_axes_0, keep_dims = variance_9_keep_dims_0, x = var_1354_cast_fp16)[name = string("variance_9_cast_fp16")];
            fp16 var_1357_to_fp16 = const()[name = string("op_1357_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_1358_cast_fp16 = add(x = variance_9_cast_fp16, y = var_1357_to_fp16)[name = string("op_1358_cast_fp16")];
            fp32 var_1359_epsilon_0 = const()[name = string("op_1359_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_1359_cast_fp16 = rsqrt(epsilon = var_1359_epsilon_0, x = var_1358_cast_fp16)[name = string("op_1359_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_29_cast_fp16 = mul(x = hidden_states_25_cast_fp16, y = var_1359_cast_fp16)[name = string("hidden_states_29_cast_fp16")];
            tensor<fp16, [512]> const_32_to_fp16 = const()[name = string("const_32_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18158464)))];
            tensor<fp16, [1, 125, 512]> h_5_cast_fp16 = mul(x = const_32_to_fp16, y = hidden_states_29_cast_fp16)[name = string("h_5_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18159552))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18683904))))[name = string("pre_transformer_layers_2_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_15_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = h_5_cast_fp16)[name = string("linear_15_cast_fp16")];
            tensor<int32, [4]> var_1385 = const()[name = string("op_1385"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_1386_cast_fp16 = reshape(shape = var_1385, x = linear_15_cast_fp16)[name = string("op_1386_cast_fp16")];
            tensor<int32, [4]> query_states_5_perm_0 = const()[name = string("query_states_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18684480))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19208832))))[name = string("pre_transformer_layers_2_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_16_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = h_5_cast_fp16)[name = string("linear_16_cast_fp16")];
            tensor<int32, [4]> var_1395 = const()[name = string("op_1395"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_1396_cast_fp16 = reshape(shape = var_1395, x = linear_16_cast_fp16)[name = string("op_1396_cast_fp16")];
            tensor<int32, [4]> key_states_5_perm_0 = const()[name = string("key_states_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19209408))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19733760))))[name = string("pre_transformer_layers_2_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_17_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = h_5_cast_fp16)[name = string("linear_17_cast_fp16")];
            tensor<int32, [4]> var_1405 = const()[name = string("op_1405"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_1406_cast_fp16 = reshape(shape = var_1405, x = linear_17_cast_fp16)[name = string("op_1406_cast_fp16")];
            tensor<int32, [4]> value_states_5_perm_0 = const()[name = string("value_states_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 125, 64]> query_states_5_cast_fp16 = transpose(perm = query_states_5_perm_0, x = var_1386_cast_fp16)[name = string("transpose_28")];
            tensor<fp16, [1, 16, 125, 64]> var_1412_cast_fp16 = mul(x = query_states_5_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_1412_cast_fp16")];
            tensor<int32, [4]> x1_9_begin_0 = const()[name = string("x1_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_9_end_0 = const()[name = string("x1_9_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_9_end_mask_0 = const()[name = string("x1_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_9_cast_fp16 = slice_by_index(begin = x1_9_begin_0, end = x1_9_end_0, end_mask = x1_9_end_mask_0, x = query_states_5_cast_fp16)[name = string("x1_9_cast_fp16")];
            tensor<int32, [4]> x2_9_begin_0 = const()[name = string("x2_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_9_end_0 = const()[name = string("x2_9_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_9_end_mask_0 = const()[name = string("x2_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_9_cast_fp16 = slice_by_index(begin = x2_9_begin_0, end = x2_9_end_0, end_mask = x2_9_end_mask_0, x = query_states_5_cast_fp16)[name = string("x2_9_cast_fp16")];
            fp16 const_37_promoted_to_fp16 = const()[name = string("const_37_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_1433_cast_fp16 = mul(x = x2_9_cast_fp16, y = const_37_promoted_to_fp16)[name = string("op_1433_cast_fp16")];
            int32 var_1435 = const()[name = string("op_1435"), val = int32(-1)];
            bool var_1436_interleave_0 = const()[name = string("op_1436_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_1436_cast_fp16 = concat(axis = var_1435, interleave = var_1436_interleave_0, values = (var_1433_cast_fp16, x1_9_cast_fp16))[name = string("op_1436_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_1439_cast_fp16 = mul(x = var_1436_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_1439_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> q_embed_5_cast_fp16 = add(x = var_1412_cast_fp16, y = var_1439_cast_fp16)[name = string("q_embed_5_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> key_states_5_cast_fp16 = transpose(perm = key_states_5_perm_0, x = var_1396_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [1, 16, 125, 64]> var_1444_cast_fp16 = mul(x = key_states_5_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_1444_cast_fp16")];
            tensor<int32, [4]> x1_11_begin_0 = const()[name = string("x1_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_11_end_0 = const()[name = string("x1_11_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_11_end_mask_0 = const()[name = string("x1_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_11_cast_fp16 = slice_by_index(begin = x1_11_begin_0, end = x1_11_end_0, end_mask = x1_11_end_mask_0, x = key_states_5_cast_fp16)[name = string("x1_11_cast_fp16")];
            tensor<int32, [4]> x2_11_begin_0 = const()[name = string("x2_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_11_end_0 = const()[name = string("x2_11_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_11_end_mask_0 = const()[name = string("x2_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_11_cast_fp16 = slice_by_index(begin = x2_11_begin_0, end = x2_11_end_0, end_mask = x2_11_end_mask_0, x = key_states_5_cast_fp16)[name = string("x2_11_cast_fp16")];
            fp16 const_40_promoted_to_fp16 = const()[name = string("const_40_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_1465_cast_fp16 = mul(x = x2_11_cast_fp16, y = const_40_promoted_to_fp16)[name = string("op_1465_cast_fp16")];
            int32 var_1467 = const()[name = string("op_1467"), val = int32(-1)];
            bool var_1468_interleave_0 = const()[name = string("op_1468_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_1468_cast_fp16 = concat(axis = var_1467, interleave = var_1468_interleave_0, values = (var_1465_cast_fp16, x1_11_cast_fp16))[name = string("op_1468_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_1471_cast_fp16 = mul(x = var_1468_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_1471_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> k_embed_5_cast_fp16 = add(x = var_1444_cast_fp16, y = var_1471_cast_fp16)[name = string("k_embed_5_cast_fp16")];
            bool var_1477_transpose_x_1 = const()[name = string("op_1477_transpose_x_1"), val = bool(false)];
            bool var_1477_transpose_y_1 = const()[name = string("op_1477_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 125, 125]> var_1477_cast_fp16 = matmul(transpose_x = var_1477_transpose_x_1, transpose_y = var_1477_transpose_y_1, x = q_embed_5_cast_fp16, y = k_embed_5_cast_fp16)[name = string("op_1477_cast_fp16")];
            fp16 var_1478_to_fp16 = const()[name = string("op_1478_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_13_cast_fp16 = mul(x = var_1477_cast_fp16, y = var_1478_to_fp16)[name = string("attn_weights_13_cast_fp16")];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_15_cast_fp16 = add(x = attn_weights_13_cast_fp16, y = op_1070_to_fp16_palettized)[name = string("attn_weights_15_cast_fp16")];
            int32 var_1495 = const()[name = string("op_1495"), val = int32(-1)];
            tensor<fp16, [1, 16, 125, 125]> var_1497_cast_fp16 = softmax(axis = var_1495, x = attn_weights_15_cast_fp16)[name = string("op_1497_cast_fp16")];
            bool attn_output_5_transpose_x_0 = const()[name = string("attn_output_5_transpose_x_0"), val = bool(false)];
            bool attn_output_5_transpose_y_0 = const()[name = string("attn_output_5_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> value_states_5_cast_fp16 = transpose(perm = value_states_5_perm_0, x = var_1406_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [1, 16, 125, 64]> attn_output_5_cast_fp16 = matmul(transpose_x = attn_output_5_transpose_x_0, transpose_y = attn_output_5_transpose_y_0, x = var_1497_cast_fp16, y = value_states_5_cast_fp16)[name = string("attn_output_5_cast_fp16")];
            tensor<int32, [4]> var_1506_perm_0 = const()[name = string("op_1506_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1508 = const()[name = string("op_1508"), val = tensor<int32, [3]>([1, 125, -1])];
            tensor<fp16, [1, 125, 16, 64]> var_1506_cast_fp16 = transpose(perm = var_1506_perm_0, x = attn_output_5_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [1, 125, 1024]> var_1509_cast_fp16 = reshape(shape = var_1508, x = var_1506_cast_fp16)[name = string("op_1509_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_2_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19734336))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20258688))))[name = string("pre_transformer_layers_2_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_18_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_2_self_attn_o_proj_weight_to_fp16_palettized, x = var_1509_cast_fp16)[name = string("linear_18_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_2_self_attn_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_2_self_attn_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20259264)))];
            tensor<fp16, [1, 125, 512]> var_1516_cast_fp16 = mul(x = pre_transformer_layers_2_self_attn_layer_scale_scale_to_fp16, y = linear_18_cast_fp16)[name = string("op_1516_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_31_cast_fp16 = add(x = hidden_states_25_cast_fp16, y = var_1516_cast_fp16)[name = string("hidden_states_31_cast_fp16")];
            fp16 var_1522_promoted_to_fp16 = const()[name = string("op_1522_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_1528_cast_fp16 = pow(x = hidden_states_31_cast_fp16, y = var_1522_promoted_to_fp16)[name = string("op_1528_cast_fp16")];
            tensor<int32, [1]> variance_11_axes_0 = const()[name = string("variance_11_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_11_keep_dims_0 = const()[name = string("variance_11_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_11_cast_fp16 = reduce_mean(axes = variance_11_axes_0, keep_dims = variance_11_keep_dims_0, x = var_1528_cast_fp16)[name = string("variance_11_cast_fp16")];
            fp16 var_1531_to_fp16 = const()[name = string("op_1531_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_1532_cast_fp16 = add(x = variance_11_cast_fp16, y = var_1531_to_fp16)[name = string("op_1532_cast_fp16")];
            fp32 var_1533_epsilon_0 = const()[name = string("op_1533_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_1533_cast_fp16 = rsqrt(epsilon = var_1533_epsilon_0, x = var_1532_cast_fp16)[name = string("op_1533_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_35_cast_fp16 = mul(x = hidden_states_31_cast_fp16, y = var_1533_cast_fp16)[name = string("hidden_states_35_cast_fp16")];
            tensor<fp16, [512]> const_41_to_fp16 = const()[name = string("const_41_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20260352)))];
            tensor<fp16, [1, 125, 512]> input_59_cast_fp16 = mul(x = const_41_to_fp16, y = hidden_states_35_cast_fp16)[name = string("input_59_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_2_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20261440))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20785792))))[name = string("pre_transformer_layers_2_mlp_gate_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_2_mlp_gate_proj_weight_to_fp16_palettized, x = input_59_cast_fp16)[name = string("linear_19_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> var_1546_cast_fp16 = silu(x = linear_19_cast_fp16)[name = string("op_1546_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_2_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20786368))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21310720))))[name = string("pre_transformer_layers_2_mlp_up_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_20_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_2_mlp_up_proj_weight_to_fp16_palettized, x = input_59_cast_fp16)[name = string("linear_20_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> input_63_cast_fp16 = mul(x = var_1546_cast_fp16, y = linear_20_cast_fp16)[name = string("input_63_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_2_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21311296))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21835648))))[name = string("pre_transformer_layers_2_mlp_down_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_21_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_2_mlp_down_proj_weight_to_fp16_palettized, x = input_63_cast_fp16)[name = string("linear_21_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_2_mlp_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_2_mlp_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21836224)))];
            tensor<fp16, [1, 125, 512]> var_1553_cast_fp16 = mul(x = pre_transformer_layers_2_mlp_layer_scale_scale_to_fp16, y = linear_21_cast_fp16)[name = string("op_1553_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_37_cast_fp16 = add(x = hidden_states_31_cast_fp16, y = var_1553_cast_fp16)[name = string("hidden_states_37_cast_fp16")];
            fp16 var_1559_promoted_to_fp16 = const()[name = string("op_1559_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_1565_cast_fp16 = pow(x = hidden_states_37_cast_fp16, y = var_1559_promoted_to_fp16)[name = string("op_1565_cast_fp16")];
            tensor<int32, [1]> variance_13_axes_0 = const()[name = string("variance_13_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_13_keep_dims_0 = const()[name = string("variance_13_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_13_cast_fp16 = reduce_mean(axes = variance_13_axes_0, keep_dims = variance_13_keep_dims_0, x = var_1565_cast_fp16)[name = string("variance_13_cast_fp16")];
            fp16 var_1568_to_fp16 = const()[name = string("op_1568_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_1569_cast_fp16 = add(x = variance_13_cast_fp16, y = var_1568_to_fp16)[name = string("op_1569_cast_fp16")];
            fp32 var_1570_epsilon_0 = const()[name = string("op_1570_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_1570_cast_fp16 = rsqrt(epsilon = var_1570_epsilon_0, x = var_1569_cast_fp16)[name = string("op_1570_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_41_cast_fp16 = mul(x = hidden_states_37_cast_fp16, y = var_1570_cast_fp16)[name = string("hidden_states_41_cast_fp16")];
            tensor<fp16, [512]> const_42_to_fp16 = const()[name = string("const_42_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21837312)))];
            tensor<fp16, [1, 125, 512]> h_7_cast_fp16 = mul(x = const_42_to_fp16, y = hidden_states_41_cast_fp16)[name = string("h_7_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21838400))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22362752))))[name = string("pre_transformer_layers_3_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_22_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = h_7_cast_fp16)[name = string("linear_22_cast_fp16")];
            tensor<int32, [4]> var_1596 = const()[name = string("op_1596"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_1597_cast_fp16 = reshape(shape = var_1596, x = linear_22_cast_fp16)[name = string("op_1597_cast_fp16")];
            tensor<int32, [4]> query_states_7_perm_0 = const()[name = string("query_states_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22363328))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22887680))))[name = string("pre_transformer_layers_3_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_23_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = h_7_cast_fp16)[name = string("linear_23_cast_fp16")];
            tensor<int32, [4]> var_1606 = const()[name = string("op_1606"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_1607_cast_fp16 = reshape(shape = var_1606, x = linear_23_cast_fp16)[name = string("op_1607_cast_fp16")];
            tensor<int32, [4]> key_states_7_perm_0 = const()[name = string("key_states_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22888256))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23412608))))[name = string("pre_transformer_layers_3_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_24_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = h_7_cast_fp16)[name = string("linear_24_cast_fp16")];
            tensor<int32, [4]> var_1616 = const()[name = string("op_1616"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_1617_cast_fp16 = reshape(shape = var_1616, x = linear_24_cast_fp16)[name = string("op_1617_cast_fp16")];
            tensor<int32, [4]> value_states_7_perm_0 = const()[name = string("value_states_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 125, 64]> query_states_7_cast_fp16 = transpose(perm = query_states_7_perm_0, x = var_1597_cast_fp16)[name = string("transpose_24")];
            tensor<fp16, [1, 16, 125, 64]> var_1623_cast_fp16 = mul(x = query_states_7_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_1623_cast_fp16")];
            tensor<int32, [4]> x1_13_begin_0 = const()[name = string("x1_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_13_end_0 = const()[name = string("x1_13_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_13_end_mask_0 = const()[name = string("x1_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_13_cast_fp16 = slice_by_index(begin = x1_13_begin_0, end = x1_13_end_0, end_mask = x1_13_end_mask_0, x = query_states_7_cast_fp16)[name = string("x1_13_cast_fp16")];
            tensor<int32, [4]> x2_13_begin_0 = const()[name = string("x2_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_13_end_0 = const()[name = string("x2_13_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_13_end_mask_0 = const()[name = string("x2_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_13_cast_fp16 = slice_by_index(begin = x2_13_begin_0, end = x2_13_end_0, end_mask = x2_13_end_mask_0, x = query_states_7_cast_fp16)[name = string("x2_13_cast_fp16")];
            fp16 const_47_promoted_to_fp16 = const()[name = string("const_47_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_1644_cast_fp16 = mul(x = x2_13_cast_fp16, y = const_47_promoted_to_fp16)[name = string("op_1644_cast_fp16")];
            int32 var_1646 = const()[name = string("op_1646"), val = int32(-1)];
            bool var_1647_interleave_0 = const()[name = string("op_1647_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_1647_cast_fp16 = concat(axis = var_1646, interleave = var_1647_interleave_0, values = (var_1644_cast_fp16, x1_13_cast_fp16))[name = string("op_1647_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_1650_cast_fp16 = mul(x = var_1647_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_1650_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> q_embed_7_cast_fp16 = add(x = var_1623_cast_fp16, y = var_1650_cast_fp16)[name = string("q_embed_7_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> key_states_7_cast_fp16 = transpose(perm = key_states_7_perm_0, x = var_1607_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [1, 16, 125, 64]> var_1655_cast_fp16 = mul(x = key_states_7_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_1655_cast_fp16")];
            tensor<int32, [4]> x1_15_begin_0 = const()[name = string("x1_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_15_end_0 = const()[name = string("x1_15_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_15_end_mask_0 = const()[name = string("x1_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_15_cast_fp16 = slice_by_index(begin = x1_15_begin_0, end = x1_15_end_0, end_mask = x1_15_end_mask_0, x = key_states_7_cast_fp16)[name = string("x1_15_cast_fp16")];
            tensor<int32, [4]> x2_15_begin_0 = const()[name = string("x2_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_15_end_0 = const()[name = string("x2_15_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_15_end_mask_0 = const()[name = string("x2_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_15_cast_fp16 = slice_by_index(begin = x2_15_begin_0, end = x2_15_end_0, end_mask = x2_15_end_mask_0, x = key_states_7_cast_fp16)[name = string("x2_15_cast_fp16")];
            fp16 const_50_promoted_to_fp16 = const()[name = string("const_50_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_1676_cast_fp16 = mul(x = x2_15_cast_fp16, y = const_50_promoted_to_fp16)[name = string("op_1676_cast_fp16")];
            int32 var_1678 = const()[name = string("op_1678"), val = int32(-1)];
            bool var_1679_interleave_0 = const()[name = string("op_1679_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_1679_cast_fp16 = concat(axis = var_1678, interleave = var_1679_interleave_0, values = (var_1676_cast_fp16, x1_15_cast_fp16))[name = string("op_1679_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_1682_cast_fp16 = mul(x = var_1679_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_1682_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> k_embed_7_cast_fp16 = add(x = var_1655_cast_fp16, y = var_1682_cast_fp16)[name = string("k_embed_7_cast_fp16")];
            bool var_1688_transpose_x_1 = const()[name = string("op_1688_transpose_x_1"), val = bool(false)];
            bool var_1688_transpose_y_1 = const()[name = string("op_1688_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 125, 125]> var_1688_cast_fp16 = matmul(transpose_x = var_1688_transpose_x_1, transpose_y = var_1688_transpose_y_1, x = q_embed_7_cast_fp16, y = k_embed_7_cast_fp16)[name = string("op_1688_cast_fp16")];
            fp16 var_1689_to_fp16 = const()[name = string("op_1689_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_19_cast_fp16 = mul(x = var_1688_cast_fp16, y = var_1689_to_fp16)[name = string("attn_weights_19_cast_fp16")];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_21_cast_fp16 = add(x = attn_weights_19_cast_fp16, y = op_1070_to_fp16_palettized)[name = string("attn_weights_21_cast_fp16")];
            int32 var_1706 = const()[name = string("op_1706"), val = int32(-1)];
            tensor<fp16, [1, 16, 125, 125]> var_1708_cast_fp16 = softmax(axis = var_1706, x = attn_weights_21_cast_fp16)[name = string("op_1708_cast_fp16")];
            bool attn_output_7_transpose_x_0 = const()[name = string("attn_output_7_transpose_x_0"), val = bool(false)];
            bool attn_output_7_transpose_y_0 = const()[name = string("attn_output_7_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> value_states_7_cast_fp16 = transpose(perm = value_states_7_perm_0, x = var_1617_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [1, 16, 125, 64]> attn_output_7_cast_fp16 = matmul(transpose_x = attn_output_7_transpose_x_0, transpose_y = attn_output_7_transpose_y_0, x = var_1708_cast_fp16, y = value_states_7_cast_fp16)[name = string("attn_output_7_cast_fp16")];
            tensor<int32, [4]> var_1717_perm_0 = const()[name = string("op_1717_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1719 = const()[name = string("op_1719"), val = tensor<int32, [3]>([1, 125, -1])];
            tensor<fp16, [1, 125, 16, 64]> var_1717_cast_fp16 = transpose(perm = var_1717_perm_0, x = attn_output_7_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [1, 125, 1024]> var_1720_cast_fp16 = reshape(shape = var_1719, x = var_1717_cast_fp16)[name = string("op_1720_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_3_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23413184))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23937536))))[name = string("pre_transformer_layers_3_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_25_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_3_self_attn_o_proj_weight_to_fp16_palettized, x = var_1720_cast_fp16)[name = string("linear_25_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_3_self_attn_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_3_self_attn_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23938112)))];
