program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3600.16.1"}, {"coremlc-version", "3600.25.1"}})]
{
    func main<ios18>(tensor<int32, [1]> cache_length, tensor<fp16, [1, 448]> decoder_key_padding_mask, tensor<fp16, [1, 1024, 1, 1500]> encoder_output_embeds, tensor<int32, [1]> input_ids, tensor<fp16, [1, 24576, 1, 448]> key_cache, tensor<fp16, [1, 448]> kv_cache_update_mask, tensor<fp16, [1, 24576, 1, 448]> value_cache) {
            int32 var_64_batch_dims_0 = const()[name = string("op_64_batch_dims_0"), val = int32(0)];
            bool var_64_validate_indices_0 = const()[name = string("op_64_validate_indices_0"), val = bool(false)];
            tensor<fp16, [51865, 1024]> embed_tokens_weight_to_fp16 = const()[name = string("embed_tokens_weight_to_fp16"), val = tensor<fp16, [51865, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            int32 greater_equal_0_y_0 = const()[name = string("greater_equal_0_y_0"), val = int32(0)];
            tensor<bool, [1]> greater_equal_0 = greater_equal(x = input_ids, 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(51865)];
            tensor<int32, [1]> add_0 = add(x = input_ids, y = slice_by_index_0)[name = string("add_0")];
            tensor<int32, [1]> select_0 = select(a = input_ids, b = add_0, cond = greater_equal_0)[name = string("select_0")];
            int32 greater_equal_0_y_0_1 = const()[name = string("greater_equal_0_y_0_1"), val = int32(0)];
            tensor<bool, [1]> greater_equal_0_1 = greater_equal(x = select_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(51865)];
            tensor<int32, [1]> add_0_1 = add(x = select_0, y = slice_by_index_0_1)[name = string("add_0_1")];
            tensor<int32, [1]> select_0_1 = select(a = select_0, b = add_0_1, cond = greater_equal_0_1)[name = string("select_0_1")];
            int32 op_64_cast_fp16_axis_0 = const()[name = string("op_64_cast_fp16_axis_0"), val = int32(0)];
            tensor<fp16, [1, 1024]> op_64_cast_fp16 = gather(axis = op_64_cast_fp16_axis_0, batch_dims = var_64_batch_dims_0, indices = select_0_1, validate_indices = var_64_validate_indices_0, x = embed_tokens_weight_to_fp16)[name = string("op_64_cast_fp16")];
            int32 var_68_axis_0 = const()[name = string("op_68_axis_0"), val = int32(0)];
            int32 var_68_batch_dims_0 = const()[name = string("op_68_batch_dims_0"), val = int32(0)];
            bool var_68_validate_indices_0 = const()[name = string("op_68_validate_indices_0"), val = bool(false)];
            tensor<fp16, [448, 1024]> embed_positions_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [448, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106219648))), lut = tensor<fp16, [28, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106563776))))[name = string("embed_positions_weight_to_fp16_palettized")];
            string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")];
            tensor<uint16, [1]> cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_0")];
            tensor<fp16, [1, 1024]> var_68_cast_fp16_cast_uint16 = gather(axis = var_68_axis_0, batch_dims = var_68_batch_dims_0, indices = cache_length_to_uint16, validate_indices = var_68_validate_indices_0, x = embed_positions_weight_to_fp16_palettized)[name = string("op_68_cast_fp16_cast_uint16")];
            tensor<fp16, [1, 1024]> hidden_states_1_cast_fp16 = add(x = op_64_cast_fp16, y = var_68_cast_fp16_cast_uint16)[name = string("hidden_states_1_cast_fp16")];
            tensor<int32, [1]> var_82_axes_0 = const()[name = string("op_82_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 1024, 1]> var_82_cast_fp16 = expand_dims(axes = var_82_axes_0, x = hidden_states_1_cast_fp16)[name = string("op_82_cast_fp16")];
            tensor<int32, [1]> inputs_1_axes_0 = const()[name = string("inputs_1_axes_0"), val = tensor<int32, [1]>([3])];
            tensor<fp16, [1, 1024, 1, 1]> inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_82_cast_fp16)[name = string("inputs_1_cast_fp16")];
            tensor<int32, [24]> tile_0 = const()[name = string("tile_0"), val = tensor<int32, [24]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106567424)))];
            int32 var_87_axis_0 = const()[name = string("op_87_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_0, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_1, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_2, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_3, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_4, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_5, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_6, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_7, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_8, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_9, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_10, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_11, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_12, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_13, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_14, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_15, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_16, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_17, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_18, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_19, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_20, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_21, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_22, tensor<fp16, [1, 1024, 1, 448]> var_87_cast_fp16_23 = split(axis = var_87_axis_0, split_sizes = tile_0, x = key_cache)[name = string("op_87_cast_fp16")];
            tensor<int32, [24]> tile_1 = const()[name = string("tile_1"), val = tensor<int32, [24]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106567616)))];
            int32 var_114_axis_0 = const()[name = string("op_114_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_0, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_1, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_2, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_3, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_4, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_5, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_6, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_7, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_8, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_9, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_10, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_11, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_12, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_13, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_14, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_15, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_16, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_17, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_18, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_19, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_20, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_21, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_22, tensor<fp16, [1, 1024, 1, 448]> var_114_cast_fp16_23 = split(axis = var_114_axis_0, split_sizes = tile_1, x = value_cache)[name = string("op_114_cast_fp16")];