            tensor<fp16, [1, 125, 512]> var_1727_cast_fp16 = mul(x = pre_transformer_layers_3_self_attn_layer_scale_scale_to_fp16, y = linear_25_cast_fp16)[name = string("op_1727_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_43_cast_fp16 = add(x = hidden_states_37_cast_fp16, y = var_1727_cast_fp16)[name = string("hidden_states_43_cast_fp16")];
            fp16 var_1733_promoted_to_fp16 = const()[name = string("op_1733_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_1739_cast_fp16 = pow(x = hidden_states_43_cast_fp16, y = var_1733_promoted_to_fp16)[name = string("op_1739_cast_fp16")];
            tensor<int32, [1]> variance_15_axes_0 = const()[name = string("variance_15_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_15_keep_dims_0 = const()[name = string("variance_15_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_15_cast_fp16 = reduce_mean(axes = variance_15_axes_0, keep_dims = variance_15_keep_dims_0, x = var_1739_cast_fp16)[name = string("variance_15_cast_fp16")];
            fp16 var_1742_to_fp16 = const()[name = string("op_1742_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_1743_cast_fp16 = add(x = variance_15_cast_fp16, y = var_1742_to_fp16)[name = string("op_1743_cast_fp16")];
            fp32 var_1744_epsilon_0 = const()[name = string("op_1744_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_1744_cast_fp16 = rsqrt(epsilon = var_1744_epsilon_0, x = var_1743_cast_fp16)[name = string("op_1744_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_47_cast_fp16 = mul(x = hidden_states_43_cast_fp16, y = var_1744_cast_fp16)[name = string("hidden_states_47_cast_fp16")];
            tensor<fp16, [512]> const_51_to_fp16 = const()[name = string("const_51_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23939200)))];
            tensor<fp16, [1, 125, 512]> input_67_cast_fp16 = mul(x = const_51_to_fp16, y = hidden_states_47_cast_fp16)[name = string("input_67_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_3_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23940288))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24464640))))[name = string("pre_transformer_layers_3_mlp_gate_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_26_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_3_mlp_gate_proj_weight_to_fp16_palettized, x = input_67_cast_fp16)[name = string("linear_26_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> var_1757_cast_fp16 = silu(x = linear_26_cast_fp16)[name = string("op_1757_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_3_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24465216))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24989568))))[name = string("pre_transformer_layers_3_mlp_up_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_27_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_3_mlp_up_proj_weight_to_fp16_palettized, x = input_67_cast_fp16)[name = string("linear_27_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> input_71_cast_fp16 = mul(x = var_1757_cast_fp16, y = linear_27_cast_fp16)[name = string("input_71_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_3_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24990144))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25514496))))[name = string("pre_transformer_layers_3_mlp_down_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_28_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_3_mlp_down_proj_weight_to_fp16_palettized, x = input_71_cast_fp16)[name = string("linear_28_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_3_mlp_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_3_mlp_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25515072)))];
            tensor<fp16, [1, 125, 512]> var_1764_cast_fp16 = mul(x = pre_transformer_layers_3_mlp_layer_scale_scale_to_fp16, y = linear_28_cast_fp16)[name = string("op_1764_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_49_cast_fp16 = add(x = hidden_states_43_cast_fp16, y = var_1764_cast_fp16)[name = string("hidden_states_49_cast_fp16")];
            fp16 var_1770_promoted_to_fp16 = const()[name = string("op_1770_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_1776_cast_fp16 = pow(x = hidden_states_49_cast_fp16, y = var_1770_promoted_to_fp16)[name = string("op_1776_cast_fp16")];
            tensor<int32, [1]> variance_17_axes_0 = const()[name = string("variance_17_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_17_keep_dims_0 = const()[name = string("variance_17_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_17_cast_fp16 = reduce_mean(axes = variance_17_axes_0, keep_dims = variance_17_keep_dims_0, x = var_1776_cast_fp16)[name = string("variance_17_cast_fp16")];
            fp16 var_1779_to_fp16 = const()[name = string("op_1779_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_1780_cast_fp16 = add(x = variance_17_cast_fp16, y = var_1779_to_fp16)[name = string("op_1780_cast_fp16")];
            fp32 var_1781_epsilon_0 = const()[name = string("op_1781_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_1781_cast_fp16 = rsqrt(epsilon = var_1781_epsilon_0, x = var_1780_cast_fp16)[name = string("op_1781_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_53_cast_fp16 = mul(x = hidden_states_49_cast_fp16, y = var_1781_cast_fp16)[name = string("hidden_states_53_cast_fp16")];
            tensor<fp16, [512]> const_52_to_fp16 = const()[name = string("const_52_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25516160)))];
            tensor<fp16, [1, 125, 512]> h_9_cast_fp16 = mul(x = const_52_to_fp16, y = hidden_states_53_cast_fp16)[name = string("h_9_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25517248))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26041600))))[name = string("pre_transformer_layers_4_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_29_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = h_9_cast_fp16)[name = string("linear_29_cast_fp16")];
            tensor<int32, [4]> var_1807 = const()[name = string("op_1807"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_1808_cast_fp16 = reshape(shape = var_1807, x = linear_29_cast_fp16)[name = string("op_1808_cast_fp16")];
            tensor<int32, [4]> query_states_9_perm_0 = const()[name = string("query_states_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26042176))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26566528))))[name = string("pre_transformer_layers_4_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_30_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = h_9_cast_fp16)[name = string("linear_30_cast_fp16")];
            tensor<int32, [4]> var_1817 = const()[name = string("op_1817"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_1818_cast_fp16 = reshape(shape = var_1817, x = linear_30_cast_fp16)[name = string("op_1818_cast_fp16")];
            tensor<int32, [4]> key_states_9_perm_0 = const()[name = string("key_states_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26567104))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27091456))))[name = string("pre_transformer_layers_4_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_31_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = h_9_cast_fp16)[name = string("linear_31_cast_fp16")];
            tensor<int32, [4]> var_1827 = const()[name = string("op_1827"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_1828_cast_fp16 = reshape(shape = var_1827, x = linear_31_cast_fp16)[name = string("op_1828_cast_fp16")];
            tensor<int32, [4]> value_states_9_perm_0 = const()[name = string("value_states_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 125, 64]> query_states_9_cast_fp16 = transpose(perm = query_states_9_perm_0, x = var_1808_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [1, 16, 125, 64]> var_1834_cast_fp16 = mul(x = query_states_9_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_1834_cast_fp16")];
            tensor<int32, [4]> x1_17_begin_0 = const()[name = string("x1_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_17_end_0 = const()[name = string("x1_17_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_17_end_mask_0 = const()[name = string("x1_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_17_cast_fp16 = slice_by_index(begin = x1_17_begin_0, end = x1_17_end_0, end_mask = x1_17_end_mask_0, x = query_states_9_cast_fp16)[name = string("x1_17_cast_fp16")];
            tensor<int32, [4]> x2_17_begin_0 = const()[name = string("x2_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_17_end_0 = const()[name = string("x2_17_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_17_end_mask_0 = const()[name = string("x2_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_17_cast_fp16 = slice_by_index(begin = x2_17_begin_0, end = x2_17_end_0, end_mask = x2_17_end_mask_0, x = query_states_9_cast_fp16)[name = string("x2_17_cast_fp16")];
            fp16 const_57_promoted_to_fp16 = const()[name = string("const_57_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_1855_cast_fp16 = mul(x = x2_17_cast_fp16, y = const_57_promoted_to_fp16)[name = string("op_1855_cast_fp16")];
            int32 var_1857 = const()[name = string("op_1857"), val = int32(-1)];
            bool var_1858_interleave_0 = const()[name = string("op_1858_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_1858_cast_fp16 = concat(axis = var_1857, interleave = var_1858_interleave_0, values = (var_1855_cast_fp16, x1_17_cast_fp16))[name = string("op_1858_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_1861_cast_fp16 = mul(x = var_1858_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_1861_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> q_embed_9_cast_fp16 = add(x = var_1834_cast_fp16, y = var_1861_cast_fp16)[name = string("q_embed_9_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> key_states_9_cast_fp16 = transpose(perm = key_states_9_perm_0, x = var_1818_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [1, 16, 125, 64]> var_1866_cast_fp16 = mul(x = key_states_9_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_1866_cast_fp16")];
            tensor<int32, [4]> x1_19_begin_0 = const()[name = string("x1_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_19_end_0 = const()[name = string("x1_19_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_19_end_mask_0 = const()[name = string("x1_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_19_cast_fp16 = slice_by_index(begin = x1_19_begin_0, end = x1_19_end_0, end_mask = x1_19_end_mask_0, x = key_states_9_cast_fp16)[name = string("x1_19_cast_fp16")];
            tensor<int32, [4]> x2_19_begin_0 = const()[name = string("x2_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_19_end_0 = const()[name = string("x2_19_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_19_end_mask_0 = const()[name = string("x2_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_19_cast_fp16 = slice_by_index(begin = x2_19_begin_0, end = x2_19_end_0, end_mask = x2_19_end_mask_0, x = key_states_9_cast_fp16)[name = string("x2_19_cast_fp16")];
            fp16 const_60_promoted_to_fp16 = const()[name = string("const_60_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_1887_cast_fp16 = mul(x = x2_19_cast_fp16, y = const_60_promoted_to_fp16)[name = string("op_1887_cast_fp16")];
            int32 var_1889 = const()[name = string("op_1889"), val = int32(-1)];
            bool var_1890_interleave_0 = const()[name = string("op_1890_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_1890_cast_fp16 = concat(axis = var_1889, interleave = var_1890_interleave_0, values = (var_1887_cast_fp16, x1_19_cast_fp16))[name = string("op_1890_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_1893_cast_fp16 = mul(x = var_1890_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_1893_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> k_embed_9_cast_fp16 = add(x = var_1866_cast_fp16, y = var_1893_cast_fp16)[name = string("k_embed_9_cast_fp16")];
            bool var_1899_transpose_x_1 = const()[name = string("op_1899_transpose_x_1"), val = bool(false)];
            bool var_1899_transpose_y_1 = const()[name = string("op_1899_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 125, 125]> var_1899_cast_fp16 = matmul(transpose_x = var_1899_transpose_x_1, transpose_y = var_1899_transpose_y_1, x = q_embed_9_cast_fp16, y = k_embed_9_cast_fp16)[name = string("op_1899_cast_fp16")];
            fp16 var_1900_to_fp16 = const()[name = string("op_1900_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_25_cast_fp16 = mul(x = var_1899_cast_fp16, y = var_1900_to_fp16)[name = string("attn_weights_25_cast_fp16")];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_27_cast_fp16 = add(x = attn_weights_25_cast_fp16, y = op_1070_to_fp16_palettized)[name = string("attn_weights_27_cast_fp16")];
            int32 var_1917 = const()[name = string("op_1917"), val = int32(-1)];
            tensor<fp16, [1, 16, 125, 125]> var_1919_cast_fp16 = softmax(axis = var_1917, x = attn_weights_27_cast_fp16)[name = string("op_1919_cast_fp16")];
            bool attn_output_9_transpose_x_0 = const()[name = string("attn_output_9_transpose_x_0"), val = bool(false)];
            bool attn_output_9_transpose_y_0 = const()[name = string("attn_output_9_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> value_states_9_cast_fp16 = transpose(perm = value_states_9_perm_0, x = var_1828_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [1, 16, 125, 64]> attn_output_9_cast_fp16 = matmul(transpose_x = attn_output_9_transpose_x_0, transpose_y = attn_output_9_transpose_y_0, x = var_1919_cast_fp16, y = value_states_9_cast_fp16)[name = string("attn_output_9_cast_fp16")];
            tensor<int32, [4]> var_1928_perm_0 = const()[name = string("op_1928_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1930 = const()[name = string("op_1930"), val = tensor<int32, [3]>([1, 125, -1])];
            tensor<fp16, [1, 125, 16, 64]> var_1928_cast_fp16 = transpose(perm = var_1928_perm_0, x = attn_output_9_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [1, 125, 1024]> var_1931_cast_fp16 = reshape(shape = var_1930, x = var_1928_cast_fp16)[name = string("op_1931_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_4_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27092032))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27616384))))[name = string("pre_transformer_layers_4_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_32_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_4_self_attn_o_proj_weight_to_fp16_palettized, x = var_1931_cast_fp16)[name = string("linear_32_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_4_self_attn_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_4_self_attn_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27616960)))];
            tensor<fp16, [1, 125, 512]> var_1938_cast_fp16 = mul(x = pre_transformer_layers_4_self_attn_layer_scale_scale_to_fp16, y = linear_32_cast_fp16)[name = string("op_1938_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_55_cast_fp16 = add(x = hidden_states_49_cast_fp16, y = var_1938_cast_fp16)[name = string("hidden_states_55_cast_fp16")];
            fp16 var_1944_promoted_to_fp16 = const()[name = string("op_1944_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_1950_cast_fp16 = pow(x = hidden_states_55_cast_fp16, y = var_1944_promoted_to_fp16)[name = string("op_1950_cast_fp16")];
            tensor<int32, [1]> variance_19_axes_0 = const()[name = string("variance_19_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_19_keep_dims_0 = const()[name = string("variance_19_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_19_cast_fp16 = reduce_mean(axes = variance_19_axes_0, keep_dims = variance_19_keep_dims_0, x = var_1950_cast_fp16)[name = string("variance_19_cast_fp16")];
            fp16 var_1953_to_fp16 = const()[name = string("op_1953_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_1954_cast_fp16 = add(x = variance_19_cast_fp16, y = var_1953_to_fp16)[name = string("op_1954_cast_fp16")];
            fp32 var_1955_epsilon_0 = const()[name = string("op_1955_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_1955_cast_fp16 = rsqrt(epsilon = var_1955_epsilon_0, x = var_1954_cast_fp16)[name = string("op_1955_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_59_cast_fp16 = mul(x = hidden_states_55_cast_fp16, y = var_1955_cast_fp16)[name = string("hidden_states_59_cast_fp16")];
            tensor<fp16, [512]> const_61_to_fp16 = const()[name = string("const_61_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27618048)))];
            tensor<fp16, [1, 125, 512]> input_75_cast_fp16 = mul(x = const_61_to_fp16, y = hidden_states_59_cast_fp16)[name = string("input_75_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_4_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27619136))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28143488))))[name = string("pre_transformer_layers_4_mlp_gate_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_33_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_4_mlp_gate_proj_weight_to_fp16_palettized, x = input_75_cast_fp16)[name = string("linear_33_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> var_1968_cast_fp16 = silu(x = linear_33_cast_fp16)[name = string("op_1968_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_4_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28144064))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28668416))))[name = string("pre_transformer_layers_4_mlp_up_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_34_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_4_mlp_up_proj_weight_to_fp16_palettized, x = input_75_cast_fp16)[name = string("linear_34_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> input_79_cast_fp16 = mul(x = var_1968_cast_fp16, y = linear_34_cast_fp16)[name = string("input_79_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_4_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28668992))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29193344))))[name = string("pre_transformer_layers_4_mlp_down_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_35_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_4_mlp_down_proj_weight_to_fp16_palettized, x = input_79_cast_fp16)[name = string("linear_35_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_4_mlp_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_4_mlp_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29193920)))];
            tensor<fp16, [1, 125, 512]> var_1975_cast_fp16 = mul(x = pre_transformer_layers_4_mlp_layer_scale_scale_to_fp16, y = linear_35_cast_fp16)[name = string("op_1975_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_61_cast_fp16 = add(x = hidden_states_55_cast_fp16, y = var_1975_cast_fp16)[name = string("hidden_states_61_cast_fp16")];
            fp16 var_1981_promoted_to_fp16 = const()[name = string("op_1981_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_1987_cast_fp16 = pow(x = hidden_states_61_cast_fp16, y = var_1981_promoted_to_fp16)[name = string("op_1987_cast_fp16")];
            tensor<int32, [1]> variance_21_axes_0 = const()[name = string("variance_21_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_21_keep_dims_0 = const()[name = string("variance_21_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_21_cast_fp16 = reduce_mean(axes = variance_21_axes_0, keep_dims = variance_21_keep_dims_0, x = var_1987_cast_fp16)[name = string("variance_21_cast_fp16")];
            fp16 var_1990_to_fp16 = const()[name = string("op_1990_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_1991_cast_fp16 = add(x = variance_21_cast_fp16, y = var_1990_to_fp16)[name = string("op_1991_cast_fp16")];
            fp32 var_1992_epsilon_0 = const()[name = string("op_1992_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_1992_cast_fp16 = rsqrt(epsilon = var_1992_epsilon_0, x = var_1991_cast_fp16)[name = string("op_1992_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_65_cast_fp16 = mul(x = hidden_states_61_cast_fp16, y = var_1992_cast_fp16)[name = string("hidden_states_65_cast_fp16")];
            tensor<fp16, [512]> const_62_to_fp16 = const()[name = string("const_62_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29195008)))];
            tensor<fp16, [1, 125, 512]> h_11_cast_fp16 = mul(x = const_62_to_fp16, y = hidden_states_65_cast_fp16)[name = string("h_11_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29196096))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29720448))))[name = string("pre_transformer_layers_5_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_36_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_5_self_attn_q_proj_weight_to_fp16_palettized, x = h_11_cast_fp16)[name = string("linear_36_cast_fp16")];
            tensor<int32, [4]> var_2018 = const()[name = string("op_2018"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_2019_cast_fp16 = reshape(shape = var_2018, x = linear_36_cast_fp16)[name = string("op_2019_cast_fp16")];
            tensor<int32, [4]> query_states_11_perm_0 = const()[name = string("query_states_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29721024))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30245376))))[name = string("pre_transformer_layers_5_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_5_self_attn_k_proj_weight_to_fp16_palettized, x = h_11_cast_fp16)[name = string("linear_37_cast_fp16")];
            tensor<int32, [4]> var_2028 = const()[name = string("op_2028"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_2029_cast_fp16 = reshape(shape = var_2028, x = linear_37_cast_fp16)[name = string("op_2029_cast_fp16")];
            tensor<int32, [4]> key_states_11_perm_0 = const()[name = string("key_states_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30245952))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30770304))))[name = string("pre_transformer_layers_5_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_38_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_5_self_attn_v_proj_weight_to_fp16_palettized, x = h_11_cast_fp16)[name = string("linear_38_cast_fp16")];
            tensor<int32, [4]> var_2038 = const()[name = string("op_2038"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_2039_cast_fp16 = reshape(shape = var_2038, x = linear_38_cast_fp16)[name = string("op_2039_cast_fp16")];
            tensor<int32, [4]> value_states_11_perm_0 = const()[name = string("value_states_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 125, 64]> query_states_11_cast_fp16 = transpose(perm = query_states_11_perm_0, x = var_2019_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [1, 16, 125, 64]> var_2045_cast_fp16 = mul(x = query_states_11_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_2045_cast_fp16")];
            tensor<int32, [4]> x1_21_begin_0 = const()[name = string("x1_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_21_end_0 = const()[name = string("x1_21_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_21_end_mask_0 = const()[name = string("x1_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_21_cast_fp16 = slice_by_index(begin = x1_21_begin_0, end = x1_21_end_0, end_mask = x1_21_end_mask_0, x = query_states_11_cast_fp16)[name = string("x1_21_cast_fp16")];
            tensor<int32, [4]> x2_21_begin_0 = const()[name = string("x2_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_21_end_0 = const()[name = string("x2_21_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_21_end_mask_0 = const()[name = string("x2_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_21_cast_fp16 = slice_by_index(begin = x2_21_begin_0, end = x2_21_end_0, end_mask = x2_21_end_mask_0, x = query_states_11_cast_fp16)[name = string("x2_21_cast_fp16")];
            fp16 const_67_promoted_to_fp16 = const()[name = string("const_67_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_2066_cast_fp16 = mul(x = x2_21_cast_fp16, y = const_67_promoted_to_fp16)[name = string("op_2066_cast_fp16")];
            int32 var_2068 = const()[name = string("op_2068"), val = int32(-1)];
            bool var_2069_interleave_0 = const()[name = string("op_2069_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_2069_cast_fp16 = concat(axis = var_2068, interleave = var_2069_interleave_0, values = (var_2066_cast_fp16, x1_21_cast_fp16))[name = string("op_2069_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_2072_cast_fp16 = mul(x = var_2069_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_2072_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> q_embed_11_cast_fp16 = add(x = var_2045_cast_fp16, y = var_2072_cast_fp16)[name = string("q_embed_11_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> key_states_11_cast_fp16 = transpose(perm = key_states_11_perm_0, x = var_2029_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [1, 16, 125, 64]> var_2077_cast_fp16 = mul(x = key_states_11_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_2077_cast_fp16")];
            tensor<int32, [4]> x1_23_begin_0 = const()[name = string("x1_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_23_end_0 = const()[name = string("x1_23_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_23_end_mask_0 = const()[name = string("x1_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_23_cast_fp16 = slice_by_index(begin = x1_23_begin_0, end = x1_23_end_0, end_mask = x1_23_end_mask_0, x = key_states_11_cast_fp16)[name = string("x1_23_cast_fp16")];
            tensor<int32, [4]> x2_23_begin_0 = const()[name = string("x2_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_23_end_0 = const()[name = string("x2_23_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_23_end_mask_0 = const()[name = string("x2_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_23_cast_fp16 = slice_by_index(begin = x2_23_begin_0, end = x2_23_end_0, end_mask = x2_23_end_mask_0, x = key_states_11_cast_fp16)[name = string("x2_23_cast_fp16")];
            fp16 const_70_promoted_to_fp16 = const()[name = string("const_70_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_2098_cast_fp16 = mul(x = x2_23_cast_fp16, y = const_70_promoted_to_fp16)[name = string("op_2098_cast_fp16")];
            int32 var_2100 = const()[name = string("op_2100"), val = int32(-1)];
            bool var_2101_interleave_0 = const()[name = string("op_2101_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_2101_cast_fp16 = concat(axis = var_2100, interleave = var_2101_interleave_0, values = (var_2098_cast_fp16, x1_23_cast_fp16))[name = string("op_2101_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_2104_cast_fp16 = mul(x = var_2101_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_2104_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> k_embed_11_cast_fp16 = add(x = var_2077_cast_fp16, y = var_2104_cast_fp16)[name = string("k_embed_11_cast_fp16")];
            bool var_2110_transpose_x_1 = const()[name = string("op_2110_transpose_x_1"), val = bool(false)];
            bool var_2110_transpose_y_1 = const()[name = string("op_2110_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 125, 125]> var_2110_cast_fp16 = matmul(transpose_x = var_2110_transpose_x_1, transpose_y = var_2110_transpose_y_1, x = q_embed_11_cast_fp16, y = k_embed_11_cast_fp16)[name = string("op_2110_cast_fp16")];
            fp16 var_2111_to_fp16 = const()[name = string("op_2111_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_31_cast_fp16 = mul(x = var_2110_cast_fp16, y = var_2111_to_fp16)[name = string("attn_weights_31_cast_fp16")];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_33_cast_fp16 = add(x = attn_weights_31_cast_fp16, y = op_1070_to_fp16_palettized)[name = string("attn_weights_33_cast_fp16")];
            int32 var_2128 = const()[name = string("op_2128"), val = int32(-1)];
            tensor<fp16, [1, 16, 125, 125]> var_2130_cast_fp16 = softmax(axis = var_2128, x = attn_weights_33_cast_fp16)[name = string("op_2130_cast_fp16")];
            bool attn_output_11_transpose_x_0 = const()[name = string("attn_output_11_transpose_x_0"), val = bool(false)];
            bool attn_output_11_transpose_y_0 = const()[name = string("attn_output_11_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> value_states_11_cast_fp16 = transpose(perm = value_states_11_perm_0, x = var_2039_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [1, 16, 125, 64]> attn_output_11_cast_fp16 = matmul(transpose_x = attn_output_11_transpose_x_0, transpose_y = attn_output_11_transpose_y_0, x = var_2130_cast_fp16, y = value_states_11_cast_fp16)[name = string("attn_output_11_cast_fp16")];
            tensor<int32, [4]> var_2139_perm_0 = const()[name = string("op_2139_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2141 = const()[name = string("op_2141"), val = tensor<int32, [3]>([1, 125, -1])];
            tensor<fp16, [1, 125, 16, 64]> var_2139_cast_fp16 = transpose(perm = var_2139_perm_0, x = attn_output_11_cast_fp16)[name = string("transpose_13")];
            tensor<fp16, [1, 125, 1024]> var_2142_cast_fp16 = reshape(shape = var_2141, x = var_2139_cast_fp16)[name = string("op_2142_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_5_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30770880))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31295232))))[name = string("pre_transformer_layers_5_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_39_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_5_self_attn_o_proj_weight_to_fp16_palettized, x = var_2142_cast_fp16)[name = string("linear_39_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_5_self_attn_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_5_self_attn_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31295808)))];
            tensor<fp16, [1, 125, 512]> var_2149_cast_fp16 = mul(x = pre_transformer_layers_5_self_attn_layer_scale_scale_to_fp16, y = linear_39_cast_fp16)[name = string("op_2149_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_67_cast_fp16 = add(x = hidden_states_61_cast_fp16, y = var_2149_cast_fp16)[name = string("hidden_states_67_cast_fp16")];
            fp16 var_2155_promoted_to_fp16 = const()[name = string("op_2155_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_2161_cast_fp16 = pow(x = hidden_states_67_cast_fp16, y = var_2155_promoted_to_fp16)[name = string("op_2161_cast_fp16")];
            tensor<int32, [1]> variance_23_axes_0 = const()[name = string("variance_23_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_23_keep_dims_0 = const()[name = string("variance_23_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_23_cast_fp16 = reduce_mean(axes = variance_23_axes_0, keep_dims = variance_23_keep_dims_0, x = var_2161_cast_fp16)[name = string("variance_23_cast_fp16")];
            fp16 var_2164_to_fp16 = const()[name = string("op_2164_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_2165_cast_fp16 = add(x = variance_23_cast_fp16, y = var_2164_to_fp16)[name = string("op_2165_cast_fp16")];
            fp32 var_2166_epsilon_0 = const()[name = string("op_2166_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_2166_cast_fp16 = rsqrt(epsilon = var_2166_epsilon_0, x = var_2165_cast_fp16)[name = string("op_2166_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_71_cast_fp16 = mul(x = hidden_states_67_cast_fp16, y = var_2166_cast_fp16)[name = string("hidden_states_71_cast_fp16")];
            tensor<fp16, [512]> const_71_to_fp16 = const()[name = string("const_71_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31296896)))];
            tensor<fp16, [1, 125, 512]> input_83_cast_fp16 = mul(x = const_71_to_fp16, y = hidden_states_71_cast_fp16)[name = string("input_83_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_5_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31297984))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31822336))))[name = string("pre_transformer_layers_5_mlp_gate_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_40_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_5_mlp_gate_proj_weight_to_fp16_palettized, x = input_83_cast_fp16)[name = string("linear_40_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> var_2179_cast_fp16 = silu(x = linear_40_cast_fp16)[name = string("op_2179_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_5_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31822912))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32347264))))[name = string("pre_transformer_layers_5_mlp_up_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_41_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_5_mlp_up_proj_weight_to_fp16_palettized, x = input_83_cast_fp16)[name = string("linear_41_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> input_87_cast_fp16 = mul(x = var_2179_cast_fp16, y = linear_41_cast_fp16)[name = string("input_87_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_5_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32347840))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32872192))))[name = string("pre_transformer_layers_5_mlp_down_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_42_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_5_mlp_down_proj_weight_to_fp16_palettized, x = input_87_cast_fp16)[name = string("linear_42_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_5_mlp_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_5_mlp_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32872768)))];
            tensor<fp16, [1, 125, 512]> var_2186_cast_fp16 = mul(x = pre_transformer_layers_5_mlp_layer_scale_scale_to_fp16, y = linear_42_cast_fp16)[name = string("op_2186_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_73_cast_fp16 = add(x = hidden_states_67_cast_fp16, y = var_2186_cast_fp16)[name = string("hidden_states_73_cast_fp16")];
            fp16 var_2192_promoted_to_fp16 = const()[name = string("op_2192_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_2198_cast_fp16 = pow(x = hidden_states_73_cast_fp16, y = var_2192_promoted_to_fp16)[name = string("op_2198_cast_fp16")];
            tensor<int32, [1]> variance_25_axes_0 = const()[name = string("variance_25_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_25_keep_dims_0 = const()[name = string("variance_25_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_25_cast_fp16 = reduce_mean(axes = variance_25_axes_0, keep_dims = variance_25_keep_dims_0, x = var_2198_cast_fp16)[name = string("variance_25_cast_fp16")];
            fp16 var_2201_to_fp16 = const()[name = string("op_2201_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_2202_cast_fp16 = add(x = variance_25_cast_fp16, y = var_2201_to_fp16)[name = string("op_2202_cast_fp16")];
            fp32 var_2203_epsilon_0 = const()[name = string("op_2203_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_2203_cast_fp16 = rsqrt(epsilon = var_2203_epsilon_0, x = var_2202_cast_fp16)[name = string("op_2203_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_77_cast_fp16 = mul(x = hidden_states_73_cast_fp16, y = var_2203_cast_fp16)[name = string("hidden_states_77_cast_fp16")];
            tensor<fp16, [512]> const_72_to_fp16 = const()[name = string("const_72_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32873856)))];
            tensor<fp16, [1, 125, 512]> h_13_cast_fp16 = mul(x = const_72_to_fp16, y = hidden_states_77_cast_fp16)[name = string("h_13_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32874944))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33399296))))[name = string("pre_transformer_layers_6_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_6_self_attn_q_proj_weight_to_fp16_palettized, x = h_13_cast_fp16)[name = string("linear_43_cast_fp16")];
            tensor<int32, [4]> var_2229 = const()[name = string("op_2229"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_2230_cast_fp16 = reshape(shape = var_2229, x = linear_43_cast_fp16)[name = string("op_2230_cast_fp16")];
            tensor<int32, [4]> query_states_13_perm_0 = const()[name = string("query_states_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33399872))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33924224))))[name = string("pre_transformer_layers_6_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_44_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_6_self_attn_k_proj_weight_to_fp16_palettized, x = h_13_cast_fp16)[name = string("linear_44_cast_fp16")];
            tensor<int32, [4]> var_2239 = const()[name = string("op_2239"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_2240_cast_fp16 = reshape(shape = var_2239, x = linear_44_cast_fp16)[name = string("op_2240_cast_fp16")];
            tensor<int32, [4]> key_states_13_perm_0 = const()[name = string("key_states_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33924800))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34449152))))[name = string("pre_transformer_layers_6_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_45_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_6_self_attn_v_proj_weight_to_fp16_palettized, x = h_13_cast_fp16)[name = string("linear_45_cast_fp16")];
            tensor<int32, [4]> var_2249 = const()[name = string("op_2249"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_2250_cast_fp16 = reshape(shape = var_2249, x = linear_45_cast_fp16)[name = string("op_2250_cast_fp16")];
            tensor<int32, [4]> value_states_13_perm_0 = const()[name = string("value_states_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 125, 64]> query_states_13_cast_fp16 = transpose(perm = query_states_13_perm_0, x = var_2230_cast_fp16)[name = string("transpose_12")];
            tensor<fp16, [1, 16, 125, 64]> var_2256_cast_fp16 = mul(x = query_states_13_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_2256_cast_fp16")];
            tensor<int32, [4]> x1_25_begin_0 = const()[name = string("x1_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_25_end_0 = const()[name = string("x1_25_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_25_end_mask_0 = const()[name = string("x1_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_25_cast_fp16 = slice_by_index(begin = x1_25_begin_0, end = x1_25_end_0, end_mask = x1_25_end_mask_0, x = query_states_13_cast_fp16)[name = string("x1_25_cast_fp16")];
            tensor<int32, [4]> x2_25_begin_0 = const()[name = string("x2_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_25_end_0 = const()[name = string("x2_25_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_25_end_mask_0 = const()[name = string("x2_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_25_cast_fp16 = slice_by_index(begin = x2_25_begin_0, end = x2_25_end_0, end_mask = x2_25_end_mask_0, x = query_states_13_cast_fp16)[name = string("x2_25_cast_fp16")];
            fp16 const_77_promoted_to_fp16 = const()[name = string("const_77_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_2277_cast_fp16 = mul(x = x2_25_cast_fp16, y = const_77_promoted_to_fp16)[name = string("op_2277_cast_fp16")];
            int32 var_2279 = const()[name = string("op_2279"), val = int32(-1)];
            bool var_2280_interleave_0 = const()[name = string("op_2280_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_2280_cast_fp16 = concat(axis = var_2279, interleave = var_2280_interleave_0, values = (var_2277_cast_fp16, x1_25_cast_fp16))[name = string("op_2280_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_2283_cast_fp16 = mul(x = var_2280_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_2283_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> q_embed_13_cast_fp16 = add(x = var_2256_cast_fp16, y = var_2283_cast_fp16)[name = string("q_embed_13_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> key_states_13_cast_fp16 = transpose(perm = key_states_13_perm_0, x = var_2240_cast_fp16)[name = string("transpose_11")];
            tensor<fp16, [1, 16, 125, 64]> var_2288_cast_fp16 = mul(x = key_states_13_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_2288_cast_fp16")];
            tensor<int32, [4]> x1_27_begin_0 = const()[name = string("x1_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_27_end_0 = const()[name = string("x1_27_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_27_end_mask_0 = const()[name = string("x1_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_27_cast_fp16 = slice_by_index(begin = x1_27_begin_0, end = x1_27_end_0, end_mask = x1_27_end_mask_0, x = key_states_13_cast_fp16)[name = string("x1_27_cast_fp16")];
            tensor<int32, [4]> x2_27_begin_0 = const()[name = string("x2_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_27_end_0 = const()[name = string("x2_27_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_27_end_mask_0 = const()[name = string("x2_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_27_cast_fp16 = slice_by_index(begin = x2_27_begin_0, end = x2_27_end_0, end_mask = x2_27_end_mask_0, x = key_states_13_cast_fp16)[name = string("x2_27_cast_fp16")];
            fp16 const_80_promoted_to_fp16 = const()[name = string("const_80_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_2309_cast_fp16 = mul(x = x2_27_cast_fp16, y = const_80_promoted_to_fp16)[name = string("op_2309_cast_fp16")];
            int32 var_2311 = const()[name = string("op_2311"), val = int32(-1)];
            bool var_2312_interleave_0 = const()[name = string("op_2312_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_2312_cast_fp16 = concat(axis = var_2311, interleave = var_2312_interleave_0, values = (var_2309_cast_fp16, x1_27_cast_fp16))[name = string("op_2312_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_2315_cast_fp16 = mul(x = var_2312_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_2315_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> k_embed_13_cast_fp16 = add(x = var_2288_cast_fp16, y = var_2315_cast_fp16)[name = string("k_embed_13_cast_fp16")];
            bool var_2321_transpose_x_1 = const()[name = string("op_2321_transpose_x_1"), val = bool(false)];
            bool var_2321_transpose_y_1 = const()[name = string("op_2321_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 125, 125]> var_2321_cast_fp16 = matmul(transpose_x = var_2321_transpose_x_1, transpose_y = var_2321_transpose_y_1, x = q_embed_13_cast_fp16, y = k_embed_13_cast_fp16)[name = string("op_2321_cast_fp16")];
            fp16 var_2322_to_fp16 = const()[name = string("op_2322_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_37_cast_fp16 = mul(x = var_2321_cast_fp16, y = var_2322_to_fp16)[name = string("attn_weights_37_cast_fp16")];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_39_cast_fp16 = add(x = attn_weights_37_cast_fp16, y = op_1070_to_fp16_palettized)[name = string("attn_weights_39_cast_fp16")];
            int32 var_2339 = const()[name = string("op_2339"), val = int32(-1)];
            tensor<fp16, [1, 16, 125, 125]> var_2341_cast_fp16 = softmax(axis = var_2339, x = attn_weights_39_cast_fp16)[name = string("op_2341_cast_fp16")];
            bool attn_output_13_transpose_x_0 = const()[name = string("attn_output_13_transpose_x_0"), val = bool(false)];
            bool attn_output_13_transpose_y_0 = const()[name = string("attn_output_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> value_states_13_cast_fp16 = transpose(perm = value_states_13_perm_0, x = var_2250_cast_fp16)[name = string("transpose_10")];
            tensor<fp16, [1, 16, 125, 64]> attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = var_2341_cast_fp16, y = value_states_13_cast_fp16)[name = string("attn_output_13_cast_fp16")];
            tensor<int32, [4]> var_2350_perm_0 = const()[name = string("op_2350_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2352 = const()[name = string("op_2352"), val = tensor<int32, [3]>([1, 125, -1])];
            tensor<fp16, [1, 125, 16, 64]> var_2350_cast_fp16 = transpose(perm = var_2350_perm_0, x = attn_output_13_cast_fp16)[name = string("transpose_9")];
            tensor<fp16, [1, 125, 1024]> var_2353_cast_fp16 = reshape(shape = var_2352, x = var_2350_cast_fp16)[name = string("op_2353_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_6_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34449728))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34974080))))[name = string("pre_transformer_layers_6_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_46_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_6_self_attn_o_proj_weight_to_fp16_palettized, x = var_2353_cast_fp16)[name = string("linear_46_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_6_self_attn_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_6_self_attn_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34974656)))];
            tensor<fp16, [1, 125, 512]> var_2360_cast_fp16 = mul(x = pre_transformer_layers_6_self_attn_layer_scale_scale_to_fp16, y = linear_46_cast_fp16)[name = string("op_2360_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_79_cast_fp16 = add(x = hidden_states_73_cast_fp16, y = var_2360_cast_fp16)[name = string("hidden_states_79_cast_fp16")];
            fp16 var_2366_promoted_to_fp16 = const()[name = string("op_2366_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_2372_cast_fp16 = pow(x = hidden_states_79_cast_fp16, y = var_2366_promoted_to_fp16)[name = string("op_2372_cast_fp16")];
            tensor<int32, [1]> variance_27_axes_0 = const()[name = string("variance_27_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_27_keep_dims_0 = const()[name = string("variance_27_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_27_cast_fp16 = reduce_mean(axes = variance_27_axes_0, keep_dims = variance_27_keep_dims_0, x = var_2372_cast_fp16)[name = string("variance_27_cast_fp16")];
            fp16 var_2375_to_fp16 = const()[name = string("op_2375_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_2376_cast_fp16 = add(x = variance_27_cast_fp16, y = var_2375_to_fp16)[name = string("op_2376_cast_fp16")];
            fp32 var_2377_epsilon_0 = const()[name = string("op_2377_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_2377_cast_fp16 = rsqrt(epsilon = var_2377_epsilon_0, x = var_2376_cast_fp16)[name = string("op_2377_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_83_cast_fp16 = mul(x = hidden_states_79_cast_fp16, y = var_2377_cast_fp16)[name = string("hidden_states_83_cast_fp16")];
            tensor<fp16, [512]> const_81_to_fp16 = const()[name = string("const_81_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34975744)))];
            tensor<fp16, [1, 125, 512]> input_91_cast_fp16 = mul(x = const_81_to_fp16, y = hidden_states_83_cast_fp16)[name = string("input_91_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_6_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34976832))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35501184))))[name = string("pre_transformer_layers_6_mlp_gate_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_47_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_6_mlp_gate_proj_weight_to_fp16_palettized, x = input_91_cast_fp16)[name = string("linear_47_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> var_2390_cast_fp16 = silu(x = linear_47_cast_fp16)[name = string("op_2390_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_6_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35501760))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36026112))))[name = string("pre_transformer_layers_6_mlp_up_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_48_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_6_mlp_up_proj_weight_to_fp16_palettized, x = input_91_cast_fp16)[name = string("linear_48_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> input_95_cast_fp16 = mul(x = var_2390_cast_fp16, y = linear_48_cast_fp16)[name = string("input_95_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_6_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36026688))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36551040))))[name = string("pre_transformer_layers_6_mlp_down_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_49_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_6_mlp_down_proj_weight_to_fp16_palettized, x = input_95_cast_fp16)[name = string("linear_49_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_6_mlp_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_6_mlp_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36551616)))];
            tensor<fp16, [1, 125, 512]> var_2397_cast_fp16 = mul(x = pre_transformer_layers_6_mlp_layer_scale_scale_to_fp16, y = linear_49_cast_fp16)[name = string("op_2397_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_85_cast_fp16 = add(x = hidden_states_79_cast_fp16, y = var_2397_cast_fp16)[name = string("hidden_states_85_cast_fp16")];
            fp16 var_2403_promoted_to_fp16 = const()[name = string("op_2403_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_2409_cast_fp16 = pow(x = hidden_states_85_cast_fp16, y = var_2403_promoted_to_fp16)[name = string("op_2409_cast_fp16")];
            tensor<int32, [1]> variance_29_axes_0 = const()[name = string("variance_29_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_29_keep_dims_0 = const()[name = string("variance_29_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_29_cast_fp16 = reduce_mean(axes = variance_29_axes_0, keep_dims = variance_29_keep_dims_0, x = var_2409_cast_fp16)[name = string("variance_29_cast_fp16")];
            fp16 var_2412_to_fp16 = const()[name = string("op_2412_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_2413_cast_fp16 = add(x = variance_29_cast_fp16, y = var_2412_to_fp16)[name = string("op_2413_cast_fp16")];
            fp32 var_2414_epsilon_0 = const()[name = string("op_2414_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_2414_cast_fp16 = rsqrt(epsilon = var_2414_epsilon_0, x = var_2413_cast_fp16)[name = string("op_2414_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_89_cast_fp16 = mul(x = hidden_states_85_cast_fp16, y = var_2414_cast_fp16)[name = string("hidden_states_89_cast_fp16")];
            tensor<fp16, [512]> const_82_to_fp16 = const()[name = string("const_82_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36552704)))];