            int32 var_144 = const()[name = string("op_144"), val = int32(3)];
            tensor<int32, [1]> out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_169_to_fp16 = const()[name = string("op_169_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_169_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")];
            tensor<fp16, [1024]> obj_1_mean_0_to_fp16 = const()[name = string("obj_1_mean_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106567808)))];
            tensor<fp16, [1024]> obj_1_variance_0_to_fp16 = const()[name = string("obj_1_variance_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106569920)))];
            tensor<fp16, [1024]> obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106572032)))];
            tensor<fp16, [1024]> obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106574144)))];
            fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_1_cast_fp16")];
            string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106576256))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107362752))))[name = string("layers_0_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107371008)))];
            tensor<fp16, [1, 1024, 1, 1]> query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = string("query_1_cast_fp16")];
            string current_key_1_pad_type_0 = const()[name = string("current_key_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_1_strides_0 = const()[name = string("current_key_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_1_pad_0 = const()[name = string("current_key_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_1_dilations_0 = const()[name = string("current_key_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_1_groups_0 = const()[name = string("current_key_1_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107373120))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108159616))))[name = string("layers_0_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = string("current_key_1_cast_fp16")];
            string current_value_1_pad_type_0 = const()[name = string("current_value_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_1_strides_0 = const()[name = string("current_value_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_1_pad_0 = const()[name = string("current_value_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_1_dilations_0 = const()[name = string("current_value_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_1_groups_0 = const()[name = string("current_value_1_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108167872))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108954368))))[name = string("layers_0_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108962624)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = string("current_value_1_cast_fp16")];
            tensor<int32, [1]> var_204_axes_0 = const()[name = string("op_204_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 448]> var_204_cast_fp16 = expand_dims(axes = var_204_axes_0, x = kv_cache_update_mask)[name = string("op_204_cast_fp16")];
            tensor<int32, [1]> var_205_axes_0 = const()[name = string("op_205_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 1, 1, 448]> var_205_cast_fp16 = expand_dims(axes = var_205_axes_0, x = var_204_cast_fp16)[name = string("op_205_cast_fp16")];
            fp16 var_145_to_fp16 = const()[name = string("op_145_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 1, 1, 448]> var_207_cast_fp16 = sub(x = var_145_to_fp16, y = var_205_cast_fp16)[name = string("op_207_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_208_cast_fp16 = mul(x = var_87_cast_fp16_0, y = var_207_cast_fp16)[name = string("op_208_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_209_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_205_cast_fp16)[name = string("op_209_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_1_cast_fp16 = add(x = var_208_cast_fp16, y = var_209_cast_fp16)[name = string("key_1_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_212_cast_fp16 = mul(x = var_114_cast_fp16_0, y = var_207_cast_fp16)[name = string("op_212_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_213_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_205_cast_fp16)[name = string("op_213_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_1_cast_fp16 = add(x = var_212_cast_fp16, y = var_213_cast_fp16)[name = string("value_1_cast_fp16")];
            tensor<int32, [4]> var_217 = const()[name = string("op_217"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_1_cast_fp16 = reshape(shape = var_217, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")];
            fp16 var_219_to_fp16 = const()[name = string("op_219_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_220_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_219_to_fp16)[name = string("op_220_cast_fp16")];
            tensor<int32, [4]> var_223 = const()[name = string("op_223"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_224_cast_fp16 = reshape(shape = var_223, x = key_1_cast_fp16)[name = string("op_224_cast_fp16")];
            bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)];
            bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_220_cast_fp16, y = var_224_cast_fp16)[name = string("mh_w_1_cast_fp16")];
            tensor<int32, [1]> var_228_axes_0 = const()[name = string("op_228_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 448]> var_228_cast_fp16 = expand_dims(axes = var_228_axes_0, x = decoder_key_padding_mask)[name = string("op_228_cast_fp16")];
            tensor<int32, [1]> var_229_axes_0 = const()[name = string("op_229_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 1, 1, 448]> var_229_cast_fp16 = expand_dims(axes = var_229_axes_0, x = var_228_cast_fp16)[name = string("op_229_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_3_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_232_cast_fp16 = softmax(axis = var_144, x = mh_w_3_cast_fp16)[name = string("op_232_cast_fp16")];
            tensor<int32, [4]> var_233 = const()[name = string("op_233"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_234_cast_fp16 = reshape(shape = var_233, x = value_1_cast_fp16)[name = string("op_234_cast_fp16")];
            bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)];
            bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_234_cast_fp16, y = var_232_cast_fp16)[name = string("attn_1_cast_fp16")];
            tensor<int32, [4]> var_237 = const()[name = string("op_237"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_1_cast_fp16 = reshape(shape = var_237, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")];