            tensor<fp16, [1, 125, 512]> h_cast_fp16 = mul(x = const_82_to_fp16, y = hidden_states_89_cast_fp16)[name = string("h_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36553792))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37078144))))[name = string("pre_transformer_layers_7_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_50_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_7_self_attn_q_proj_weight_to_fp16_palettized, x = h_cast_fp16)[name = string("linear_50_cast_fp16")];
            tensor<int32, [4]> var_2440 = const()[name = string("op_2440"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_2441_cast_fp16 = reshape(shape = var_2440, x = linear_50_cast_fp16)[name = string("op_2441_cast_fp16")];
            tensor<int32, [4]> query_states_perm_0 = const()[name = string("query_states_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37078720))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37603072))))[name = string("pre_transformer_layers_7_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_51_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_7_self_attn_k_proj_weight_to_fp16_palettized, x = h_cast_fp16)[name = string("linear_51_cast_fp16")];
            tensor<int32, [4]> var_2450 = const()[name = string("op_2450"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_2451_cast_fp16 = reshape(shape = var_2450, x = linear_51_cast_fp16)[name = string("op_2451_cast_fp16")];
            tensor<int32, [4]> key_states_perm_0 = const()[name = string("key_states_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1024, 512]> pre_transformer_layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37603648))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38128000))))[name = string("pre_transformer_layers_7_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_52_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_7_self_attn_v_proj_weight_to_fp16_palettized, x = h_cast_fp16)[name = string("linear_52_cast_fp16")];
            tensor<int32, [4]> var_2460 = const()[name = string("op_2460"), val = tensor<int32, [4]>([1, 125, -1, 64])];
            tensor<fp16, [1, 125, 16, 64]> var_2461_cast_fp16 = reshape(shape = var_2460, x = linear_52_cast_fp16)[name = string("op_2461_cast_fp16")];
            tensor<int32, [4]> value_states_perm_0 = const()[name = string("value_states_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 125, 64]> query_states_cast_fp16 = transpose(perm = query_states_perm_0, x = var_2441_cast_fp16)[name = string("transpose_8")];
            tensor<fp16, [1, 16, 125, 64]> var_2467_cast_fp16 = mul(x = query_states_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_2467_cast_fp16")];
            tensor<int32, [4]> x1_29_begin_0 = const()[name = string("x1_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_29_end_0 = const()[name = string("x1_29_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_29_end_mask_0 = const()[name = string("x1_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_29_cast_fp16 = slice_by_index(begin = x1_29_begin_0, end = x1_29_end_0, end_mask = x1_29_end_mask_0, x = query_states_cast_fp16)[name = string("x1_29_cast_fp16")];
            tensor<int32, [4]> x2_29_begin_0 = const()[name = string("x2_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_29_end_0 = const()[name = string("x2_29_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_29_end_mask_0 = const()[name = string("x2_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_29_cast_fp16 = slice_by_index(begin = x2_29_begin_0, end = x2_29_end_0, end_mask = x2_29_end_mask_0, x = query_states_cast_fp16)[name = string("x2_29_cast_fp16")];
            fp16 const_87_promoted_to_fp16 = const()[name = string("const_87_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_2488_cast_fp16 = mul(x = x2_29_cast_fp16, y = const_87_promoted_to_fp16)[name = string("op_2488_cast_fp16")];
            int32 var_2490 = const()[name = string("op_2490"), val = int32(-1)];
            bool var_2491_interleave_0 = const()[name = string("op_2491_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_2491_cast_fp16 = concat(axis = var_2490, interleave = var_2491_interleave_0, values = (var_2488_cast_fp16, x1_29_cast_fp16))[name = string("op_2491_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_2494_cast_fp16 = mul(x = var_2491_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_2494_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> q_embed_cast_fp16 = add(x = var_2467_cast_fp16, y = var_2494_cast_fp16)[name = string("q_embed_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> key_states_cast_fp16 = transpose(perm = key_states_perm_0, x = var_2451_cast_fp16)[name = string("transpose_7")];
            tensor<fp16, [1, 16, 125, 64]> var_2499_cast_fp16 = mul(x = key_states_cast_fp16, y = op_989_to_fp16_palettized)[name = string("op_2499_cast_fp16")];
            tensor<int32, [4]> x1_begin_0 = const()[name = string("x1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_end_0 = const()[name = string("x1_end_0"), val = tensor<int32, [4]>([1, 16, 125, 32])];
            tensor<bool, [4]> x1_end_mask_0 = const()[name = string("x1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 125, 32]> x1_cast_fp16 = slice_by_index(begin = x1_begin_0, end = x1_end_0, end_mask = x1_end_mask_0, x = key_states_cast_fp16)[name = string("x1_cast_fp16")];
            tensor<int32, [4]> x2_begin_0 = const()[name = string("x2_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_end_0 = const()[name = string("x2_end_0"), val = tensor<int32, [4]>([1, 16, 125, 64])];
            tensor<bool, [4]> x2_end_mask_0 = const()[name = string("x2_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 125, 32]> x2_cast_fp16 = slice_by_index(begin = x2_begin_0, end = x2_end_0, end_mask = x2_end_mask_0, x = key_states_cast_fp16)[name = string("x2_cast_fp16")];
            fp16 const_90_promoted_to_fp16 = const()[name = string("const_90_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 125, 32]> var_2520_cast_fp16 = mul(x = x2_cast_fp16, y = const_90_promoted_to_fp16)[name = string("op_2520_cast_fp16")];
            int32 var_2522 = const()[name = string("op_2522"), val = int32(-1)];
            bool var_2523_interleave_0 = const()[name = string("op_2523_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> var_2523_cast_fp16 = concat(axis = var_2522, interleave = var_2523_interleave_0, values = (var_2520_cast_fp16, x1_cast_fp16))[name = string("op_2523_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> var_2526_cast_fp16 = mul(x = var_2523_cast_fp16, y = op_1016_to_fp16_palettized)[name = string("op_2526_cast_fp16")];
            tensor<fp16, [1, 16, 125, 64]> k_embed_cast_fp16 = add(x = var_2499_cast_fp16, y = var_2526_cast_fp16)[name = string("k_embed_cast_fp16")];
            bool var_2532_transpose_x_1 = const()[name = string("op_2532_transpose_x_1"), val = bool(false)];
            bool var_2532_transpose_y_1 = const()[name = string("op_2532_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 125, 125]> var_2532_cast_fp16 = matmul(transpose_x = var_2532_transpose_x_1, transpose_y = var_2532_transpose_y_1, x = q_embed_cast_fp16, y = k_embed_cast_fp16)[name = string("op_2532_cast_fp16")];
            fp16 var_2533_to_fp16 = const()[name = string("op_2533_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_43_cast_fp16 = mul(x = var_2532_cast_fp16, y = var_2533_to_fp16)[name = string("attn_weights_43_cast_fp16")];
            tensor<fp16, [1, 16, 125, 125]> attn_weights_45_cast_fp16 = add(x = attn_weights_43_cast_fp16, y = op_1070_to_fp16_palettized)[name = string("attn_weights_45_cast_fp16")];
            int32 var_2550 = const()[name = string("op_2550"), val = int32(-1)];
            tensor<fp16, [1, 16, 125, 125]> var_2552_cast_fp16 = softmax(axis = var_2550, x = attn_weights_45_cast_fp16)[name = string("op_2552_cast_fp16")];
            bool attn_output_transpose_x_0 = const()[name = string("attn_output_transpose_x_0"), val = bool(false)];
            bool attn_output_transpose_y_0 = const()[name = string("attn_output_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 125, 64]> value_states_cast_fp16 = transpose(perm = value_states_perm_0, x = var_2461_cast_fp16)[name = string("transpose_6")];
            tensor<fp16, [1, 16, 125, 64]> attn_output_cast_fp16 = matmul(transpose_x = attn_output_transpose_x_0, transpose_y = attn_output_transpose_y_0, x = var_2552_cast_fp16, y = value_states_cast_fp16)[name = string("attn_output_cast_fp16")];
            tensor<int32, [4]> var_2561_perm_0 = const()[name = string("op_2561_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2563 = const()[name = string("op_2563"), val = tensor<int32, [3]>([1, 125, -1])];
            tensor<fp16, [1, 125, 16, 64]> var_2561_cast_fp16 = transpose(perm = var_2561_perm_0, x = attn_output_cast_fp16)[name = string("transpose_5")];
            tensor<fp16, [1, 125, 1024]> var_2564_cast_fp16 = reshape(shape = var_2563, x = var_2561_cast_fp16)[name = string("op_2564_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_7_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38128576))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38652928))))[name = string("pre_transformer_layers_7_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_53_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_7_self_attn_o_proj_weight_to_fp16_palettized, x = var_2564_cast_fp16)[name = string("linear_53_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_7_self_attn_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_7_self_attn_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38653504)))];
            tensor<fp16, [1, 125, 512]> var_2571_cast_fp16 = mul(x = pre_transformer_layers_7_self_attn_layer_scale_scale_to_fp16, y = linear_53_cast_fp16)[name = string("op_2571_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_91_cast_fp16 = add(x = hidden_states_85_cast_fp16, y = var_2571_cast_fp16)[name = string("hidden_states_91_cast_fp16")];
            fp16 var_2577_promoted_to_fp16 = const()[name = string("op_2577_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_2583_cast_fp16 = pow(x = hidden_states_91_cast_fp16, y = var_2577_promoted_to_fp16)[name = string("op_2583_cast_fp16")];
            tensor<int32, [1]> variance_31_axes_0 = const()[name = string("variance_31_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_31_keep_dims_0 = const()[name = string("variance_31_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_31_cast_fp16 = reduce_mean(axes = variance_31_axes_0, keep_dims = variance_31_keep_dims_0, x = var_2583_cast_fp16)[name = string("variance_31_cast_fp16")];
            fp16 var_2586_to_fp16 = const()[name = string("op_2586_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_2587_cast_fp16 = add(x = variance_31_cast_fp16, y = var_2586_to_fp16)[name = string("op_2587_cast_fp16")];
            fp32 var_2588_epsilon_0 = const()[name = string("op_2588_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_2588_cast_fp16 = rsqrt(epsilon = var_2588_epsilon_0, x = var_2587_cast_fp16)[name = string("op_2588_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_95_cast_fp16 = mul(x = hidden_states_91_cast_fp16, y = var_2588_cast_fp16)[name = string("hidden_states_95_cast_fp16")];
            tensor<fp16, [512]> const_91_to_fp16 = const()[name = string("const_91_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38654592)))];
            tensor<fp16, [1, 125, 512]> input_99_cast_fp16 = mul(x = const_91_to_fp16, y = hidden_states_95_cast_fp16)[name = string("input_99_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_7_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38655680))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39180032))))[name = string("pre_transformer_layers_7_mlp_gate_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_54_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_7_mlp_gate_proj_weight_to_fp16_palettized, x = input_99_cast_fp16)[name = string("linear_54_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> var_2601_cast_fp16 = silu(x = linear_54_cast_fp16)[name = string("op_2601_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_layers_7_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39180608))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39704960))))[name = string("pre_transformer_layers_7_mlp_up_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 1024]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = pre_transformer_layers_7_mlp_up_proj_weight_to_fp16_palettized, x = input_99_cast_fp16)[name = string("linear_55_cast_fp16")];
            tensor<fp16, [1, 125, 1024]> input_103_cast_fp16 = mul(x = var_2601_cast_fp16, y = linear_55_cast_fp16)[name = string("input_103_cast_fp16")];
            tensor<fp16, [512, 1024]> pre_transformer_layers_7_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39705536))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40229888))))[name = string("pre_transformer_layers_7_mlp_down_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 125, 512]> linear_56_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = pre_transformer_layers_7_mlp_down_proj_weight_to_fp16_palettized, x = input_103_cast_fp16)[name = string("linear_56_cast_fp16")];
            tensor<fp16, [512]> pre_transformer_layers_7_mlp_layer_scale_scale_to_fp16 = const()[name = string("pre_transformer_layers_7_mlp_layer_scale_scale_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40230464)))];
            tensor<fp16, [1, 125, 512]> var_2608_cast_fp16 = mul(x = pre_transformer_layers_7_mlp_layer_scale_scale_to_fp16, y = linear_56_cast_fp16)[name = string("op_2608_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_97_cast_fp16 = add(x = hidden_states_91_cast_fp16, y = var_2608_cast_fp16)[name = string("hidden_states_97_cast_fp16")];
            fp16 var_2614_promoted_to_fp16 = const()[name = string("op_2614_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 125, 512]> var_2620_cast_fp16 = pow(x = hidden_states_97_cast_fp16, y = var_2614_promoted_to_fp16)[name = string("op_2620_cast_fp16")];
            tensor<int32, [1]> variance_axes_0 = const()[name = string("variance_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_keep_dims_0 = const()[name = string("variance_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 125, 1]> variance_cast_fp16 = reduce_mean(axes = variance_axes_0, keep_dims = variance_keep_dims_0, x = var_2620_cast_fp16)[name = string("variance_cast_fp16")];
            fp16 var_2623_to_fp16 = const()[name = string("op_2623_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 125, 1]> var_2624_cast_fp16 = add(x = variance_cast_fp16, y = var_2623_to_fp16)[name = string("op_2624_cast_fp16")];
            fp32 var_2625_epsilon_0 = const()[name = string("op_2625_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 125, 1]> var_2625_cast_fp16 = rsqrt(epsilon = var_2625_epsilon_0, x = var_2624_cast_fp16)[name = string("op_2625_cast_fp16")];
            tensor<fp16, [1, 125, 512]> hidden_states_101_cast_fp16 = mul(x = hidden_states_97_cast_fp16, y = var_2625_cast_fp16)[name = string("hidden_states_101_cast_fp16")];
            tensor<fp16, [512]> const_92_to_fp16 = const()[name = string("const_92_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40231552)))];
            tensor<fp16, [1, 125, 512]> input_105_cast_fp16 = mul(x = const_92_to_fp16, y = hidden_states_101_cast_fp16)[name = string("input_105_cast_fp16")];
            tensor<fp16, [1024, 512]> pre_transformer_output_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40232640))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40756992))))[name = string("pre_transformer_output_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> pre_transformer_output_proj_bias_to_fp16 = const()[name = string("pre_transformer_output_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40757568)))];
            tensor<fp16, [1, 125, 1024]> linear_57_cast_fp16 = linear(bias = pre_transformer_output_proj_bias_to_fp16, weight = pre_transformer_output_proj_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = string("linear_57_cast_fp16")];
            tensor<int32, [3]> var_2635 = const()[name = string("op_2635"), val = tensor<int32, [3]>([0, 2, 1])];
            string hidden_state_3_pad_type_0 = const()[name = string("hidden_state_3_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> hidden_state_3_strides_0 = const()[name = string("hidden_state_3_strides_0"), val = tensor<int32, [1]>([2])];
            tensor<int32, [2]> hidden_state_3_pad_0 = const()[name = string("hidden_state_3_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> hidden_state_3_dilations_0 = const()[name = string("hidden_state_3_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 hidden_state_3_groups_0 = const()[name = string("hidden_state_3_groups_0"), val = int32(1)];
            tensor<int32, [3]> hidden_state_3_has_output_shape_output_shape_0 = const()[name = string("hidden_state_3_has_output_shape_output_shape_0"), val = tensor<int32, [3]>([1, 1024, 250])];
            tensor<fp16, [1024, 1024, 2]> upsample_0_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 1024, 2]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40759680))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42856896))))[name = string("upsample_0_0_conv_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> upsample_0_0_conv_bias_to_fp16 = const()[name = string("upsample_0_0_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42857472)))];
            tensor<fp16, [1, 1024, 125]> input_107_cast_fp16 = transpose(perm = var_2635, x = linear_57_cast_fp16)[name = string("transpose_4")];
            tensor<fp16, [1, 1024, 250]> hidden_state_3_has_output_shape_cast_fp16 = conv_transpose(bias = upsample_0_0_conv_bias_to_fp16, dilations = hidden_state_3_dilations_0, groups = hidden_state_3_groups_0, output_shape = hidden_state_3_has_output_shape_output_shape_0, pad = hidden_state_3_pad_0, pad_type = hidden_state_3_pad_type_0, strides = hidden_state_3_strides_0, weight = upsample_0_0_conv_weight_to_fp16_palettized, x = input_107_cast_fp16)[name = string("hidden_state_3_has_output_shape_cast_fp16")];
            tensor<int32, [6]> input_109_pad_0 = const()[name = string("input_109_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string input_109_mode_0 = const()[name = string("input_109_mode_0"), val = string("constant")];
            fp16 const_94_to_fp16 = const()[name = string("const_94_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 256]> input_109_cast_fp16 = pad(constant_val = const_94_to_fp16, mode = input_109_mode_0, pad = input_109_pad_0, x = hidden_state_3_has_output_shape_cast_fp16)[name = string("input_109_cast_fp16")];
            string var_2681_pad_type_0 = const()[name = string("op_2681_pad_type_0"), val = string("valid")];
            int32 var_2681_groups_0 = const()[name = string("op_2681_groups_0"), val = int32(1024)];
            tensor<int32, [1]> var_2681_strides_0 = const()[name = string("op_2681_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2681_pad_0 = const()[name = string("op_2681_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_2681_dilations_0 = const()[name = string("op_2681_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 7]> upsample_0_1_dwconv_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42859584))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42866816))))[name = string("upsample_0_1_dwconv_conv_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> upsample_0_1_dwconv_conv_bias_to_fp16 = const()[name = string("upsample_0_1_dwconv_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42867392)))];
            tensor<fp16, [1, 1024, 250]> var_2681_cast_fp16 = conv(bias = upsample_0_1_dwconv_conv_bias_to_fp16, dilations = var_2681_dilations_0, groups = var_2681_groups_0, pad = var_2681_pad_0, pad_type = var_2681_pad_type_0, strides = var_2681_strides_0, weight = upsample_0_1_dwconv_conv_weight_to_fp16_palettized, x = input_109_cast_fp16)[name = string("op_2681_cast_fp16")];
            tensor<int32, [3]> var_2683 = const()[name = string("op_2683"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_113_axes_0 = const()[name = string("input_113_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> upsample_0_1_norm_weight_to_fp16 = const()[name = string("upsample_0_1_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42869504)))];
            tensor<fp16, [1024]> upsample_0_1_norm_bias_to_fp16 = const()[name = string("upsample_0_1_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42871616)))];
            fp16 var_2652_to_fp16 = const()[name = string("op_2652_to_fp16"), val = fp16(0x1.1p-20)];
            tensor<fp16, [1, 250, 1024]> input_111_cast_fp16 = transpose(perm = var_2683, x = var_2681_cast_fp16)[name = string("transpose_3")];
            tensor<fp16, [1, 250, 1024]> input_113_cast_fp16 = layer_norm(axes = input_113_axes_0, beta = upsample_0_1_norm_bias_to_fp16, epsilon = var_2652_to_fp16, gamma = upsample_0_1_norm_weight_to_fp16, x = input_111_cast_fp16)[name = string("input_113_cast_fp16")];
            tensor<fp16, [4096, 1024]> upsample_0_1_pwconv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42873728))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47068096))))[name = string("upsample_0_1_pwconv1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> upsample_0_1_pwconv1_bias_to_fp16 = const()[name = string("upsample_0_1_pwconv1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47068672)))];
            tensor<fp16, [1, 250, 4096]> linear_58_cast_fp16 = linear(bias = upsample_0_1_pwconv1_bias_to_fp16, weight = upsample_0_1_pwconv1_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = string("linear_58_cast_fp16")];
            string input_117_mode_0 = const()[name = string("input_117_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 250, 4096]> input_117_cast_fp16 = gelu(mode = input_117_mode_0, x = linear_58_cast_fp16)[name = string("input_117_cast_fp16")];
            tensor<fp16, [1024, 4096]> upsample_0_1_pwconv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47076928))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51271296))))[name = string("upsample_0_1_pwconv2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> upsample_0_1_pwconv2_bias_to_fp16 = const()[name = string("upsample_0_1_pwconv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51271872)))];
            tensor<fp16, [1, 250, 1024]> linear_59_cast_fp16 = linear(bias = upsample_0_1_pwconv2_bias_to_fp16, weight = upsample_0_1_pwconv2_weight_to_fp16_palettized, x = input_117_cast_fp16)[name = string("linear_59_cast_fp16")];
            tensor<fp16, [1024]> upsample_0_1_gamma_to_fp16 = const()[name = string("upsample_0_1_gamma_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51273984)))];
            tensor<fp16, [1, 250, 1024]> hidden_states_107_cast_fp16 = mul(x = upsample_0_1_gamma_to_fp16, y = linear_59_cast_fp16)[name = string("hidden_states_107_cast_fp16")];
            tensor<int32, [3]> var_2697 = const()[name = string("op_2697"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 1024, 250]> hidden_states_109_cast_fp16 = transpose(perm = var_2697, x = hidden_states_107_cast_fp16)[name = string("transpose_2")];
            tensor<fp16, [1, 1024, 250]> input_119_cast_fp16 = add(x = hidden_state_3_has_output_shape_cast_fp16, y = hidden_states_109_cast_fp16)[name = string("input_119_cast_fp16")];
            string hidden_state_7_pad_type_0 = const()[name = string("hidden_state_7_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> hidden_state_7_strides_0 = const()[name = string("hidden_state_7_strides_0"), val = tensor<int32, [1]>([2])];
            tensor<int32, [2]> hidden_state_7_pad_0 = const()[name = string("hidden_state_7_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> hidden_state_7_dilations_0 = const()[name = string("hidden_state_7_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 hidden_state_7_groups_0 = const()[name = string("hidden_state_7_groups_0"), val = int32(1)];
            tensor<int32, [3]> hidden_state_7_has_output_shape_output_shape_0 = const()[name = string("hidden_state_7_has_output_shape_output_shape_0"), val = tensor<int32, [3]>([1, 1024, 500])];
            tensor<fp16, [1024, 1024, 2]> upsample_1_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 1024, 2]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51276096))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53373312))))[name = string("upsample_1_0_conv_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> upsample_1_0_conv_bias_to_fp16 = const()[name = string("upsample_1_0_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53373888)))];
            tensor<fp16, [1, 1024, 500]> hidden_state_7_has_output_shape_cast_fp16 = conv_transpose(bias = upsample_1_0_conv_bias_to_fp16, dilations = hidden_state_7_dilations_0, groups = hidden_state_7_groups_0, output_shape = hidden_state_7_has_output_shape_output_shape_0, pad = hidden_state_7_pad_0, pad_type = hidden_state_7_pad_type_0, strides = hidden_state_7_strides_0, weight = upsample_1_0_conv_weight_to_fp16_palettized, x = input_119_cast_fp16)[name = string("hidden_state_7_has_output_shape_cast_fp16")];
            tensor<int32, [6]> input_121_pad_0 = const()[name = string("input_121_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string input_121_mode_0 = const()[name = string("input_121_mode_0"), val = string("constant")];
            fp16 const_96_to_fp16 = const()[name = string("const_96_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 506]> input_121_cast_fp16 = pad(constant_val = const_96_to_fp16, mode = input_121_mode_0, pad = input_121_pad_0, x = hidden_state_7_has_output_shape_cast_fp16)[name = string("input_121_cast_fp16")];
            string var_2744_pad_type_0 = const()[name = string("op_2744_pad_type_0"), val = string("valid")];
            int32 var_2744_groups_0 = const()[name = string("op_2744_groups_0"), val = int32(1024)];
            tensor<int32, [1]> var_2744_strides_0 = const()[name = string("op_2744_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2744_pad_0 = const()[name = string("op_2744_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_2744_dilations_0 = const()[name = string("op_2744_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 7]> upsample_1_1_dwconv_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53376000))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53383232))))[name = string("upsample_1_1_dwconv_conv_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> upsample_1_1_dwconv_conv_bias_to_fp16 = const()[name = string("upsample_1_1_dwconv_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53383808)))];
            tensor<fp16, [1, 1024, 500]> var_2744_cast_fp16 = conv(bias = upsample_1_1_dwconv_conv_bias_to_fp16, dilations = var_2744_dilations_0, groups = var_2744_groups_0, pad = var_2744_pad_0, pad_type = var_2744_pad_type_0, strides = var_2744_strides_0, weight = upsample_1_1_dwconv_conv_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = string("op_2744_cast_fp16")];
            tensor<int32, [3]> var_2746 = const()[name = string("op_2746"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_125_axes_0 = const()[name = string("input_125_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> upsample_1_1_norm_weight_to_fp16 = const()[name = string("upsample_1_1_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53385920)))];
            tensor<fp16, [1024]> upsample_1_1_norm_bias_to_fp16 = const()[name = string("upsample_1_1_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53388032)))];
            fp16 var_2715_to_fp16 = const()[name = string("op_2715_to_fp16"), val = fp16(0x1.1p-20)];
            tensor<fp16, [1, 500, 1024]> input_123_cast_fp16 = transpose(perm = var_2746, x = var_2744_cast_fp16)[name = string("transpose_1")];
            tensor<fp16, [1, 500, 1024]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = upsample_1_1_norm_bias_to_fp16, epsilon = var_2715_to_fp16, gamma = upsample_1_1_norm_weight_to_fp16, x = input_123_cast_fp16)[name = string("input_125_cast_fp16")];
            tensor<fp16, [4096, 1024]> upsample_1_1_pwconv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53390144))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57584512))))[name = string("upsample_1_1_pwconv1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> upsample_1_1_pwconv1_bias_to_fp16 = const()[name = string("upsample_1_1_pwconv1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57585088)))];
            tensor<fp16, [1, 500, 4096]> linear_60_cast_fp16 = linear(bias = upsample_1_1_pwconv1_bias_to_fp16, weight = upsample_1_1_pwconv1_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = string("linear_60_cast_fp16")];
            string input_129_mode_0 = const()[name = string("input_129_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 500, 4096]> input_129_cast_fp16 = gelu(mode = input_129_mode_0, x = linear_60_cast_fp16)[name = string("input_129_cast_fp16")];
            tensor<fp16, [1024, 4096]> upsample_1_1_pwconv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57593344))), lut = tensor<fp16, [1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61787712))))[name = string("upsample_1_1_pwconv2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> upsample_1_1_pwconv2_bias_to_fp16 = const()[name = string("upsample_1_1_pwconv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61788288)))];
            tensor<fp16, [1, 500, 1024]> linear_61_cast_fp16 = linear(bias = upsample_1_1_pwconv2_bias_to_fp16, weight = upsample_1_1_pwconv2_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = string("linear_61_cast_fp16")];
            tensor<fp16, [1024]> upsample_1_1_gamma_to_fp16 = const()[name = string("upsample_1_1_gamma_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61790400)))];
            tensor<fp16, [1, 500, 1024]> hidden_states_115_cast_fp16 = mul(x = upsample_1_1_gamma_to_fp16, y = linear_61_cast_fp16)[name = string("hidden_states_115_cast_fp16")];
            tensor<int32, [3]> var_2760 = const()[name = string("op_2760"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 1024, 500]> hidden_states_117_cast_fp16 = transpose(perm = var_2760, x = hidden_states_115_cast_fp16)[name = string("transpose_0")];
            tensor<fp16, [1, 1024, 500]> hidden_state_11_cast_fp16 = add(x = hidden_state_7_has_output_shape_cast_fp16, y = hidden_states_117_cast_fp16)[name = string("hidden_state_11_cast_fp16")];
            tensor<int32, [6]> input_131_pad_0 = const()[name = string("input_131_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string input_131_mode_0 = const()[name = string("input_131_mode_0"), val = string("constant")];
            fp16 const_98_to_fp16 = const()[name = string("const_98_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 506]> input_131_cast_fp16 = pad(constant_val = const_98_to_fp16, mode = input_131_mode_0, pad = input_131_pad_0, x = hidden_state_11_cast_fp16)[name = string("input_131_cast_fp16")];
            string var_2785_pad_type_0 = const()[name = string("op_2785_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_2785_strides_0 = const()[name = string("op_2785_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2785_pad_0 = const()[name = string("op_2785_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_2785_dilations_0 = const()[name = string("op_2785_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_2785_groups_0 = const()[name = string("op_2785_groups_0"), val = int32(1)];
            tensor<fp16, [1536, 1024, 7]> audio_decoder_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1536, 1024, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61792512))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72802624))))[name = string("audio_decoder_0_conv_weight_to_fp16_palettized")];
            tensor<fp16, [1536]> audio_decoder_0_conv_bias_to_fp16 = const()[name = string("audio_decoder_0_conv_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72803200)))];
            tensor<fp16, [1, 1536, 500]> var_2785_cast_fp16 = conv(bias = audio_decoder_0_conv_bias_to_fp16, dilations = var_2785_dilations_0, groups = var_2785_groups_0, pad = var_2785_pad_0, pad_type = var_2785_pad_type_0, strides = var_2785_strides_0, weight = audio_decoder_0_conv_weight_to_fp16_palettized, x = input_131_cast_fp16)[name = string("op_2785_cast_fp16")];
            tensor<fp16, [1, 1536, 1]> alpha_3_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 1536, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72806336))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72807936))))[name = string("alpha_3_to_fp16_palettized")];
            tensor<fp16, [1, 1536, 500]> var_2825_cast_fp16 = mul(x = var_2785_cast_fp16, y = alpha_3_to_fp16_palettized)[name = string("op_2825_cast_fp16")];
            tensor<fp16, [1, 1536, 500]> var_2826_cast_fp16 = sin(x = var_2825_cast_fp16)[name = string("op_2826_cast_fp16")];
            fp16 var_2801_promoted_to_fp16 = const()[name = string("op_2801_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 1536, 500]> var_2827_cast_fp16 = pow(x = var_2826_cast_fp16, y = var_2801_promoted_to_fp16)[name = string("op_2827_cast_fp16")];
            tensor<fp16, [1, 1536, 1]> op_2822_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 1536, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72808512))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72810112))))[name = string("op_2822_to_fp16_palettized")];
            tensor<fp16, [1, 1536, 500]> var_2828_cast_fp16 = mul(x = op_2822_to_fp16_palettized, y = var_2827_cast_fp16)[name = string("op_2828_cast_fp16")];
            tensor<fp16, [1, 1536, 500]> input_133_cast_fp16 = add(x = var_2785_cast_fp16, y = var_2828_cast_fp16)[name = string("input_133_cast_fp16")];
            string hidden_state_13_pad_type_0 = const()[name = string("hidden_state_13_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> hidden_state_13_strides_0 = const()[name = string("hidden_state_13_strides_0"), val = tensor<int32, [1]>([8])];
            tensor<int32, [2]> hidden_state_13_pad_0 = const()[name = string("hidden_state_13_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> hidden_state_13_dilations_0 = const()[name = string("hidden_state_13_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 hidden_state_13_groups_0 = const()[name = string("hidden_state_13_groups_0"), val = int32(1)];
            tensor<int32, [3]> hidden_state_13_has_output_shape_output_shape_0 = const()[name = string("hidden_state_13_has_output_shape_output_shape_0"), val = tensor<int32, [3]>([1, 768, 4008])];
            tensor<fp16, [1536, 768, 16]> audio_decoder_1_block_1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1536, 768, 16]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72810688))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91685120))))[name = string("audio_decoder_1_block_1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [768]> audio_decoder_1_block_1_conv_bias_to_fp16 = const()[name = string("audio_decoder_1_block_1_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91685696)))];
            tensor<fp16, [1, 768, 4008]> hidden_state_13_has_output_shape_cast_fp16 = conv_transpose(bias = audio_decoder_1_block_1_conv_bias_to_fp16, dilations = hidden_state_13_dilations_0, groups = hidden_state_13_groups_0, output_shape = hidden_state_13_has_output_shape_output_shape_0, pad = hidden_state_13_pad_0, pad_type = hidden_state_13_pad_type_0, strides = hidden_state_13_strides_0, weight = audio_decoder_1_block_1_conv_weight_to_fp16_palettized, x = input_133_cast_fp16)[name = string("hidden_state_13_has_output_shape_cast_fp16")];
            tensor<int32, [3]> hidden_state_15_begin_0 = const()[name = string("hidden_state_15_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> hidden_state_15_end_0 = const()[name = string("hidden_state_15_end_0"), val = tensor<int32, [3]>([1, 768, 4000])];
            tensor<bool, [3]> hidden_state_15_end_mask_0 = const()[name = string("hidden_state_15_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [1, 768, 4000]> hidden_state_15_cast_fp16 = slice_by_index(begin = hidden_state_15_begin_0, end = hidden_state_15_end_0, end_mask = hidden_state_15_end_mask_0, x = hidden_state_13_has_output_shape_cast_fp16)[name = string("hidden_state_15_cast_fp16")];
            tensor<fp16, [1, 768, 1]> alpha_7_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91687296))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91688128))))[name = string("alpha_7_to_fp16_palettized")];
            tensor<fp16, [1, 768, 4000]> var_2862_cast_fp16 = mul(x = hidden_state_15_cast_fp16, y = alpha_7_to_fp16_palettized)[name = string("op_2862_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> var_2863_cast_fp16 = sin(x = var_2862_cast_fp16)[name = string("op_2863_cast_fp16")];
            fp16 var_2801_promoted_1_to_fp16 = const()[name = string("op_2801_promoted_1_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 768, 4000]> var_2864_cast_fp16 = pow(x = var_2863_cast_fp16, y = var_2801_promoted_1_to_fp16)[name = string("op_2864_cast_fp16")];
            tensor<fp16, [1, 768, 1]> op_2859_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91688704))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91689536))))[name = string("op_2859_to_fp16_palettized")];
            tensor<fp16, [1, 768, 4000]> var_2865_cast_fp16 = mul(x = op_2859_to_fp16_palettized, y = var_2864_cast_fp16)[name = string("op_2865_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> hidden_state_17_cast_fp16 = add(x = hidden_state_15_cast_fp16, y = var_2865_cast_fp16)[name = string("hidden_state_17_cast_fp16")];
            tensor<int32, [6]> input_135_pad_0 = const()[name = string("input_135_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string input_135_mode_0 = const()[name = string("input_135_mode_0"), val = string("constant")];
            fp16 const_101_to_fp16 = const()[name = string("const_101_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 4006]> input_135_cast_fp16 = pad(constant_val = const_101_to_fp16, mode = input_135_mode_0, pad = input_135_pad_0, x = hidden_state_17_cast_fp16)[name = string("input_135_cast_fp16")];
            string var_2880_pad_type_0 = const()[name = string("op_2880_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_2880_strides_0 = const()[name = string("op_2880_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2880_pad_0 = const()[name = string("op_2880_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_2880_dilations_0 = const()[name = string("op_2880_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_2880_groups_0 = const()[name = string("op_2880_groups_0"), val = int32(1)];
            tensor<fp16, [768, 768, 7]> audio_decoder_1_block_2_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [768, 768, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91690112))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95818944))))[name = string("audio_decoder_1_block_2_conv1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [768]> audio_decoder_1_block_2_conv1_conv_bias_to_fp16 = const()[name = string("audio_decoder_1_block_2_conv1_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95819520)))];
            tensor<fp16, [1, 768, 4000]> var_2880_cast_fp16 = conv(bias = audio_decoder_1_block_2_conv1_conv_bias_to_fp16, dilations = var_2880_dilations_0, groups = var_2880_groups_0, pad = var_2880_pad_0, pad_type = var_2880_pad_type_0, strides = var_2880_strides_0, weight = audio_decoder_1_block_2_conv1_conv_weight_to_fp16_palettized, x = input_135_cast_fp16)[name = string("op_2880_cast_fp16")];
            tensor<fp16, [1, 768, 1]> alpha_11_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95821120))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95821952))))[name = string("alpha_11_to_fp16_palettized")];
            tensor<fp16, [1, 768, 4000]> var_2895_cast_fp16 = mul(x = var_2880_cast_fp16, y = alpha_11_to_fp16_palettized)[name = string("op_2895_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> var_2896_cast_fp16 = sin(x = var_2895_cast_fp16)[name = string("op_2896_cast_fp16")];
            fp16 var_2801_promoted_2_to_fp16 = const()[name = string("op_2801_promoted_2_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 768, 4000]> var_2897_cast_fp16 = pow(x = var_2896_cast_fp16, y = var_2801_promoted_2_to_fp16)[name = string("op_2897_cast_fp16")];
            tensor<fp16, [1, 768, 1]> op_2892_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95822528))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95823360))))[name = string("op_2892_to_fp16_palettized")];
            tensor<fp16, [1, 768, 4000]> var_2898_cast_fp16 = mul(x = op_2892_to_fp16_palettized, y = var_2897_cast_fp16)[name = string("op_2898_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> hidden_state_19_cast_fp16 = add(x = var_2880_cast_fp16, y = var_2898_cast_fp16)[name = string("hidden_state_19_cast_fp16")];
            string var_2913_pad_type_0 = const()[name = string("op_2913_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_2913_strides_0 = const()[name = string("op_2913_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2913_pad_0 = const()[name = string("op_2913_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_2913_dilations_0 = const()[name = string("op_2913_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_2913_groups_0 = const()[name = string("op_2913_groups_0"), val = int32(1)];
            tensor<fp16, [768, 768, 1]> audio_decoder_1_block_2_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [768, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95823936))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96413824))))[name = string("audio_decoder_1_block_2_conv2_conv_weight_to_fp16_palettized")];
            tensor<fp16, [768]> audio_decoder_1_block_2_conv2_conv_bias_to_fp16 = const()[name = string("audio_decoder_1_block_2_conv2_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96414400)))];
            tensor<fp16, [1, 768, 4000]> var_2913_cast_fp16 = conv(bias = audio_decoder_1_block_2_conv2_conv_bias_to_fp16, dilations = var_2913_dilations_0, groups = var_2913_groups_0, pad = var_2913_pad_0, pad_type = var_2913_pad_type_0, strides = var_2913_strides_0, weight = audio_decoder_1_block_2_conv2_conv_weight_to_fp16_palettized, x = hidden_state_19_cast_fp16)[name = string("op_2913_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> hidden_states_125_cast_fp16 = add(x = var_2913_cast_fp16, y = hidden_state_15_cast_fp16)[name = string("hidden_states_125_cast_fp16")];
            tensor<fp16, [1, 768, 1]> alpha_15_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96416000))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96416832))))[name = string("alpha_15_to_fp16_palettized")];
            tensor<fp16, [1, 768, 4000]> var_2933_cast_fp16 = mul(x = hidden_states_125_cast_fp16, y = alpha_15_to_fp16_palettized)[name = string("op_2933_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> var_2934_cast_fp16 = sin(x = var_2933_cast_fp16)[name = string("op_2934_cast_fp16")];
            fp16 var_2801_promoted_3_to_fp16 = const()[name = string("op_2801_promoted_3_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 768, 4000]> var_2935_cast_fp16 = pow(x = var_2934_cast_fp16, y = var_2801_promoted_3_to_fp16)[name = string("op_2935_cast_fp16")];
            tensor<fp16, [1, 768, 1]> op_2930_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96417408))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96418240))))[name = string("op_2930_to_fp16_palettized")];
            tensor<fp16, [1, 768, 4000]> var_2936_cast_fp16 = mul(x = op_2930_to_fp16_palettized, y = var_2935_cast_fp16)[name = string("op_2936_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> hidden_state_23_cast_fp16 = add(x = hidden_states_125_cast_fp16, y = var_2936_cast_fp16)[name = string("hidden_state_23_cast_fp16")];
            tensor<int32, [6]> input_139_pad_0 = const()[name = string("input_139_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 18, 0])];
            string input_139_mode_0 = const()[name = string("input_139_mode_0"), val = string("constant")];
            fp16 const_105_to_fp16 = const()[name = string("const_105_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 4018]> input_139_cast_fp16 = pad(constant_val = const_105_to_fp16, mode = input_139_mode_0, pad = input_139_pad_0, x = hidden_state_23_cast_fp16)[name = string("input_139_cast_fp16")];
            string var_2951_pad_type_0 = const()[name = string("op_2951_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_2951_dilations_0 = const()[name = string("op_2951_dilations_0"), val = tensor<int32, [1]>([3])];
            tensor<int32, [1]> var_2951_strides_0 = const()[name = string("op_2951_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2951_pad_0 = const()[name = string("op_2951_pad_0"), val = tensor<int32, [2]>([0, 0])];
            int32 var_2951_groups_0 = const()[name = string("op_2951_groups_0"), val = int32(1)];
            tensor<fp16, [768, 768, 7]> audio_decoder_1_block_3_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [768, 768, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96418816))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100547648))))[name = string("audio_decoder_1_block_3_conv1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [768]> audio_decoder_1_block_3_conv1_conv_bias_to_fp16 = const()[name = string("audio_decoder_1_block_3_conv1_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100548224)))];
            tensor<fp16, [1, 768, 4000]> var_2951_cast_fp16 = conv(bias = audio_decoder_1_block_3_conv1_conv_bias_to_fp16, dilations = var_2951_dilations_0, groups = var_2951_groups_0, pad = var_2951_pad_0, pad_type = var_2951_pad_type_0, strides = var_2951_strides_0, weight = audio_decoder_1_block_3_conv1_conv_weight_to_fp16_palettized, x = input_139_cast_fp16)[name = string("op_2951_cast_fp16")];
            tensor<fp16, [1, 768, 1]> alpha_19_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100549824))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100550656))))[name = string("alpha_19_to_fp16_palettized")];
            tensor<fp16, [1, 768, 4000]> var_2966_cast_fp16 = mul(x = var_2951_cast_fp16, y = alpha_19_to_fp16_palettized)[name = string("op_2966_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> var_2967_cast_fp16 = sin(x = var_2966_cast_fp16)[name = string("op_2967_cast_fp16")];
            fp16 var_2801_promoted_4_to_fp16 = const()[name = string("op_2801_promoted_4_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 768, 4000]> var_2968_cast_fp16 = pow(x = var_2967_cast_fp16, y = var_2801_promoted_4_to_fp16)[name = string("op_2968_cast_fp16")];
            tensor<fp16, [1, 768, 1]> op_2963_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100551232))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100552064))))[name = string("op_2963_to_fp16_palettized")];
            tensor<fp16, [1, 768, 4000]> var_2969_cast_fp16 = mul(x = op_2963_to_fp16_palettized, y = var_2968_cast_fp16)[name = string("op_2969_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> hidden_state_25_cast_fp16 = add(x = var_2951_cast_fp16, y = var_2969_cast_fp16)[name = string("hidden_state_25_cast_fp16")];
            string var_2984_pad_type_0 = const()[name = string("op_2984_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_2984_strides_0 = const()[name = string("op_2984_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2984_pad_0 = const()[name = string("op_2984_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_2984_dilations_0 = const()[name = string("op_2984_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_2984_groups_0 = const()[name = string("op_2984_groups_0"), val = int32(1)];
            tensor<fp16, [768, 768, 1]> audio_decoder_1_block_3_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [768, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100552640))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101142528))))[name = string("audio_decoder_1_block_3_conv2_conv_weight_to_fp16_palettized")];
            tensor<fp16, [768]> audio_decoder_1_block_3_conv2_conv_bias_to_fp16 = const()[name = string("audio_decoder_1_block_3_conv2_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101143104)))];
            tensor<fp16, [1, 768, 4000]> var_2984_cast_fp16 = conv(bias = audio_decoder_1_block_3_conv2_conv_bias_to_fp16, dilations = var_2984_dilations_0, groups = var_2984_groups_0, pad = var_2984_pad_0, pad_type = var_2984_pad_type_0, strides = var_2984_strides_0, weight = audio_decoder_1_block_3_conv2_conv_weight_to_fp16_palettized, x = hidden_state_25_cast_fp16)[name = string("op_2984_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> hidden_states_129_cast_fp16 = add(x = var_2984_cast_fp16, y = hidden_states_125_cast_fp16)[name = string("hidden_states_129_cast_fp16")];
            tensor<fp16, [1, 768, 1]> alpha_23_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101144704))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101145536))))[name = string("alpha_23_to_fp16_palettized")];
            tensor<fp16, [1, 768, 4000]> var_3004_cast_fp16 = mul(x = hidden_states_129_cast_fp16, y = alpha_23_to_fp16_palettized)[name = string("op_3004_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> var_3005_cast_fp16 = sin(x = var_3004_cast_fp16)[name = string("op_3005_cast_fp16")];
            fp16 var_2801_promoted_5_to_fp16 = const()[name = string("op_2801_promoted_5_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 768, 4000]> var_3006_cast_fp16 = pow(x = var_3005_cast_fp16, y = var_2801_promoted_5_to_fp16)[name = string("op_3006_cast_fp16")];
            tensor<fp16, [1, 768, 1]> op_3001_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101146112))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101146944))))[name = string("op_3001_to_fp16_palettized")];
            tensor<fp16, [1, 768, 4000]> var_3007_cast_fp16 = mul(x = op_3001_to_fp16_palettized, y = var_3006_cast_fp16)[name = string("op_3007_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> hidden_state_29_cast_fp16 = add(x = hidden_states_129_cast_fp16, y = var_3007_cast_fp16)[name = string("hidden_state_29_cast_fp16")];