            string obj_7_pad_type_0 = const()[name = string("obj_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_7_strides_0 = const()[name = string("obj_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_7_pad_0 = const()[name = string("obj_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_7_dilations_0 = const()[name = string("obj_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_7_groups_0 = const()[name = string("obj_7_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108964736))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109751232))))[name = string("layers_0_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109759488)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_7_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16_palettized, x = input_1_cast_fp16)[name = string("obj_7_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_7_cast_fp16)[name = string("inputs_3_cast_fp16")];
            tensor<int32, [1]> out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_259_to_fp16 = const()[name = string("op_259_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_259_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")];
            tensor<fp16, [1024]> obj_9_gamma_0_to_fp16 = const()[name = string("obj_9_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109761600)))];
            tensor<fp16, [1024]> obj_9_beta_0_to_fp16 = const()[name = string("obj_9_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109763712)))];
            fp16 obj_9_epsilon_0_to_fp16 = const()[name = string("obj_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_9_cast_fp16")];
            string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109765824))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110552320))))[name = string("layers_0_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110560576)))];
            tensor<fp16, [1, 1024, 1, 1]> query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_9_cast_fp16)[name = string("query_3_cast_fp16")];
            string key_3_pad_type_0 = const()[name = string("key_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_3_strides_0 = const()[name = string("key_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_3_pad_0 = const()[name = string("key_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_3_dilations_0 = const()[name = string("key_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_3_groups_0 = const()[name = string("key_3_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110562688))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111349184))))[name = string("layers_0_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_0_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_3_cast_fp16")];
            string value_3_pad_type_0 = const()[name = string("value_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_3_strides_0 = const()[name = string("value_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_3_pad_0 = const()[name = string("value_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_3_dilations_0 = const()[name = string("value_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_3_groups_0 = const()[name = string("value_3_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111357440))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112143936))))[name = string("layers_0_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112152192)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_3_cast_fp16 = conv(bias = layers_0_encoder_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_0_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_3_cast_fp16")];
            tensor<int32, [4]> var_295 = const()[name = string("op_295"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_3_cast_fp16 = reshape(shape = var_295, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")];
            fp16 var_297_to_fp16 = const()[name = string("op_297_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_298_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_297_to_fp16)[name = string("op_298_cast_fp16")];
            tensor<int32, [4]> var_301 = const()[name = string("op_301"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_302_cast_fp16 = reshape(shape = var_301, x = key_3_cast_fp16)[name = string("op_302_cast_fp16")];
            bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)];
            bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_298_cast_fp16, y = var_302_cast_fp16)[name = string("mh_w_5_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_13_cast_fp16 = softmax(axis = var_144, x = mh_w_5_cast_fp16)[name = string("obj_13_cast_fp16")];
            tensor<int32, [4]> var_306 = const()[name = string("op_306"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_307_cast_fp16 = reshape(shape = var_306, x = value_3_cast_fp16)[name = string("op_307_cast_fp16")];
            bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)];
            bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_307_cast_fp16, y = obj_13_cast_fp16)[name = string("attn_3_cast_fp16")];
            tensor<int32, [4]> var_310 = const()[name = string("op_310"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_3_cast_fp16 = reshape(shape = var_310, x = attn_3_cast_fp16)[name = string("input_3_cast_fp16")];
            string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112154304))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112940800))))[name = string("layers_0_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112949056)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_11_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_3_cast_fp16)[name = string("obj_11_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_5_cast_fp16")];
            tensor<int32, [1]> out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_328_to_fp16 = const()[name = string("op_328_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_328_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")];
            tensor<fp16, [1024]> input_5_gamma_0_to_fp16 = const()[name = string("input_5_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112951168)))];
            tensor<fp16, [1024]> input_5_beta_0_to_fp16 = const()[name = string("input_5_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112953280)))];
            fp16 input_5_epsilon_0_to_fp16 = const()[name = string("input_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_5_cast_fp16")];
            string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_0_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112955392))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116101184))))[name = string("layers_0_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116134016)))];
            tensor<fp16, [1, 4096, 1, 1]> input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16_palettized, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")];
            string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")];