            tensor<int32, [6]> input_143_pad_0 = const()[name = string("input_143_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 54, 0])];
            string input_143_mode_0 = const()[name = string("input_143_mode_0"), val = string("constant")];
            fp16 const_109_to_fp16 = const()[name = string("const_109_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 4054]> input_143_cast_fp16 = pad(constant_val = const_109_to_fp16, mode = input_143_mode_0, pad = input_143_pad_0, x = hidden_state_29_cast_fp16)[name = string("input_143_cast_fp16")];
            string var_3022_pad_type_0 = const()[name = string("op_3022_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3022_dilations_0 = const()[name = string("op_3022_dilations_0"), val = tensor<int32, [1]>([9])];
            tensor<int32, [1]> var_3022_strides_0 = const()[name = string("op_3022_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3022_pad_0 = const()[name = string("op_3022_pad_0"), val = tensor<int32, [2]>([0, 0])];
            int32 var_3022_groups_0 = const()[name = string("op_3022_groups_0"), val = int32(1)];
            tensor<fp16, [768, 768, 7]> audio_decoder_1_block_4_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [768, 768, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101147520))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105276352))))[name = string("audio_decoder_1_block_4_conv1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [768]> audio_decoder_1_block_4_conv1_conv_bias_to_fp16 = const()[name = string("audio_decoder_1_block_4_conv1_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105276928)))];
            tensor<fp16, [1, 768, 4000]> var_3022_cast_fp16 = conv(bias = audio_decoder_1_block_4_conv1_conv_bias_to_fp16, dilations = var_3022_dilations_0, groups = var_3022_groups_0, pad = var_3022_pad_0, pad_type = var_3022_pad_type_0, strides = var_3022_strides_0, weight = audio_decoder_1_block_4_conv1_conv_weight_to_fp16_palettized, x = input_143_cast_fp16)[name = string("op_3022_cast_fp16")];
            tensor<fp16, [1, 768, 1]> alpha_27_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105278528))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105279360))))[name = string("alpha_27_to_fp16_palettized")];
            tensor<fp16, [1, 768, 4000]> var_3037_cast_fp16 = mul(x = var_3022_cast_fp16, y = alpha_27_to_fp16_palettized)[name = string("op_3037_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> var_3038_cast_fp16 = sin(x = var_3037_cast_fp16)[name = string("op_3038_cast_fp16")];
            fp16 var_2801_promoted_6_to_fp16 = const()[name = string("op_2801_promoted_6_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 768, 4000]> var_3039_cast_fp16 = pow(x = var_3038_cast_fp16, y = var_2801_promoted_6_to_fp16)[name = string("op_3039_cast_fp16")];
            tensor<fp16, [1, 768, 1]> op_3034_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105279936))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105280768))))[name = string("op_3034_to_fp16_palettized")];
            tensor<fp16, [1, 768, 4000]> var_3040_cast_fp16 = mul(x = op_3034_to_fp16_palettized, y = var_3039_cast_fp16)[name = string("op_3040_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> hidden_state_31_cast_fp16 = add(x = var_3022_cast_fp16, y = var_3040_cast_fp16)[name = string("hidden_state_31_cast_fp16")];
            string var_3055_pad_type_0 = const()[name = string("op_3055_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3055_strides_0 = const()[name = string("op_3055_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3055_pad_0 = const()[name = string("op_3055_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_3055_dilations_0 = const()[name = string("op_3055_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_3055_groups_0 = const()[name = string("op_3055_groups_0"), val = int32(1)];
            tensor<fp16, [768, 768, 1]> audio_decoder_1_block_4_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [768, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105281344))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105871232))))[name = string("audio_decoder_1_block_4_conv2_conv_weight_to_fp16_palettized")];
            tensor<fp16, [768]> audio_decoder_1_block_4_conv2_conv_bias_to_fp16 = const()[name = string("audio_decoder_1_block_4_conv2_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105871808)))];
            tensor<fp16, [1, 768, 4000]> var_3055_cast_fp16 = conv(bias = audio_decoder_1_block_4_conv2_conv_bias_to_fp16, dilations = var_3055_dilations_0, groups = var_3055_groups_0, pad = var_3055_pad_0, pad_type = var_3055_pad_type_0, strides = var_3055_strides_0, weight = audio_decoder_1_block_4_conv2_conv_weight_to_fp16_palettized, x = hidden_state_31_cast_fp16)[name = string("op_3055_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> hidden_states_133_cast_fp16 = add(x = var_3055_cast_fp16, y = hidden_states_129_cast_fp16)[name = string("hidden_states_133_cast_fp16")];
            tensor<fp16, [1, 768, 1]> alpha_31_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105873408))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105874240))))[name = string("alpha_31_to_fp16_palettized")];
            tensor<fp16, [1, 768, 4000]> var_3096_cast_fp16 = mul(x = hidden_states_133_cast_fp16, y = alpha_31_to_fp16_palettized)[name = string("op_3096_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> var_3097_cast_fp16 = sin(x = var_3096_cast_fp16)[name = string("op_3097_cast_fp16")];
            fp16 var_3072_promoted_to_fp16 = const()[name = string("op_3072_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 768, 4000]> var_3098_cast_fp16 = pow(x = var_3097_cast_fp16, y = var_3072_promoted_to_fp16)[name = string("op_3098_cast_fp16")];
            tensor<fp16, [1, 768, 1]> op_3093_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 768, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105874816))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105875648))))[name = string("op_3093_to_fp16_palettized")];
            tensor<fp16, [1, 768, 4000]> var_3099_cast_fp16 = mul(x = op_3093_to_fp16_palettized, y = var_3098_cast_fp16)[name = string("op_3099_cast_fp16")];
            tensor<fp16, [1, 768, 4000]> input_147_cast_fp16 = add(x = hidden_states_133_cast_fp16, y = var_3099_cast_fp16)[name = string("input_147_cast_fp16")];
            string hidden_state_35_pad_type_0 = const()[name = string("hidden_state_35_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> hidden_state_35_strides_0 = const()[name = string("hidden_state_35_strides_0"), val = tensor<int32, [1]>([5])];
            tensor<int32, [2]> hidden_state_35_pad_0 = const()[name = string("hidden_state_35_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> hidden_state_35_dilations_0 = const()[name = string("hidden_state_35_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 hidden_state_35_groups_0 = const()[name = string("hidden_state_35_groups_0"), val = int32(1)];
            tensor<int32, [3]> hidden_state_35_has_output_shape_output_shape_0 = const()[name = string("hidden_state_35_has_output_shape_output_shape_0"), val = tensor<int32, [3]>([1, 384, 20005])];
            tensor<fp16, [768, 384, 10]> audio_decoder_2_block_1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [768, 384, 10]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105876224))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108825408))))[name = string("audio_decoder_2_block_1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [384]> audio_decoder_2_block_1_conv_bias_to_fp16 = const()[name = string("audio_decoder_2_block_1_conv_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108825984)))];
            tensor<fp16, [1, 384, 20005]> hidden_state_35_has_output_shape_cast_fp16 = conv_transpose(bias = audio_decoder_2_block_1_conv_bias_to_fp16, dilations = hidden_state_35_dilations_0, groups = hidden_state_35_groups_0, output_shape = hidden_state_35_has_output_shape_output_shape_0, pad = hidden_state_35_pad_0, pad_type = hidden_state_35_pad_type_0, strides = hidden_state_35_strides_0, weight = audio_decoder_2_block_1_conv_weight_to_fp16_palettized, x = input_147_cast_fp16)[name = string("hidden_state_35_has_output_shape_cast_fp16")];
            tensor<int32, [3]> hidden_state_37_begin_0 = const()[name = string("hidden_state_37_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> hidden_state_37_end_0 = const()[name = string("hidden_state_37_end_0"), val = tensor<int32, [3]>([1, 384, 20000])];
            tensor<bool, [3]> hidden_state_37_end_mask_0 = const()[name = string("hidden_state_37_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [1, 384, 20000]> hidden_state_37_cast_fp16 = slice_by_index(begin = hidden_state_37_begin_0, end = hidden_state_37_end_0, end_mask = hidden_state_37_end_mask_0, x = hidden_state_35_has_output_shape_cast_fp16)[name = string("hidden_state_37_cast_fp16")];
            tensor<fp16, [1, 384, 1]> alpha_35_to_fp16 = const()[name = string("alpha_35_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108826816)))];
            tensor<fp16, [1, 384, 20000]> var_3133_cast_fp16 = mul(x = hidden_state_37_cast_fp16, y = alpha_35_to_fp16)[name = string("op_3133_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> var_3134_cast_fp16 = sin(x = var_3133_cast_fp16)[name = string("op_3134_cast_fp16")];
            fp16 var_3072_promoted_1_to_fp16 = const()[name = string("op_3072_promoted_1_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 384, 20000]> var_3135_cast_fp16 = pow(x = var_3134_cast_fp16, y = var_3072_promoted_1_to_fp16)[name = string("op_3135_cast_fp16")];
            tensor<fp16, [1, 384, 1]> var_3130_to_fp16 = const()[name = string("op_3130_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108827648)))];
            tensor<fp16, [1, 384, 20000]> var_3136_cast_fp16 = mul(x = var_3130_to_fp16, y = var_3135_cast_fp16)[name = string("op_3136_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> hidden_state_39_cast_fp16 = add(x = hidden_state_37_cast_fp16, y = var_3136_cast_fp16)[name = string("hidden_state_39_cast_fp16")];
            tensor<int32, [6]> input_149_pad_0 = const()[name = string("input_149_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string input_149_mode_0 = const()[name = string("input_149_mode_0"), val = string("constant")];
            fp16 const_114_to_fp16 = const()[name = string("const_114_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 384, 20006]> input_149_cast_fp16 = pad(constant_val = const_114_to_fp16, mode = input_149_mode_0, pad = input_149_pad_0, x = hidden_state_39_cast_fp16)[name = string("input_149_cast_fp16")];
            string var_3151_pad_type_0 = const()[name = string("op_3151_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3151_strides_0 = const()[name = string("op_3151_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3151_pad_0 = const()[name = string("op_3151_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_3151_dilations_0 = const()[name = string("op_3151_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_3151_groups_0 = const()[name = string("op_3151_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 7]> audio_decoder_2_block_2_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [384, 384, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108828480))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109860736))))[name = string("audio_decoder_2_block_2_conv1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [384]> audio_decoder_2_block_2_conv1_conv_bias_to_fp16 = const()[name = string("audio_decoder_2_block_2_conv1_conv_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109861312)))];
            tensor<fp16, [1, 384, 20000]> var_3151_cast_fp16 = conv(bias = audio_decoder_2_block_2_conv1_conv_bias_to_fp16, dilations = var_3151_dilations_0, groups = var_3151_groups_0, pad = var_3151_pad_0, pad_type = var_3151_pad_type_0, strides = var_3151_strides_0, weight = audio_decoder_2_block_2_conv1_conv_weight_to_fp16_palettized, x = input_149_cast_fp16)[name = string("op_3151_cast_fp16")];
            tensor<fp16, [1, 384, 1]> alpha_39_to_fp16 = const()[name = string("alpha_39_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109862144)))];
            tensor<fp16, [1, 384, 20000]> var_3166_cast_fp16 = mul(x = var_3151_cast_fp16, y = alpha_39_to_fp16)[name = string("op_3166_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> var_3167_cast_fp16 = sin(x = var_3166_cast_fp16)[name = string("op_3167_cast_fp16")];
            fp16 var_3072_promoted_2_to_fp16 = const()[name = string("op_3072_promoted_2_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 384, 20000]> var_3168_cast_fp16 = pow(x = var_3167_cast_fp16, y = var_3072_promoted_2_to_fp16)[name = string("op_3168_cast_fp16")];
            tensor<fp16, [1, 384, 1]> var_3163_to_fp16 = const()[name = string("op_3163_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109862976)))];
            tensor<fp16, [1, 384, 20000]> var_3169_cast_fp16 = mul(x = var_3163_to_fp16, y = var_3168_cast_fp16)[name = string("op_3169_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> hidden_state_41_cast_fp16 = add(x = var_3151_cast_fp16, y = var_3169_cast_fp16)[name = string("hidden_state_41_cast_fp16")];
            string var_3184_pad_type_0 = const()[name = string("op_3184_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3184_strides_0 = const()[name = string("op_3184_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3184_pad_0 = const()[name = string("op_3184_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_3184_dilations_0 = const()[name = string("op_3184_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_3184_groups_0 = const()[name = string("op_3184_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1]> audio_decoder_2_block_2_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [384, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109863808))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110011328))))[name = string("audio_decoder_2_block_2_conv2_conv_weight_to_fp16_palettized")];
            tensor<fp16, [384]> audio_decoder_2_block_2_conv2_conv_bias_to_fp16 = const()[name = string("audio_decoder_2_block_2_conv2_conv_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110011904)))];
            tensor<fp16, [1, 384, 20000]> var_3184_cast_fp16 = conv(bias = audio_decoder_2_block_2_conv2_conv_bias_to_fp16, dilations = var_3184_dilations_0, groups = var_3184_groups_0, pad = var_3184_pad_0, pad_type = var_3184_pad_type_0, strides = var_3184_strides_0, weight = audio_decoder_2_block_2_conv2_conv_weight_to_fp16_palettized, x = hidden_state_41_cast_fp16)[name = string("op_3184_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> hidden_states_139_cast_fp16 = add(x = var_3184_cast_fp16, y = hidden_state_37_cast_fp16)[name = string("hidden_states_139_cast_fp16")];
            tensor<fp16, [1, 384, 1]> alpha_43_to_fp16 = const()[name = string("alpha_43_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110012736)))];
            tensor<fp16, [1, 384, 20000]> var_3204_cast_fp16 = mul(x = hidden_states_139_cast_fp16, y = alpha_43_to_fp16)[name = string("op_3204_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> var_3205_cast_fp16 = sin(x = var_3204_cast_fp16)[name = string("op_3205_cast_fp16")];
            fp16 var_3072_promoted_3_to_fp16 = const()[name = string("op_3072_promoted_3_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 384, 20000]> var_3206_cast_fp16 = pow(x = var_3205_cast_fp16, y = var_3072_promoted_3_to_fp16)[name = string("op_3206_cast_fp16")];
            tensor<fp16, [1, 384, 1]> var_3201_to_fp16 = const()[name = string("op_3201_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110013568)))];
            tensor<fp16, [1, 384, 20000]> var_3207_cast_fp16 = mul(x = var_3201_to_fp16, y = var_3206_cast_fp16)[name = string("op_3207_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> hidden_state_45_cast_fp16 = add(x = hidden_states_139_cast_fp16, y = var_3207_cast_fp16)[name = string("hidden_state_45_cast_fp16")];
            tensor<int32, [6]> input_153_pad_0 = const()[name = string("input_153_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 18, 0])];
            string input_153_mode_0 = const()[name = string("input_153_mode_0"), val = string("constant")];
            fp16 const_118_to_fp16 = const()[name = string("const_118_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 384, 20018]> input_153_cast_fp16 = pad(constant_val = const_118_to_fp16, mode = input_153_mode_0, pad = input_153_pad_0, x = hidden_state_45_cast_fp16)[name = string("input_153_cast_fp16")];
            string var_3222_pad_type_0 = const()[name = string("op_3222_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3222_dilations_0 = const()[name = string("op_3222_dilations_0"), val = tensor<int32, [1]>([3])];
            tensor<int32, [1]> var_3222_strides_0 = const()[name = string("op_3222_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3222_pad_0 = const()[name = string("op_3222_pad_0"), val = tensor<int32, [2]>([0, 0])];
            int32 var_3222_groups_0 = const()[name = string("op_3222_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 7]> audio_decoder_2_block_3_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [384, 384, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110014400))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111046656))))[name = string("audio_decoder_2_block_3_conv1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [384]> audio_decoder_2_block_3_conv1_conv_bias_to_fp16 = const()[name = string("audio_decoder_2_block_3_conv1_conv_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111047232)))];
            tensor<fp16, [1, 384, 20000]> var_3222_cast_fp16 = conv(bias = audio_decoder_2_block_3_conv1_conv_bias_to_fp16, dilations = var_3222_dilations_0, groups = var_3222_groups_0, pad = var_3222_pad_0, pad_type = var_3222_pad_type_0, strides = var_3222_strides_0, weight = audio_decoder_2_block_3_conv1_conv_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = string("op_3222_cast_fp16")];
            tensor<fp16, [1, 384, 1]> alpha_47_to_fp16 = const()[name = string("alpha_47_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111048064)))];
            tensor<fp16, [1, 384, 20000]> var_3237_cast_fp16 = mul(x = var_3222_cast_fp16, y = alpha_47_to_fp16)[name = string("op_3237_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> var_3238_cast_fp16 = sin(x = var_3237_cast_fp16)[name = string("op_3238_cast_fp16")];
            fp16 var_3072_promoted_4_to_fp16 = const()[name = string("op_3072_promoted_4_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 384, 20000]> var_3239_cast_fp16 = pow(x = var_3238_cast_fp16, y = var_3072_promoted_4_to_fp16)[name = string("op_3239_cast_fp16")];
            tensor<fp16, [1, 384, 1]> var_3234_to_fp16 = const()[name = string("op_3234_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111048896)))];
            tensor<fp16, [1, 384, 20000]> var_3240_cast_fp16 = mul(x = var_3234_to_fp16, y = var_3239_cast_fp16)[name = string("op_3240_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> hidden_state_47_cast_fp16 = add(x = var_3222_cast_fp16, y = var_3240_cast_fp16)[name = string("hidden_state_47_cast_fp16")];
            string var_3255_pad_type_0 = const()[name = string("op_3255_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3255_strides_0 = const()[name = string("op_3255_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3255_pad_0 = const()[name = string("op_3255_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_3255_dilations_0 = const()[name = string("op_3255_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_3255_groups_0 = const()[name = string("op_3255_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1]> audio_decoder_2_block_3_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [384, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111049728))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111197248))))[name = string("audio_decoder_2_block_3_conv2_conv_weight_to_fp16_palettized")];
            tensor<fp16, [384]> audio_decoder_2_block_3_conv2_conv_bias_to_fp16 = const()[name = string("audio_decoder_2_block_3_conv2_conv_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111197824)))];
            tensor<fp16, [1, 384, 20000]> var_3255_cast_fp16 = conv(bias = audio_decoder_2_block_3_conv2_conv_bias_to_fp16, dilations = var_3255_dilations_0, groups = var_3255_groups_0, pad = var_3255_pad_0, pad_type = var_3255_pad_type_0, strides = var_3255_strides_0, weight = audio_decoder_2_block_3_conv2_conv_weight_to_fp16_palettized, x = hidden_state_47_cast_fp16)[name = string("op_3255_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> hidden_states_143_cast_fp16 = add(x = var_3255_cast_fp16, y = hidden_states_139_cast_fp16)[name = string("hidden_states_143_cast_fp16")];
            tensor<fp16, [1, 384, 1]> alpha_51_to_fp16 = const()[name = string("alpha_51_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111198656)))];
            tensor<fp16, [1, 384, 20000]> var_3275_cast_fp16 = mul(x = hidden_states_143_cast_fp16, y = alpha_51_to_fp16)[name = string("op_3275_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> var_3276_cast_fp16 = sin(x = var_3275_cast_fp16)[name = string("op_3276_cast_fp16")];
            fp16 var_3072_promoted_5_to_fp16 = const()[name = string("op_3072_promoted_5_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 384, 20000]> var_3277_cast_fp16 = pow(x = var_3276_cast_fp16, y = var_3072_promoted_5_to_fp16)[name = string("op_3277_cast_fp16")];
            tensor<fp16, [1, 384, 1]> var_3272_to_fp16 = const()[name = string("op_3272_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111199488)))];
            tensor<fp16, [1, 384, 20000]> var_3278_cast_fp16 = mul(x = var_3272_to_fp16, y = var_3277_cast_fp16)[name = string("op_3278_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> hidden_state_51_cast_fp16 = add(x = hidden_states_143_cast_fp16, y = var_3278_cast_fp16)[name = string("hidden_state_51_cast_fp16")];
            tensor<int32, [6]> input_157_pad_0 = const()[name = string("input_157_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 54, 0])];
            string input_157_mode_0 = const()[name = string("input_157_mode_0"), val = string("constant")];
            fp16 const_122_to_fp16 = const()[name = string("const_122_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 384, 20054]> input_157_cast_fp16 = pad(constant_val = const_122_to_fp16, mode = input_157_mode_0, pad = input_157_pad_0, x = hidden_state_51_cast_fp16)[name = string("input_157_cast_fp16")];
            string var_3293_pad_type_0 = const()[name = string("op_3293_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3293_dilations_0 = const()[name = string("op_3293_dilations_0"), val = tensor<int32, [1]>([9])];
            tensor<int32, [1]> var_3293_strides_0 = const()[name = string("op_3293_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3293_pad_0 = const()[name = string("op_3293_pad_0"), val = tensor<int32, [2]>([0, 0])];
            int32 var_3293_groups_0 = const()[name = string("op_3293_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 7]> audio_decoder_2_block_4_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [384, 384, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111200320))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112232576))))[name = string("audio_decoder_2_block_4_conv1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [384]> audio_decoder_2_block_4_conv1_conv_bias_to_fp16 = const()[name = string("audio_decoder_2_block_4_conv1_conv_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112233152)))];
            tensor<fp16, [1, 384, 20000]> var_3293_cast_fp16 = conv(bias = audio_decoder_2_block_4_conv1_conv_bias_to_fp16, dilations = var_3293_dilations_0, groups = var_3293_groups_0, pad = var_3293_pad_0, pad_type = var_3293_pad_type_0, strides = var_3293_strides_0, weight = audio_decoder_2_block_4_conv1_conv_weight_to_fp16_palettized, x = input_157_cast_fp16)[name = string("op_3293_cast_fp16")];
            tensor<fp16, [1, 384, 1]> alpha_55_to_fp16 = const()[name = string("alpha_55_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112233984)))];
            tensor<fp16, [1, 384, 20000]> var_3308_cast_fp16 = mul(x = var_3293_cast_fp16, y = alpha_55_to_fp16)[name = string("op_3308_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> var_3309_cast_fp16 = sin(x = var_3308_cast_fp16)[name = string("op_3309_cast_fp16")];
            fp16 var_3072_promoted_6_to_fp16 = const()[name = string("op_3072_promoted_6_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 384, 20000]> var_3310_cast_fp16 = pow(x = var_3309_cast_fp16, y = var_3072_promoted_6_to_fp16)[name = string("op_3310_cast_fp16")];
            tensor<fp16, [1, 384, 1]> var_3305_to_fp16 = const()[name = string("op_3305_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112234816)))];
            tensor<fp16, [1, 384, 20000]> var_3311_cast_fp16 = mul(x = var_3305_to_fp16, y = var_3310_cast_fp16)[name = string("op_3311_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> hidden_state_53_cast_fp16 = add(x = var_3293_cast_fp16, y = var_3311_cast_fp16)[name = string("hidden_state_53_cast_fp16")];
            string var_3326_pad_type_0 = const()[name = string("op_3326_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3326_strides_0 = const()[name = string("op_3326_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3326_pad_0 = const()[name = string("op_3326_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_3326_dilations_0 = const()[name = string("op_3326_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_3326_groups_0 = const()[name = string("op_3326_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1]> audio_decoder_2_block_4_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [384, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112235648))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112383168))))[name = string("audio_decoder_2_block_4_conv2_conv_weight_to_fp16_palettized")];
            tensor<fp16, [384]> audio_decoder_2_block_4_conv2_conv_bias_to_fp16 = const()[name = string("audio_decoder_2_block_4_conv2_conv_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112383744)))];
            tensor<fp16, [1, 384, 20000]> var_3326_cast_fp16 = conv(bias = audio_decoder_2_block_4_conv2_conv_bias_to_fp16, dilations = var_3326_dilations_0, groups = var_3326_groups_0, pad = var_3326_pad_0, pad_type = var_3326_pad_type_0, strides = var_3326_strides_0, weight = audio_decoder_2_block_4_conv2_conv_weight_to_fp16_palettized, x = hidden_state_53_cast_fp16)[name = string("op_3326_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> hidden_states_147_cast_fp16 = add(x = var_3326_cast_fp16, y = hidden_states_143_cast_fp16)[name = string("hidden_states_147_cast_fp16")];
            tensor<fp16, [1, 384, 1]> alpha_59_to_fp16 = const()[name = string("alpha_59_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112384576)))];
            tensor<fp16, [1, 384, 20000]> var_3367_cast_fp16 = mul(x = hidden_states_147_cast_fp16, y = alpha_59_to_fp16)[name = string("op_3367_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> var_3368_cast_fp16 = sin(x = var_3367_cast_fp16)[name = string("op_3368_cast_fp16")];
            fp16 var_3343_promoted_to_fp16 = const()[name = string("op_3343_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 384, 20000]> var_3369_cast_fp16 = pow(x = var_3368_cast_fp16, y = var_3343_promoted_to_fp16)[name = string("op_3369_cast_fp16")];
            tensor<fp16, [1, 384, 1]> var_3364_to_fp16 = const()[name = string("op_3364_to_fp16"), val = tensor<fp16, [1, 384, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112385408)))];
            tensor<fp16, [1, 384, 20000]> var_3370_cast_fp16 = mul(x = var_3364_to_fp16, y = var_3369_cast_fp16)[name = string("op_3370_cast_fp16")];
            tensor<fp16, [1, 384, 20000]> input_161_cast_fp16 = add(x = hidden_states_147_cast_fp16, y = var_3370_cast_fp16)[name = string("input_161_cast_fp16")];
            string hidden_state_57_pad_type_0 = const()[name = string("hidden_state_57_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> hidden_state_57_strides_0 = const()[name = string("hidden_state_57_strides_0"), val = tensor<int32, [1]>([4])];
            tensor<int32, [2]> hidden_state_57_pad_0 = const()[name = string("hidden_state_57_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> hidden_state_57_dilations_0 = const()[name = string("hidden_state_57_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 hidden_state_57_groups_0 = const()[name = string("hidden_state_57_groups_0"), val = int32(1)];
            tensor<int32, [3]> hidden_state_57_has_output_shape_output_shape_0 = const()[name = string("hidden_state_57_has_output_shape_output_shape_0"), val = tensor<int32, [3]>([1, 192, 80004])];
            tensor<fp16, [384, 192, 8]> audio_decoder_3_block_1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [384, 192, 8]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112386240))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112976128))))[name = string("audio_decoder_3_block_1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [192]> audio_decoder_3_block_1_conv_bias_to_fp16 = const()[name = string("audio_decoder_3_block_1_conv_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112976704)))];
            tensor<fp16, [1, 192, 80004]> hidden_state_57_has_output_shape_cast_fp16 = conv_transpose(bias = audio_decoder_3_block_1_conv_bias_to_fp16, dilations = hidden_state_57_dilations_0, groups = hidden_state_57_groups_0, output_shape = hidden_state_57_has_output_shape_output_shape_0, pad = hidden_state_57_pad_0, pad_type = hidden_state_57_pad_type_0, strides = hidden_state_57_strides_0, weight = audio_decoder_3_block_1_conv_weight_to_fp16_palettized, x = input_161_cast_fp16)[name = string("hidden_state_57_has_output_shape_cast_fp16")];
            tensor<int32, [3]> hidden_state_59_begin_0 = const()[name = string("hidden_state_59_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> hidden_state_59_end_0 = const()[name = string("hidden_state_59_end_0"), val = tensor<int32, [3]>([1, 192, 80000])];
            tensor<bool, [3]> hidden_state_59_end_mask_0 = const()[name = string("hidden_state_59_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [1, 192, 80000]> hidden_state_59_cast_fp16 = slice_by_index(begin = hidden_state_59_begin_0, end = hidden_state_59_end_0, end_mask = hidden_state_59_end_mask_0, x = hidden_state_57_has_output_shape_cast_fp16)[name = string("hidden_state_59_cast_fp16")];
            tensor<fp16, [1, 192, 1]> alpha_63_to_fp16 = const()[name = string("alpha_63_to_fp16"), val = tensor<fp16, [1, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112977152)))];
            tensor<fp16, [1, 192, 80000]> var_3404_cast_fp16 = mul(x = hidden_state_59_cast_fp16, y = alpha_63_to_fp16)[name = string("op_3404_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> var_3405_cast_fp16 = sin(x = var_3404_cast_fp16)[name = string("op_3405_cast_fp16")];
            fp16 var_3343_promoted_1_to_fp16 = const()[name = string("op_3343_promoted_1_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 192, 80000]> var_3406_cast_fp16 = pow(x = var_3405_cast_fp16, y = var_3343_promoted_1_to_fp16)[name = string("op_3406_cast_fp16")];
            tensor<fp16, [1, 192, 1]> var_3401_to_fp16 = const()[name = string("op_3401_to_fp16"), val = tensor<fp16, [1, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112977600)))];
            tensor<fp16, [1, 192, 80000]> var_3407_cast_fp16 = mul(x = var_3401_to_fp16, y = var_3406_cast_fp16)[name = string("op_3407_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> hidden_state_61_cast_fp16 = add(x = hidden_state_59_cast_fp16, y = var_3407_cast_fp16)[name = string("hidden_state_61_cast_fp16")];
            tensor<int32, [6]> input_163_pad_0 = const()[name = string("input_163_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string input_163_mode_0 = const()[name = string("input_163_mode_0"), val = string("constant")];
            fp16 const_127_to_fp16 = const()[name = string("const_127_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 192, 80006]> input_163_cast_fp16 = pad(constant_val = const_127_to_fp16, mode = input_163_mode_0, pad = input_163_pad_0, x = hidden_state_61_cast_fp16)[name = string("input_163_cast_fp16")];
            string var_3422_pad_type_0 = const()[name = string("op_3422_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3422_strides_0 = const()[name = string("op_3422_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3422_pad_0 = const()[name = string("op_3422_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_3422_dilations_0 = const()[name = string("op_3422_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_3422_groups_0 = const()[name = string("op_3422_groups_0"), val = int32(1)];
            tensor<fp16, [192, 192, 7]> audio_decoder_3_block_2_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [192, 192, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112978048))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113236160))))[name = string("audio_decoder_3_block_2_conv1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [192]> audio_decoder_3_block_2_conv1_conv_bias_to_fp16 = const()[name = string("audio_decoder_3_block_2_conv1_conv_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113236736)))];
            tensor<fp16, [1, 192, 80000]> var_3422_cast_fp16 = conv(bias = audio_decoder_3_block_2_conv1_conv_bias_to_fp16, dilations = var_3422_dilations_0, groups = var_3422_groups_0, pad = var_3422_pad_0, pad_type = var_3422_pad_type_0, strides = var_3422_strides_0, weight = audio_decoder_3_block_2_conv1_conv_weight_to_fp16_palettized, x = input_163_cast_fp16)[name = string("op_3422_cast_fp16")];
            tensor<fp16, [1, 192, 1]> alpha_67_to_fp16 = const()[name = string("alpha_67_to_fp16"), val = tensor<fp16, [1, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113237184)))];
            tensor<fp16, [1, 192, 80000]> var_3437_cast_fp16 = mul(x = var_3422_cast_fp16, y = alpha_67_to_fp16)[name = string("op_3437_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> var_3438_cast_fp16 = sin(x = var_3437_cast_fp16)[name = string("op_3438_cast_fp16")];
            fp16 var_3343_promoted_2_to_fp16 = const()[name = string("op_3343_promoted_2_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 192, 80000]> var_3439_cast_fp16 = pow(x = var_3438_cast_fp16, y = var_3343_promoted_2_to_fp16)[name = string("op_3439_cast_fp16")];
            tensor<fp16, [1, 192, 1]> var_3434_to_fp16 = const()[name = string("op_3434_to_fp16"), val = tensor<fp16, [1, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113237632)))];
            tensor<fp16, [1, 192, 80000]> var_3440_cast_fp16 = mul(x = var_3434_to_fp16, y = var_3439_cast_fp16)[name = string("op_3440_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> hidden_state_63_cast_fp16 = add(x = var_3422_cast_fp16, y = var_3440_cast_fp16)[name = string("hidden_state_63_cast_fp16")];
            string var_3455_pad_type_0 = const()[name = string("op_3455_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3455_strides_0 = const()[name = string("op_3455_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3455_pad_0 = const()[name = string("op_3455_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_3455_dilations_0 = const()[name = string("op_3455_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_3455_groups_0 = const()[name = string("op_3455_groups_0"), val = int32(1)];
            tensor<fp16, [192, 192, 1]> audio_decoder_3_block_2_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [192, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113238080))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113275008))))[name = string("audio_decoder_3_block_2_conv2_conv_weight_to_fp16_palettized")];
            tensor<fp16, [192]> audio_decoder_3_block_2_conv2_conv_bias_to_fp16 = const()[name = string("audio_decoder_3_block_2_conv2_conv_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113275584)))];
            tensor<fp16, [1, 192, 80000]> var_3455_cast_fp16 = conv(bias = audio_decoder_3_block_2_conv2_conv_bias_to_fp16, dilations = var_3455_dilations_0, groups = var_3455_groups_0, pad = var_3455_pad_0, pad_type = var_3455_pad_type_0, strides = var_3455_strides_0, weight = audio_decoder_3_block_2_conv2_conv_weight_to_fp16_palettized, x = hidden_state_63_cast_fp16)[name = string("op_3455_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> hidden_states_153_cast_fp16 = add(x = var_3455_cast_fp16, y = hidden_state_59_cast_fp16)[name = string("hidden_states_153_cast_fp16")];
            tensor<fp16, [1, 192, 1]> alpha_71_to_fp16 = const()[name = string("alpha_71_to_fp16"), val = tensor<fp16, [1, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113276032)))];
            tensor<fp16, [1, 192, 80000]> var_3475_cast_fp16 = mul(x = hidden_states_153_cast_fp16, y = alpha_71_to_fp16)[name = string("op_3475_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> var_3476_cast_fp16 = sin(x = var_3475_cast_fp16)[name = string("op_3476_cast_fp16")];
            fp16 var_3343_promoted_3_to_fp16 = const()[name = string("op_3343_promoted_3_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 192, 80000]> var_3477_cast_fp16 = pow(x = var_3476_cast_fp16, y = var_3343_promoted_3_to_fp16)[name = string("op_3477_cast_fp16")];
            tensor<fp16, [1, 192, 1]> var_3472_to_fp16 = const()[name = string("op_3472_to_fp16"), val = tensor<fp16, [1, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113276480)))];
            tensor<fp16, [1, 192, 80000]> var_3478_cast_fp16 = mul(x = var_3472_to_fp16, y = var_3477_cast_fp16)[name = string("op_3478_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> hidden_state_67_cast_fp16 = add(x = hidden_states_153_cast_fp16, y = var_3478_cast_fp16)[name = string("hidden_state_67_cast_fp16")];
            tensor<int32, [6]> input_167_pad_0 = const()[name = string("input_167_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 18, 0])];
            string input_167_mode_0 = const()[name = string("input_167_mode_0"), val = string("constant")];
            fp16 const_131_to_fp16 = const()[name = string("const_131_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 192, 80018]> input_167_cast_fp16 = pad(constant_val = const_131_to_fp16, mode = input_167_mode_0, pad = input_167_pad_0, x = hidden_state_67_cast_fp16)[name = string("input_167_cast_fp16")];
            string var_3493_pad_type_0 = const()[name = string("op_3493_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3493_dilations_0 = const()[name = string("op_3493_dilations_0"), val = tensor<int32, [1]>([3])];
            tensor<int32, [1]> var_3493_strides_0 = const()[name = string("op_3493_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3493_pad_0 = const()[name = string("op_3493_pad_0"), val = tensor<int32, [2]>([0, 0])];
            int32 var_3493_groups_0 = const()[name = string("op_3493_groups_0"), val = int32(1)];
            tensor<fp16, [192, 192, 7]> audio_decoder_3_block_3_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [192, 192, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113276928))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113535040))))[name = string("audio_decoder_3_block_3_conv1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [192]> audio_decoder_3_block_3_conv1_conv_bias_to_fp16 = const()[name = string("audio_decoder_3_block_3_conv1_conv_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113535616)))];
            tensor<fp16, [1, 192, 80000]> var_3493_cast_fp16 = conv(bias = audio_decoder_3_block_3_conv1_conv_bias_to_fp16, dilations = var_3493_dilations_0, groups = var_3493_groups_0, pad = var_3493_pad_0, pad_type = var_3493_pad_type_0, strides = var_3493_strides_0, weight = audio_decoder_3_block_3_conv1_conv_weight_to_fp16_palettized, x = input_167_cast_fp16)[name = string("op_3493_cast_fp16")];
            tensor<fp16, [1, 192, 1]> alpha_75_to_fp16 = const()[name = string("alpha_75_to_fp16"), val = tensor<fp16, [1, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113536064)))];
            tensor<fp16, [1, 192, 80000]> var_3508_cast_fp16 = mul(x = var_3493_cast_fp16, y = alpha_75_to_fp16)[name = string("op_3508_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> var_3509_cast_fp16 = sin(x = var_3508_cast_fp16)[name = string("op_3509_cast_fp16")];
            fp16 var_3343_promoted_4_to_fp16 = const()[name = string("op_3343_promoted_4_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 192, 80000]> var_3510_cast_fp16 = pow(x = var_3509_cast_fp16, y = var_3343_promoted_4_to_fp16)[name = string("op_3510_cast_fp16")];
            tensor<fp16, [1, 192, 1]> var_3505_to_fp16 = const()[name = string("op_3505_to_fp16"), val = tensor<fp16, [1, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113536512)))];
            tensor<fp16, [1, 192, 80000]> var_3511_cast_fp16 = mul(x = var_3505_to_fp16, y = var_3510_cast_fp16)[name = string("op_3511_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> hidden_state_69_cast_fp16 = add(x = var_3493_cast_fp16, y = var_3511_cast_fp16)[name = string("hidden_state_69_cast_fp16")];
            string var_3526_pad_type_0 = const()[name = string("op_3526_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3526_strides_0 = const()[name = string("op_3526_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3526_pad_0 = const()[name = string("op_3526_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_3526_dilations_0 = const()[name = string("op_3526_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_3526_groups_0 = const()[name = string("op_3526_groups_0"), val = int32(1)];
            tensor<fp16, [192, 192, 1]> audio_decoder_3_block_3_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [192, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113536960))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113573888))))[name = string("audio_decoder_3_block_3_conv2_conv_weight_to_fp16_palettized")];
            tensor<fp16, [192]> audio_decoder_3_block_3_conv2_conv_bias_to_fp16 = const()[name = string("audio_decoder_3_block_3_conv2_conv_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113574464)))];
            tensor<fp16, [1, 192, 80000]> var_3526_cast_fp16 = conv(bias = audio_decoder_3_block_3_conv2_conv_bias_to_fp16, dilations = var_3526_dilations_0, groups = var_3526_groups_0, pad = var_3526_pad_0, pad_type = var_3526_pad_type_0, strides = var_3526_strides_0, weight = audio_decoder_3_block_3_conv2_conv_weight_to_fp16_palettized, x = hidden_state_69_cast_fp16)[name = string("op_3526_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> hidden_states_157_cast_fp16 = add(x = var_3526_cast_fp16, y = hidden_states_153_cast_fp16)[name = string("hidden_states_157_cast_fp16")];
            tensor<fp16, [1, 192, 1]> alpha_79_to_fp16 = const()[name = string("alpha_79_to_fp16"), val = tensor<fp16, [1, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113574912)))];
            tensor<fp16, [1, 192, 80000]> var_3546_cast_fp16 = mul(x = hidden_states_157_cast_fp16, y = alpha_79_to_fp16)[name = string("op_3546_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> var_3547_cast_fp16 = sin(x = var_3546_cast_fp16)[name = string("op_3547_cast_fp16")];
            fp16 var_3343_promoted_5_to_fp16 = const()[name = string("op_3343_promoted_5_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 192, 80000]> var_3548_cast_fp16 = pow(x = var_3547_cast_fp16, y = var_3343_promoted_5_to_fp16)[name = string("op_3548_cast_fp16")];
            tensor<fp16, [1, 192, 1]> var_3543_to_fp16 = const()[name = string("op_3543_to_fp16"), val = tensor<fp16, [1, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113575360)))];
            tensor<fp16, [1, 192, 80000]> var_3549_cast_fp16 = mul(x = var_3543_to_fp16, y = var_3548_cast_fp16)[name = string("op_3549_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> hidden_state_73_cast_fp16 = add(x = hidden_states_157_cast_fp16, y = var_3549_cast_fp16)[name = string("hidden_state_73_cast_fp16")];
            tensor<int32, [6]> input_171_pad_0 = const()[name = string("input_171_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 54, 0])];
            string input_171_mode_0 = const()[name = string("input_171_mode_0"), val = string("constant")];
            fp16 const_135_to_fp16 = const()[name = string("const_135_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 192, 80054]> input_171_cast_fp16 = pad(constant_val = const_135_to_fp16, mode = input_171_mode_0, pad = input_171_pad_0, x = hidden_state_73_cast_fp16)[name = string("input_171_cast_fp16")];
            string var_3564_pad_type_0 = const()[name = string("op_3564_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3564_dilations_0 = const()[name = string("op_3564_dilations_0"), val = tensor<int32, [1]>([9])];
            tensor<int32, [1]> var_3564_strides_0 = const()[name = string("op_3564_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3564_pad_0 = const()[name = string("op_3564_pad_0"), val = tensor<int32, [2]>([0, 0])];
            int32 var_3564_groups_0 = const()[name = string("op_3564_groups_0"), val = int32(1)];
            tensor<fp16, [192, 192, 7]> audio_decoder_3_block_4_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [192, 192, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113575808))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113833920))))[name = string("audio_decoder_3_block_4_conv1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [192]> audio_decoder_3_block_4_conv1_conv_bias_to_fp16 = const()[name = string("audio_decoder_3_block_4_conv1_conv_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113834496)))];
            tensor<fp16, [1, 192, 80000]> var_3564_cast_fp16 = conv(bias = audio_decoder_3_block_4_conv1_conv_bias_to_fp16, dilations = var_3564_dilations_0, groups = var_3564_groups_0, pad = var_3564_pad_0, pad_type = var_3564_pad_type_0, strides = var_3564_strides_0, weight = audio_decoder_3_block_4_conv1_conv_weight_to_fp16_palettized, x = input_171_cast_fp16)[name = string("op_3564_cast_fp16")];
            tensor<fp16, [1, 192, 1]> alpha_83_to_fp16 = const()[name = string("alpha_83_to_fp16"), val = tensor<fp16, [1, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113834944)))];
            tensor<fp16, [1, 192, 80000]> var_3579_cast_fp16 = mul(x = var_3564_cast_fp16, y = alpha_83_to_fp16)[name = string("op_3579_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> var_3580_cast_fp16 = sin(x = var_3579_cast_fp16)[name = string("op_3580_cast_fp16")];
            fp16 var_3343_promoted_6_to_fp16 = const()[name = string("op_3343_promoted_6_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 192, 80000]> var_3581_cast_fp16 = pow(x = var_3580_cast_fp16, y = var_3343_promoted_6_to_fp16)[name = string("op_3581_cast_fp16")];
            tensor<fp16, [1, 192, 1]> var_3576_to_fp16 = const()[name = string("op_3576_to_fp16"), val = tensor<fp16, [1, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113835392)))];
            tensor<fp16, [1, 192, 80000]> var_3582_cast_fp16 = mul(x = var_3576_to_fp16, y = var_3581_cast_fp16)[name = string("op_3582_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> hidden_state_75_cast_fp16 = add(x = var_3564_cast_fp16, y = var_3582_cast_fp16)[name = string("hidden_state_75_cast_fp16")];
            string var_3597_pad_type_0 = const()[name = string("op_3597_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3597_strides_0 = const()[name = string("op_3597_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3597_pad_0 = const()[name = string("op_3597_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_3597_dilations_0 = const()[name = string("op_3597_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_3597_groups_0 = const()[name = string("op_3597_groups_0"), val = int32(1)];
            tensor<fp16, [192, 192, 1]> audio_decoder_3_block_4_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [192, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113835840))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113872768))))[name = string("audio_decoder_3_block_4_conv2_conv_weight_to_fp16_palettized")];
            tensor<fp16, [192]> audio_decoder_3_block_4_conv2_conv_bias_to_fp16 = const()[name = string("audio_decoder_3_block_4_conv2_conv_bias_to_fp16"), val = tensor<fp16, [192]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113873344)))];
            tensor<fp16, [1, 192, 80000]> var_3597_cast_fp16 = conv(bias = audio_decoder_3_block_4_conv2_conv_bias_to_fp16, dilations = var_3597_dilations_0, groups = var_3597_groups_0, pad = var_3597_pad_0, pad_type = var_3597_pad_type_0, strides = var_3597_strides_0, weight = audio_decoder_3_block_4_conv2_conv_weight_to_fp16_palettized, x = hidden_state_75_cast_fp16)[name = string("op_3597_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> hidden_states_161_cast_fp16 = add(x = var_3597_cast_fp16, y = hidden_states_157_cast_fp16)[name = string("hidden_states_161_cast_fp16")];
            tensor<fp16, [1, 192, 1]> alpha_87_to_fp16 = const()[name = string("alpha_87_to_fp16"), val = tensor<fp16, [1, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113873792)))];
            tensor<fp16, [1, 192, 80000]> var_3637_cast_fp16 = mul(x = hidden_states_161_cast_fp16, y = alpha_87_to_fp16)[name = string("op_3637_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> var_3638_cast_fp16 = sin(x = var_3637_cast_fp16)[name = string("op_3638_cast_fp16")];
            fp16 var_3613_promoted_to_fp16 = const()[name = string("op_3613_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 192, 80000]> var_3639_cast_fp16 = pow(x = var_3638_cast_fp16, y = var_3613_promoted_to_fp16)[name = string("op_3639_cast_fp16")];