            string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_0_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116142272))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119288064))))[name = string("layers_0_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119296320)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16_palettized, x = input_9_cast_fp16)[name = string("hidden_states_3_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = string("inputs_7_cast_fp16")];
            int32 var_363 = const()[name = string("op_363"), val = int32(3)];
            tensor<int32, [1]> out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_388_to_fp16 = const()[name = string("op_388_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_388_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")];
            tensor<fp16, [1024]> obj_15_gamma_0_to_fp16 = const()[name = string("obj_15_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119298432)))];
            tensor<fp16, [1024]> obj_15_beta_0_to_fp16 = const()[name = string("obj_15_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119300544)))];
            fp16 obj_15_epsilon_0_to_fp16 = const()[name = string("obj_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_15_cast_fp16 = batch_norm(beta = obj_15_beta_0_to_fp16, epsilon = obj_15_epsilon_0_to_fp16, gamma = obj_15_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("obj_15_cast_fp16")];
            string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119302656))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120089152))))[name = string("layers_1_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120097408)))];
            tensor<fp16, [1, 1024, 1, 1]> query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = string("query_5_cast_fp16")];
            string current_key_3_pad_type_0 = const()[name = string("current_key_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_3_strides_0 = const()[name = string("current_key_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_3_pad_0 = const()[name = string("current_key_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_3_dilations_0 = const()[name = string("current_key_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_3_groups_0 = const()[name = string("current_key_3_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120099520))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120886016))))[name = string("layers_1_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_3_cast_fp16 = conv(dilations = current_key_3_dilations_0, groups = current_key_3_groups_0, pad = current_key_3_pad_0, pad_type = current_key_3_pad_type_0, strides = current_key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = string("current_key_3_cast_fp16")];
            string current_value_3_pad_type_0 = const()[name = string("current_value_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_3_strides_0 = const()[name = string("current_value_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_3_pad_0 = const()[name = string("current_value_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_3_dilations_0 = const()[name = string("current_value_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_3_groups_0 = const()[name = string("current_value_3_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120894272))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121680768))))[name = string("layers_1_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121689024)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_3_dilations_0, groups = current_value_3_groups_0, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = current_value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = string("current_value_3_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_427_cast_fp16 = mul(x = var_87_cast_fp16_1, y = var_207_cast_fp16)[name = string("op_427_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_428_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_205_cast_fp16)[name = string("op_428_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_5_cast_fp16 = add(x = var_427_cast_fp16, y = var_428_cast_fp16)[name = string("key_5_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_431_cast_fp16 = mul(x = var_114_cast_fp16_1, y = var_207_cast_fp16)[name = string("op_431_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_432_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_205_cast_fp16)[name = string("op_432_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_5_cast_fp16 = add(x = var_431_cast_fp16, y = var_432_cast_fp16)[name = string("value_5_cast_fp16")];
            tensor<int32, [4]> var_436 = const()[name = string("op_436"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_5_cast_fp16 = reshape(shape = var_436, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")];
            fp16 var_438_to_fp16 = const()[name = string("op_438_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_439_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_438_to_fp16)[name = string("op_439_cast_fp16")];
            tensor<int32, [4]> var_442 = const()[name = string("op_442"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_443_cast_fp16 = reshape(shape = var_442, x = key_5_cast_fp16)[name = string("op_443_cast_fp16")];
            bool mh_w_7_transpose_x_0 = const()[name = string("mh_w_7_transpose_x_0"), val = bool(true)];
            bool mh_w_7_transpose_y_0 = const()[name = string("mh_w_7_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_7_cast_fp16 = matmul(transpose_x = mh_w_7_transpose_x_0, transpose_y = mh_w_7_transpose_y_0, x = var_439_cast_fp16, y = var_443_cast_fp16)[name = string("mh_w_7_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_9_cast_fp16 = add(x = mh_w_7_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_9_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_451_cast_fp16 = softmax(axis = var_363, x = mh_w_9_cast_fp16)[name = string("op_451_cast_fp16")];
            tensor<int32, [4]> var_452 = const()[name = string("op_452"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_453_cast_fp16 = reshape(shape = var_452, x = value_5_cast_fp16)[name = string("op_453_cast_fp16")];
            bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)];
            bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_453_cast_fp16, y = var_451_cast_fp16)[name = string("attn_5_cast_fp16")];
            tensor<int32, [4]> var_456 = const()[name = string("op_456"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_11_cast_fp16 = reshape(shape = var_456, x = attn_5_cast_fp16)[name = string("input_11_cast_fp16")];
            string obj_21_pad_type_0 = const()[name = string("obj_21_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_21_strides_0 = const()[name = string("obj_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_21_pad_0 = const()[name = string("obj_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_21_dilations_0 = const()[name = string("obj_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_21_groups_0 = const()[name = string("obj_21_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121691136))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122477632))))[name = string("layers_1_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122485888)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_21_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16_palettized, x = input_11_cast_fp16)[name = string("obj_21_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_9_cast_fp16")];