            tensor<fp16, [1, 192, 1]> var_3634_to_fp16 = const()[name = string("op_3634_to_fp16"), val = tensor<fp16, [1, 192, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113874240)))];
            tensor<fp16, [1, 192, 80000]> var_3640_cast_fp16 = mul(x = var_3634_to_fp16, y = var_3639_cast_fp16)[name = string("op_3640_cast_fp16")];
            tensor<fp16, [1, 192, 80000]> input_175_cast_fp16 = add(x = hidden_states_161_cast_fp16, y = var_3640_cast_fp16)[name = string("input_175_cast_fp16")];
            string hidden_state_79_pad_type_0 = const()[name = string("hidden_state_79_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> hidden_state_79_strides_0 = const()[name = string("hidden_state_79_strides_0"), val = tensor<int32, [1]>([3])];
            tensor<int32, [2]> hidden_state_79_pad_0 = const()[name = string("hidden_state_79_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> hidden_state_79_dilations_0 = const()[name = string("hidden_state_79_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 hidden_state_79_groups_0 = const()[name = string("hidden_state_79_groups_0"), val = int32(1)];
            tensor<int32, [3]> hidden_state_79_has_output_shape_output_shape_0 = const()[name = string("hidden_state_79_has_output_shape_output_shape_0"), val = tensor<int32, [3]>([1, 96, 240003])];
            tensor<fp16, [192, 96, 6]> audio_decoder_4_block_1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [192, 96, 6]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113874688))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113985344))))[name = string("audio_decoder_4_block_1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [96]> audio_decoder_4_block_1_conv_bias_to_fp16 = const()[name = string("audio_decoder_4_block_1_conv_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113985920)))];
            tensor<fp16, [1, 96, 240003]> hidden_state_79_has_output_shape_cast_fp16 = conv_transpose(bias = audio_decoder_4_block_1_conv_bias_to_fp16, dilations = hidden_state_79_dilations_0, groups = hidden_state_79_groups_0, output_shape = hidden_state_79_has_output_shape_output_shape_0, pad = hidden_state_79_pad_0, pad_type = hidden_state_79_pad_type_0, strides = hidden_state_79_strides_0, weight = audio_decoder_4_block_1_conv_weight_to_fp16_palettized, x = input_175_cast_fp16)[name = string("hidden_state_79_has_output_shape_cast_fp16")];
            tensor<int32, [3]> hidden_state_81_begin_0 = const()[name = string("hidden_state_81_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> hidden_state_81_end_0 = const()[name = string("hidden_state_81_end_0"), val = tensor<int32, [3]>([1, 96, 240000])];
            tensor<bool, [3]> hidden_state_81_end_mask_0 = const()[name = string("hidden_state_81_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [1, 96, 240000]> hidden_state_81_cast_fp16 = slice_by_index(begin = hidden_state_81_begin_0, end = hidden_state_81_end_0, end_mask = hidden_state_81_end_mask_0, x = hidden_state_79_has_output_shape_cast_fp16)[name = string("hidden_state_81_cast_fp16")];
            tensor<fp16, [1, 96, 1]> alpha_91_to_fp16 = const()[name = string("alpha_91_to_fp16"), val = tensor<fp16, [1, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113986176)))];
            tensor<fp16, [1, 96, 240000]> var_3674_cast_fp16 = mul(x = hidden_state_81_cast_fp16, y = alpha_91_to_fp16)[name = string("op_3674_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> var_3675_cast_fp16 = sin(x = var_3674_cast_fp16)[name = string("op_3675_cast_fp16")];
            fp16 var_3613_promoted_1_to_fp16 = const()[name = string("op_3613_promoted_1_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 240000]> var_3676_cast_fp16 = pow(x = var_3675_cast_fp16, y = var_3613_promoted_1_to_fp16)[name = string("op_3676_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_3671_to_fp16 = const()[name = string("op_3671_to_fp16"), val = tensor<fp16, [1, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113986432)))];
            tensor<fp16, [1, 96, 240000]> var_3677_cast_fp16 = mul(x = var_3671_to_fp16, y = var_3676_cast_fp16)[name = string("op_3677_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> hidden_state_83_cast_fp16 = add(x = hidden_state_81_cast_fp16, y = var_3677_cast_fp16)[name = string("hidden_state_83_cast_fp16")];
            tensor<int32, [6]> input_177_pad_0 = const()[name = string("input_177_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string input_177_mode_0 = const()[name = string("input_177_mode_0"), val = string("constant")];
            fp16 const_140_to_fp16 = const()[name = string("const_140_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 96, 240006]> input_177_cast_fp16 = pad(constant_val = const_140_to_fp16, mode = input_177_mode_0, pad = input_177_pad_0, x = hidden_state_83_cast_fp16)[name = string("input_177_cast_fp16")];
            string var_3692_pad_type_0 = const()[name = string("op_3692_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3692_strides_0 = const()[name = string("op_3692_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3692_pad_0 = const()[name = string("op_3692_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_3692_dilations_0 = const()[name = string("op_3692_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_3692_groups_0 = const()[name = string("op_3692_groups_0"), val = int32(1)];
            tensor<fp16, [96, 96, 7]> audio_decoder_4_block_2_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [96, 96, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113986688))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114051264))))[name = string("audio_decoder_4_block_2_conv1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [96]> audio_decoder_4_block_2_conv1_conv_bias_to_fp16 = const()[name = string("audio_decoder_4_block_2_conv1_conv_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114051840)))];
            tensor<fp16, [1, 96, 240000]> var_3692_cast_fp16 = conv(bias = audio_decoder_4_block_2_conv1_conv_bias_to_fp16, dilations = var_3692_dilations_0, groups = var_3692_groups_0, pad = var_3692_pad_0, pad_type = var_3692_pad_type_0, strides = var_3692_strides_0, weight = audio_decoder_4_block_2_conv1_conv_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = string("op_3692_cast_fp16")];
            tensor<fp16, [1, 96, 1]> alpha_95_to_fp16 = const()[name = string("alpha_95_to_fp16"), val = tensor<fp16, [1, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114052096)))];
            tensor<fp16, [1, 96, 240000]> var_3707_cast_fp16 = mul(x = var_3692_cast_fp16, y = alpha_95_to_fp16)[name = string("op_3707_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> var_3708_cast_fp16 = sin(x = var_3707_cast_fp16)[name = string("op_3708_cast_fp16")];
            fp16 var_3613_promoted_2_to_fp16 = const()[name = string("op_3613_promoted_2_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 240000]> var_3709_cast_fp16 = pow(x = var_3708_cast_fp16, y = var_3613_promoted_2_to_fp16)[name = string("op_3709_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_3704_to_fp16 = const()[name = string("op_3704_to_fp16"), val = tensor<fp16, [1, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114052352)))];
            tensor<fp16, [1, 96, 240000]> var_3710_cast_fp16 = mul(x = var_3704_to_fp16, y = var_3709_cast_fp16)[name = string("op_3710_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> hidden_state_85_cast_fp16 = add(x = var_3692_cast_fp16, y = var_3710_cast_fp16)[name = string("hidden_state_85_cast_fp16")];
            string var_3725_pad_type_0 = const()[name = string("op_3725_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3725_strides_0 = const()[name = string("op_3725_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3725_pad_0 = const()[name = string("op_3725_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_3725_dilations_0 = const()[name = string("op_3725_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_3725_groups_0 = const()[name = string("op_3725_groups_0"), val = int32(1)];
            tensor<fp16, [96, 96, 1]> audio_decoder_4_block_2_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [96, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114052608))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114061888))))[name = string("audio_decoder_4_block_2_conv2_conv_weight_to_fp16_palettized")];
            tensor<fp16, [96]> audio_decoder_4_block_2_conv2_conv_bias_to_fp16 = const()[name = string("audio_decoder_4_block_2_conv2_conv_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114062464)))];
            tensor<fp16, [1, 96, 240000]> var_3725_cast_fp16 = conv(bias = audio_decoder_4_block_2_conv2_conv_bias_to_fp16, dilations = var_3725_dilations_0, groups = var_3725_groups_0, pad = var_3725_pad_0, pad_type = var_3725_pad_type_0, strides = var_3725_strides_0, weight = audio_decoder_4_block_2_conv2_conv_weight_to_fp16_palettized, x = hidden_state_85_cast_fp16)[name = string("op_3725_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> hidden_states_167_cast_fp16 = add(x = var_3725_cast_fp16, y = hidden_state_81_cast_fp16)[name = string("hidden_states_167_cast_fp16")];
            tensor<fp16, [1, 96, 1]> alpha_99_to_fp16 = const()[name = string("alpha_99_to_fp16"), val = tensor<fp16, [1, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114062720)))];
            tensor<fp16, [1, 96, 240000]> var_3745_cast_fp16 = mul(x = hidden_states_167_cast_fp16, y = alpha_99_to_fp16)[name = string("op_3745_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> var_3746_cast_fp16 = sin(x = var_3745_cast_fp16)[name = string("op_3746_cast_fp16")];
            fp16 var_3613_promoted_3_to_fp16 = const()[name = string("op_3613_promoted_3_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 240000]> var_3747_cast_fp16 = pow(x = var_3746_cast_fp16, y = var_3613_promoted_3_to_fp16)[name = string("op_3747_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_3742_to_fp16 = const()[name = string("op_3742_to_fp16"), val = tensor<fp16, [1, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114062976)))];
            tensor<fp16, [1, 96, 240000]> var_3748_cast_fp16 = mul(x = var_3742_to_fp16, y = var_3747_cast_fp16)[name = string("op_3748_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> hidden_state_89_cast_fp16 = add(x = hidden_states_167_cast_fp16, y = var_3748_cast_fp16)[name = string("hidden_state_89_cast_fp16")];
            tensor<int32, [6]> input_181_pad_0 = const()[name = string("input_181_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 18, 0])];
            string input_181_mode_0 = const()[name = string("input_181_mode_0"), val = string("constant")];
            fp16 const_144_to_fp16 = const()[name = string("const_144_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 96, 240018]> input_181_cast_fp16 = pad(constant_val = const_144_to_fp16, mode = input_181_mode_0, pad = input_181_pad_0, x = hidden_state_89_cast_fp16)[name = string("input_181_cast_fp16")];
            string var_3763_pad_type_0 = const()[name = string("op_3763_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3763_dilations_0 = const()[name = string("op_3763_dilations_0"), val = tensor<int32, [1]>([3])];
            tensor<int32, [1]> var_3763_strides_0 = const()[name = string("op_3763_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3763_pad_0 = const()[name = string("op_3763_pad_0"), val = tensor<int32, [2]>([0, 0])];
            int32 var_3763_groups_0 = const()[name = string("op_3763_groups_0"), val = int32(1)];
            tensor<fp16, [96, 96, 7]> audio_decoder_4_block_3_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [96, 96, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114063232))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114127808))))[name = string("audio_decoder_4_block_3_conv1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [96]> audio_decoder_4_block_3_conv1_conv_bias_to_fp16 = const()[name = string("audio_decoder_4_block_3_conv1_conv_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114128384)))];
            tensor<fp16, [1, 96, 240000]> var_3763_cast_fp16 = conv(bias = audio_decoder_4_block_3_conv1_conv_bias_to_fp16, dilations = var_3763_dilations_0, groups = var_3763_groups_0, pad = var_3763_pad_0, pad_type = var_3763_pad_type_0, strides = var_3763_strides_0, weight = audio_decoder_4_block_3_conv1_conv_weight_to_fp16_palettized, x = input_181_cast_fp16)[name = string("op_3763_cast_fp16")];
            tensor<fp16, [1, 96, 1]> alpha_103_to_fp16 = const()[name = string("alpha_103_to_fp16"), val = tensor<fp16, [1, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114128640)))];
            tensor<fp16, [1, 96, 240000]> var_3778_cast_fp16 = mul(x = var_3763_cast_fp16, y = alpha_103_to_fp16)[name = string("op_3778_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> var_3779_cast_fp16 = sin(x = var_3778_cast_fp16)[name = string("op_3779_cast_fp16")];
            fp16 var_3613_promoted_4_to_fp16 = const()[name = string("op_3613_promoted_4_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 240000]> var_3780_cast_fp16 = pow(x = var_3779_cast_fp16, y = var_3613_promoted_4_to_fp16)[name = string("op_3780_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_3775_to_fp16 = const()[name = string("op_3775_to_fp16"), val = tensor<fp16, [1, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114128896)))];
            tensor<fp16, [1, 96, 240000]> var_3781_cast_fp16 = mul(x = var_3775_to_fp16, y = var_3780_cast_fp16)[name = string("op_3781_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> hidden_state_91_cast_fp16 = add(x = var_3763_cast_fp16, y = var_3781_cast_fp16)[name = string("hidden_state_91_cast_fp16")];
            string var_3796_pad_type_0 = const()[name = string("op_3796_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3796_strides_0 = const()[name = string("op_3796_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3796_pad_0 = const()[name = string("op_3796_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_3796_dilations_0 = const()[name = string("op_3796_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_3796_groups_0 = const()[name = string("op_3796_groups_0"), val = int32(1)];
            tensor<fp16, [96, 96, 1]> audio_decoder_4_block_3_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [96, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114129152))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114138432))))[name = string("audio_decoder_4_block_3_conv2_conv_weight_to_fp16_palettized")];
            tensor<fp16, [96]> audio_decoder_4_block_3_conv2_conv_bias_to_fp16 = const()[name = string("audio_decoder_4_block_3_conv2_conv_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114139008)))];
            tensor<fp16, [1, 96, 240000]> var_3796_cast_fp16 = conv(bias = audio_decoder_4_block_3_conv2_conv_bias_to_fp16, dilations = var_3796_dilations_0, groups = var_3796_groups_0, pad = var_3796_pad_0, pad_type = var_3796_pad_type_0, strides = var_3796_strides_0, weight = audio_decoder_4_block_3_conv2_conv_weight_to_fp16_palettized, x = hidden_state_91_cast_fp16)[name = string("op_3796_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> hidden_states_171_cast_fp16 = add(x = var_3796_cast_fp16, y = hidden_states_167_cast_fp16)[name = string("hidden_states_171_cast_fp16")];
            tensor<fp16, [1, 96, 1]> alpha_107_to_fp16 = const()[name = string("alpha_107_to_fp16"), val = tensor<fp16, [1, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114139264)))];
            tensor<fp16, [1, 96, 240000]> var_3816_cast_fp16 = mul(x = hidden_states_171_cast_fp16, y = alpha_107_to_fp16)[name = string("op_3816_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> var_3817_cast_fp16 = sin(x = var_3816_cast_fp16)[name = string("op_3817_cast_fp16")];
            fp16 var_3613_promoted_5_to_fp16 = const()[name = string("op_3613_promoted_5_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 240000]> var_3818_cast_fp16 = pow(x = var_3817_cast_fp16, y = var_3613_promoted_5_to_fp16)[name = string("op_3818_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_3813_to_fp16 = const()[name = string("op_3813_to_fp16"), val = tensor<fp16, [1, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114139520)))];
            tensor<fp16, [1, 96, 240000]> var_3819_cast_fp16 = mul(x = var_3813_to_fp16, y = var_3818_cast_fp16)[name = string("op_3819_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> hidden_state_95_cast_fp16 = add(x = hidden_states_171_cast_fp16, y = var_3819_cast_fp16)[name = string("hidden_state_95_cast_fp16")];
            tensor<int32, [6]> input_185_pad_0 = const()[name = string("input_185_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 54, 0])];
            string input_185_mode_0 = const()[name = string("input_185_mode_0"), val = string("constant")];
            fp16 const_148_to_fp16 = const()[name = string("const_148_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 96, 240054]> input_185_cast_fp16 = pad(constant_val = const_148_to_fp16, mode = input_185_mode_0, pad = input_185_pad_0, x = hidden_state_95_cast_fp16)[name = string("input_185_cast_fp16")];
            string var_3834_pad_type_0 = const()[name = string("op_3834_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3834_dilations_0 = const()[name = string("op_3834_dilations_0"), val = tensor<int32, [1]>([9])];
            tensor<int32, [1]> var_3834_strides_0 = const()[name = string("op_3834_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3834_pad_0 = const()[name = string("op_3834_pad_0"), val = tensor<int32, [2]>([0, 0])];
            int32 var_3834_groups_0 = const()[name = string("op_3834_groups_0"), val = int32(1)];
            tensor<fp16, [96, 96, 7]> audio_decoder_4_block_4_conv1_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [96, 96, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114139776))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114204352))))[name = string("audio_decoder_4_block_4_conv1_conv_weight_to_fp16_palettized")];
            tensor<fp16, [96]> audio_decoder_4_block_4_conv1_conv_bias_to_fp16 = const()[name = string("audio_decoder_4_block_4_conv1_conv_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114204928)))];
            tensor<fp16, [1, 96, 240000]> var_3834_cast_fp16 = conv(bias = audio_decoder_4_block_4_conv1_conv_bias_to_fp16, dilations = var_3834_dilations_0, groups = var_3834_groups_0, pad = var_3834_pad_0, pad_type = var_3834_pad_type_0, strides = var_3834_strides_0, weight = audio_decoder_4_block_4_conv1_conv_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = string("op_3834_cast_fp16")];
            tensor<fp16, [1, 96, 1]> alpha_111_to_fp16 = const()[name = string("alpha_111_to_fp16"), val = tensor<fp16, [1, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114205184)))];
            tensor<fp16, [1, 96, 240000]> var_3849_cast_fp16 = mul(x = var_3834_cast_fp16, y = alpha_111_to_fp16)[name = string("op_3849_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> var_3850_cast_fp16 = sin(x = var_3849_cast_fp16)[name = string("op_3850_cast_fp16")];
            fp16 var_3613_promoted_6_to_fp16 = const()[name = string("op_3613_promoted_6_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 240000]> var_3851_cast_fp16 = pow(x = var_3850_cast_fp16, y = var_3613_promoted_6_to_fp16)[name = string("op_3851_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_3846_to_fp16 = const()[name = string("op_3846_to_fp16"), val = tensor<fp16, [1, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114205440)))];
            tensor<fp16, [1, 96, 240000]> var_3852_cast_fp16 = mul(x = var_3846_to_fp16, y = var_3851_cast_fp16)[name = string("op_3852_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> hidden_state_97_cast_fp16 = add(x = var_3834_cast_fp16, y = var_3852_cast_fp16)[name = string("hidden_state_97_cast_fp16")];
            string var_3867_pad_type_0 = const()[name = string("op_3867_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3867_strides_0 = const()[name = string("op_3867_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3867_pad_0 = const()[name = string("op_3867_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_3867_dilations_0 = const()[name = string("op_3867_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_3867_groups_0 = const()[name = string("op_3867_groups_0"), val = int32(1)];
            tensor<fp16, [96, 96, 1]> audio_decoder_4_block_4_conv2_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [96, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114205696))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114214976))))[name = string("audio_decoder_4_block_4_conv2_conv_weight_to_fp16_palettized")];
            tensor<fp16, [96]> audio_decoder_4_block_4_conv2_conv_bias_to_fp16 = const()[name = string("audio_decoder_4_block_4_conv2_conv_bias_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114215552)))];
            tensor<fp16, [1, 96, 240000]> var_3867_cast_fp16 = conv(bias = audio_decoder_4_block_4_conv2_conv_bias_to_fp16, dilations = var_3867_dilations_0, groups = var_3867_groups_0, pad = var_3867_pad_0, pad_type = var_3867_pad_type_0, strides = var_3867_strides_0, weight = audio_decoder_4_block_4_conv2_conv_weight_to_fp16_palettized, x = hidden_state_97_cast_fp16)[name = string("op_3867_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> hidden_states_cast_fp16 = add(x = var_3867_cast_fp16, y = hidden_states_171_cast_fp16)[name = string("hidden_states_cast_fp16")];
            tensor<fp16, [1, 96, 1]> alpha_115_to_fp16 = const()[name = string("alpha_115_to_fp16"), val = tensor<fp16, [1, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114215808)))];
            tensor<fp16, [1, 96, 240000]> var_3887_cast_fp16 = mul(x = hidden_states_cast_fp16, y = alpha_115_to_fp16)[name = string("op_3887_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> var_3888_cast_fp16 = sin(x = var_3887_cast_fp16)[name = string("op_3888_cast_fp16")];
            fp16 var_3870_promoted_to_fp16 = const()[name = string("op_3870_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 240000]> var_3889_cast_fp16 = pow(x = var_3888_cast_fp16, y = var_3870_promoted_to_fp16)[name = string("op_3889_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_3884_to_fp16 = const()[name = string("op_3884_to_fp16"), val = tensor<fp16, [1, 96, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114216064)))];
            tensor<fp16, [1, 96, 240000]> var_3890_cast_fp16 = mul(x = var_3884_to_fp16, y = var_3889_cast_fp16)[name = string("op_3890_cast_fp16")];
            tensor<fp16, [1, 96, 240000]> hidden_state_cast_fp16 = add(x = hidden_states_cast_fp16, y = var_3890_cast_fp16)[name = string("hidden_state_cast_fp16")];
            tensor<int32, [6]> input_pad_0 = const()[name = string("input_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string input_mode_0 = const()[name = string("input_mode_0"), val = string("constant")];
            fp16 const_152_to_fp16 = const()[name = string("const_152_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 96, 240006]> input_cast_fp16 = pad(constant_val = const_152_to_fp16, mode = input_mode_0, pad = input_pad_0, x = hidden_state_cast_fp16)[name = string("input_cast_fp16")];
            string var_3914_pad_type_0 = const()[name = string("op_3914_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_3914_strides_0 = const()[name = string("op_3914_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_3914_pad_0 = const()[name = string("op_3914_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_3914_dilations_0 = const()[name = string("op_3914_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_3914_groups_0 = const()[name = string("op_3914_groups_0"), val = int32(1)];
            tensor<fp16, [1, 96, 7]> audio_decoder_6_conv_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint8, [1, 96, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114216320))), lut = tensor<fp16, [1, 1, 1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114217088))))[name = string("audio_decoder_6_conv_weight_to_fp16_palettized")];
            tensor<fp16, [1]> audio_decoder_6_conv_bias_to_fp16 = const()[name = string("audio_decoder_6_conv_bias_to_fp16"), val = tensor<fp16, [1]>([-0x1.1p-19])];
            tensor<fp16, [1, 1, 240000]> var_3914_cast_fp16 = conv(bias = audio_decoder_6_conv_bias_to_fp16, dilations = var_3914_dilations_0, groups = var_3914_groups_0, pad = var_3914_pad_0, pad_type = var_3914_pad_type_0, strides = var_3914_strides_0, weight = audio_decoder_6_conv_weight_to_fp16_palettized, x = input_cast_fp16)[name = string("op_3914_cast_fp16")];
            fp16 var_3916_promoted_to_fp16 = const()[name = string("op_3916_promoted_to_fp16"), val = fp16(-0x1p+0)];
            fp16 var_3917_promoted_to_fp16 = const()[name = string("op_3917_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 1, 240000]> audio = clip(alpha = var_3916_promoted_to_fp16, beta = var_3917_promoted_to_fp16, x = var_3914_cast_fp16)[name = string("clip_0_cast_fp16")];
        } -> (audio);
}