            tensor<int32, [1]> out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_478_to_fp16 = const()[name = string("op_478_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_478_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")];
            tensor<fp16, [1024]> obj_23_gamma_0_to_fp16 = const()[name = string("obj_23_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122488000)))];
            tensor<fp16, [1024]> obj_23_beta_0_to_fp16 = const()[name = string("obj_23_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122490112)))];
            fp16 obj_23_epsilon_0_to_fp16 = const()[name = string("obj_23_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_23_cast_fp16 = batch_norm(beta = obj_23_beta_0_to_fp16, epsilon = obj_23_epsilon_0_to_fp16, gamma = obj_23_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_23_cast_fp16")];
            string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122492224))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123278720))))[name = string("layers_1_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123286976)))];
            tensor<fp16, [1, 1024, 1, 1]> query_7_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_23_cast_fp16)[name = string("query_7_cast_fp16")];
            string key_7_pad_type_0 = const()[name = string("key_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_7_strides_0 = const()[name = string("key_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_7_pad_0 = const()[name = string("key_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_7_dilations_0 = const()[name = string("key_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_7_groups_0 = const()[name = string("key_7_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123289088))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124075584))))[name = string("layers_1_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_1_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_7_cast_fp16")];
            string value_7_pad_type_0 = const()[name = string("value_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_7_strides_0 = const()[name = string("value_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_7_pad_0 = const()[name = string("value_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_7_dilations_0 = const()[name = string("value_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_7_groups_0 = const()[name = string("value_7_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124083840))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124870336))))[name = string("layers_1_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124878592)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_7_cast_fp16 = conv(bias = layers_1_encoder_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_1_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_7_cast_fp16")];
            tensor<int32, [4]> var_514 = const()[name = string("op_514"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_7_cast_fp16 = reshape(shape = var_514, x = query_7_cast_fp16)[name = string("mh_q_7_cast_fp16")];
            fp16 var_516_to_fp16 = const()[name = string("op_516_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_517_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_516_to_fp16)[name = string("op_517_cast_fp16")];
            tensor<int32, [4]> var_520 = const()[name = string("op_520"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_521_cast_fp16 = reshape(shape = var_520, x = key_7_cast_fp16)[name = string("op_521_cast_fp16")];
            bool mh_w_11_transpose_x_0 = const()[name = string("mh_w_11_transpose_x_0"), val = bool(true)];
            bool mh_w_11_transpose_y_0 = const()[name = string("mh_w_11_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_11_cast_fp16 = matmul(transpose_x = mh_w_11_transpose_x_0, transpose_y = mh_w_11_transpose_y_0, x = var_517_cast_fp16, y = var_521_cast_fp16)[name = string("mh_w_11_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_27_cast_fp16 = softmax(axis = var_363, x = mh_w_11_cast_fp16)[name = string("obj_27_cast_fp16")];
            tensor<int32, [4]> var_525 = const()[name = string("op_525"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_526_cast_fp16 = reshape(shape = var_525, x = value_7_cast_fp16)[name = string("op_526_cast_fp16")];
            bool attn_7_transpose_x_0 = const()[name = string("attn_7_transpose_x_0"), val = bool(false)];
            bool attn_7_transpose_y_0 = const()[name = string("attn_7_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_526_cast_fp16, y = obj_27_cast_fp16)[name = string("attn_7_cast_fp16")];
            tensor<int32, [4]> var_529 = const()[name = string("op_529"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_13_cast_fp16 = reshape(shape = var_529, x = attn_7_cast_fp16)[name = string("input_13_cast_fp16")];
            string obj_25_pad_type_0 = const()[name = string("obj_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_25_strides_0 = const()[name = string("obj_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_25_pad_0 = const()[name = string("obj_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_25_dilations_0 = const()[name = string("obj_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_25_groups_0 = const()[name = string("obj_25_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124880704))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125667200))))[name = string("layers_1_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125675456)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_25_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_25_dilations_0, groups = obj_25_groups_0, pad = obj_25_pad_0, pad_type = obj_25_pad_type_0, strides = obj_25_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_13_cast_fp16)[name = string("obj_25_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_25_cast_fp16)[name = string("inputs_11_cast_fp16")];
            tensor<int32, [1]> out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_547_to_fp16 = const()[name = string("op_547_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_547_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")];
            tensor<fp16, [1024]> input_15_gamma_0_to_fp16 = const()[name = string("input_15_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125677568)))];
            tensor<fp16, [1024]> input_15_beta_0_to_fp16 = const()[name = string("input_15_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125679680)))];
            fp16 input_15_epsilon_0_to_fp16 = const()[name = string("input_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_15_cast_fp16")];
            string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_17_groups_0 = const()[name = string("input_17_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_1_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125681792))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128827584))))[name = string("layers_1_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128860416)))];
            tensor<fp16, [1, 4096, 1, 1]> input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = string("input_17_cast_fp16")];
            string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = string("input_19_cast_fp16")];
            string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_1_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128868672))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132014464))))[name = string("layers_1_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132022720)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16_palettized, x = input_19_cast_fp16)[name = string("hidden_states_5_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_13_cast_fp16")];
            int32 var_582 = const()[name = string("op_582"), val = int32(3)];
            tensor<int32, [1]> out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_607_to_fp16 = const()[name = string("op_607_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_607_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")];
            tensor<fp16, [1024]> obj_29_gamma_0_to_fp16 = const()[name = string("obj_29_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132024832)))];
            tensor<fp16, [1024]> obj_29_beta_0_to_fp16 = const()[name = string("obj_29_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132026944)))];
            fp16 obj_29_epsilon_0_to_fp16 = const()[name = string("obj_29_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_29_cast_fp16 = batch_norm(beta = obj_29_beta_0_to_fp16, epsilon = obj_29_epsilon_0_to_fp16, gamma = obj_29_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_29_cast_fp16")];
            string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132029056))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132815552))))[name = string("layers_2_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132823808)))];
            tensor<fp16, [1, 1024, 1, 1]> query_9_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = obj_29_cast_fp16)[name = string("query_9_cast_fp16")];
            string current_key_5_pad_type_0 = const()[name = string("current_key_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_5_strides_0 = const()[name = string("current_key_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_5_pad_0 = const()[name = string("current_key_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_5_dilations_0 = const()[name = string("current_key_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_5_groups_0 = const()[name = string("current_key_5_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132825920))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133612416))))[name = string("layers_2_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_5_cast_fp16 = conv(dilations = current_key_5_dilations_0, groups = current_key_5_groups_0, pad = current_key_5_pad_0, pad_type = current_key_5_pad_type_0, strides = current_key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = obj_29_cast_fp16)[name = string("current_key_5_cast_fp16")];
            string current_value_5_pad_type_0 = const()[name = string("current_value_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_5_strides_0 = const()[name = string("current_value_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_5_pad_0 = const()[name = string("current_value_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_5_dilations_0 = const()[name = string("current_value_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_5_groups_0 = const()[name = string("current_value_5_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133620672))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134407168))))[name = string("layers_2_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134415424)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = current_value_5_dilations_0, groups = current_value_5_groups_0, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = current_value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = obj_29_cast_fp16)[name = string("current_value_5_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_646_cast_fp16 = mul(x = var_87_cast_fp16_2, y = var_207_cast_fp16)[name = string("op_646_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_647_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_205_cast_fp16)[name = string("op_647_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_9_cast_fp16 = add(x = var_646_cast_fp16, y = var_647_cast_fp16)[name = string("key_9_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_650_cast_fp16 = mul(x = var_114_cast_fp16_2, y = var_207_cast_fp16)[name = string("op_650_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_651_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_205_cast_fp16)[name = string("op_651_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_9_cast_fp16 = add(x = var_650_cast_fp16, y = var_651_cast_fp16)[name = string("value_9_cast_fp16")];
            tensor<int32, [4]> var_655 = const()[name = string("op_655"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_9_cast_fp16 = reshape(shape = var_655, x = query_9_cast_fp16)[name = string("mh_q_9_cast_fp16")];
            fp16 var_657_to_fp16 = const()[name = string("op_657_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_658_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_657_to_fp16)[name = string("op_658_cast_fp16")];
            tensor<int32, [4]> var_661 = const()[name = string("op_661"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_662_cast_fp16 = reshape(shape = var_661, x = key_9_cast_fp16)[name = string("op_662_cast_fp16")];
            bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)];
            bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_658_cast_fp16, y = var_662_cast_fp16)[name = string("mh_w_13_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_15_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_670_cast_fp16 = softmax(axis = var_582, x = mh_w_15_cast_fp16)[name = string("op_670_cast_fp16")];
            tensor<int32, [4]> var_671 = const()[name = string("op_671"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_672_cast_fp16 = reshape(shape = var_671, x = value_9_cast_fp16)[name = string("op_672_cast_fp16")];
            bool attn_9_transpose_x_0 = const()[name = string("attn_9_transpose_x_0"), val = bool(false)];
            bool attn_9_transpose_y_0 = const()[name = string("attn_9_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_672_cast_fp16, y = var_670_cast_fp16)[name = string("attn_9_cast_fp16")];
            tensor<int32, [4]> var_675 = const()[name = string("op_675"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_21_cast_fp16 = reshape(shape = var_675, x = attn_9_cast_fp16)[name = string("input_21_cast_fp16")];
            string obj_35_pad_type_0 = const()[name = string("obj_35_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_35_strides_0 = const()[name = string("obj_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_35_pad_0 = const()[name = string("obj_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_35_dilations_0 = const()[name = string("obj_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_35_groups_0 = const()[name = string("obj_35_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134417536))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135204032))))[name = string("layers_2_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135212288)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_35_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_35_dilations_0, groups = obj_35_groups_0, pad = obj_35_pad_0, pad_type = obj_35_pad_type_0, strides = obj_35_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16_palettized, x = input_21_cast_fp16)[name = string("obj_35_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_35_cast_fp16)[name = string("inputs_15_cast_fp16")];
            tensor<int32, [1]> out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_697_to_fp16 = const()[name = string("op_697_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_697_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")];
            tensor<fp16, [1024]> obj_37_gamma_0_to_fp16 = const()[name = string("obj_37_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135214400)))];
            tensor<fp16, [1024]> obj_37_beta_0_to_fp16 = const()[name = string("obj_37_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135216512)))];
            fp16 obj_37_epsilon_0_to_fp16 = const()[name = string("obj_37_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_37_cast_fp16 = batch_norm(beta = obj_37_beta_0_to_fp16, epsilon = obj_37_epsilon_0_to_fp16, gamma = obj_37_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("obj_37_cast_fp16")];
            string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135218624))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136005120))))[name = string("layers_2_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_2_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136013376)))];
            tensor<fp16, [1, 1024, 1, 1]> query_11_cast_fp16 = conv(bias = layers_2_encoder_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_2_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_37_cast_fp16)[name = string("query_11_cast_fp16")];
            string key_11_pad_type_0 = const()[name = string("key_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_11_strides_0 = const()[name = string("key_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_11_pad_0 = const()[name = string("key_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_11_dilations_0 = const()[name = string("key_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_11_groups_0 = const()[name = string("key_11_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136015488))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136801984))))[name = string("layers_2_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_11_cast_fp16 = conv(dilations = key_11_dilations_0, groups = key_11_groups_0, pad = key_11_pad_0, pad_type = key_11_pad_type_0, strides = key_11_strides_0, weight = layers_2_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_11_cast_fp16")];
            string value_11_pad_type_0 = const()[name = string("value_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_11_strides_0 = const()[name = string("value_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_11_pad_0 = const()[name = string("value_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_11_dilations_0 = const()[name = string("value_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_11_groups_0 = const()[name = string("value_11_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136810240))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137596736))))[name = string("layers_2_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137604992)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_11_cast_fp16 = conv(bias = layers_2_encoder_attn_v_proj_bias_to_fp16, dilations = value_11_dilations_0, groups = value_11_groups_0, pad = value_11_pad_0, pad_type = value_11_pad_type_0, strides = value_11_strides_0, weight = layers_2_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_11_cast_fp16")];
            tensor<int32, [4]> var_733 = const()[name = string("op_733"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_11_cast_fp16 = reshape(shape = var_733, x = query_11_cast_fp16)[name = string("mh_q_11_cast_fp16")];
            fp16 var_735_to_fp16 = const()[name = string("op_735_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_736_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_735_to_fp16)[name = string("op_736_cast_fp16")];
            tensor<int32, [4]> var_739 = const()[name = string("op_739"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_740_cast_fp16 = reshape(shape = var_739, x = key_11_cast_fp16)[name = string("op_740_cast_fp16")];
            bool mh_w_17_transpose_x_0 = const()[name = string("mh_w_17_transpose_x_0"), val = bool(true)];
            bool mh_w_17_transpose_y_0 = const()[name = string("mh_w_17_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_736_cast_fp16, y = var_740_cast_fp16)[name = string("mh_w_17_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_41_cast_fp16 = softmax(axis = var_582, x = mh_w_17_cast_fp16)[name = string("obj_41_cast_fp16")];
            tensor<int32, [4]> var_744 = const()[name = string("op_744"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_745_cast_fp16 = reshape(shape = var_744, x = value_11_cast_fp16)[name = string("op_745_cast_fp16")];
            bool attn_11_transpose_x_0 = const()[name = string("attn_11_transpose_x_0"), val = bool(false)];
            bool attn_11_transpose_y_0 = const()[name = string("attn_11_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_745_cast_fp16, y = obj_41_cast_fp16)[name = string("attn_11_cast_fp16")];
            tensor<int32, [4]> var_748 = const()[name = string("op_748"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_23_cast_fp16 = reshape(shape = var_748, x = attn_11_cast_fp16)[name = string("input_23_cast_fp16")];
            string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137607104))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138393600))))[name = string("layers_2_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_2_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138401856)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_39_cast_fp16 = conv(bias = layers_2_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_2_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_23_cast_fp16)[name = string("obj_39_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_17_cast_fp16")];
            tensor<int32, [1]> out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_766_to_fp16 = const()[name = string("op_766_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_766_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")];
            tensor<fp16, [1024]> input_25_gamma_0_to_fp16 = const()[name = string("input_25_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138403968)))];
            tensor<fp16, [1024]> input_25_beta_0_to_fp16 = const()[name = string("input_25_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138406080)))];
            fp16 input_25_epsilon_0_to_fp16 = const()[name = string("input_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("input_25_cast_fp16")];
            string input_27_pad_type_0 = const()[name = string("input_27_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_27_strides_0 = const()[name = string("input_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_27_pad_0 = const()[name = string("input_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_27_dilations_0 = const()[name = string("input_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_27_groups_0 = const()[name = string("input_27_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_2_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138408192))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141553984))))[name = string("layers_2_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141586816)))];
            tensor<fp16, [1, 4096, 1, 1]> input_27_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = layers_2_fc1_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = string("input_27_cast_fp16")];
            string input_29_mode_0 = const()[name = string("input_29_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")];
            string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_2_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141595072))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144740864))))[name = string("layers_2_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144749120)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_7_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_2_fc2_weight_to_fp16_palettized, x = input_29_cast_fp16)[name = string("hidden_states_7_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_19_cast_fp16")];
            int32 var_801 = const()[name = string("op_801"), val = int32(3)];
            tensor<int32, [1]> out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_826_to_fp16 = const()[name = string("op_826_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_826_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")];
            tensor<fp16, [1024]> obj_43_gamma_0_to_fp16 = const()[name = string("obj_43_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144751232)))];
            tensor<fp16, [1024]> obj_43_beta_0_to_fp16 = const()[name = string("obj_43_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144753344)))];
            fp16 obj_43_epsilon_0_to_fp16 = const()[name = string("obj_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("obj_43_cast_fp16")];
            string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144755456))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145541952))))[name = string("layers_3_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145550208)))];
            tensor<fp16, [1, 1024, 1, 1]> query_13_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = string("query_13_cast_fp16")];
            string current_key_7_pad_type_0 = const()[name = string("current_key_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_7_strides_0 = const()[name = string("current_key_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_7_pad_0 = const()[name = string("current_key_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_7_dilations_0 = const()[name = string("current_key_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_7_groups_0 = const()[name = string("current_key_7_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145552320))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146338816))))[name = string("layers_3_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_7_cast_fp16 = conv(dilations = current_key_7_dilations_0, groups = current_key_7_groups_0, pad = current_key_7_pad_0, pad_type = current_key_7_pad_type_0, strides = current_key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = string("current_key_7_cast_fp16")];
            string current_value_7_pad_type_0 = const()[name = string("current_value_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_7_strides_0 = const()[name = string("current_value_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_7_pad_0 = const()[name = string("current_value_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_7_dilations_0 = const()[name = string("current_value_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_7_groups_0 = const()[name = string("current_value_7_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146347072))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147133568))))[name = string("layers_3_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147141824)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = current_value_7_dilations_0, groups = current_value_7_groups_0, pad = current_value_7_pad_0, pad_type = current_value_7_pad_type_0, strides = current_value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = string("current_value_7_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_865_cast_fp16 = mul(x = var_87_cast_fp16_3, y = var_207_cast_fp16)[name = string("op_865_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_866_cast_fp16 = mul(x = current_key_7_cast_fp16, y = var_205_cast_fp16)[name = string("op_866_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_13_cast_fp16 = add(x = var_865_cast_fp16, y = var_866_cast_fp16)[name = string("key_13_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_869_cast_fp16 = mul(x = var_114_cast_fp16_3, y = var_207_cast_fp16)[name = string("op_869_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_870_cast_fp16 = mul(x = current_value_7_cast_fp16, y = var_205_cast_fp16)[name = string("op_870_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_13_cast_fp16 = add(x = var_869_cast_fp16, y = var_870_cast_fp16)[name = string("value_13_cast_fp16")];
            tensor<int32, [4]> var_874 = const()[name = string("op_874"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_13_cast_fp16 = reshape(shape = var_874, x = query_13_cast_fp16)[name = string("mh_q_13_cast_fp16")];
            fp16 var_876_to_fp16 = const()[name = string("op_876_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_877_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_876_to_fp16)[name = string("op_877_cast_fp16")];
            tensor<int32, [4]> var_880 = const()[name = string("op_880"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_881_cast_fp16 = reshape(shape = var_880, x = key_13_cast_fp16)[name = string("op_881_cast_fp16")];
            bool mh_w_19_transpose_x_0 = const()[name = string("mh_w_19_transpose_x_0"), val = bool(true)];
            bool mh_w_19_transpose_y_0 = const()[name = string("mh_w_19_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_19_cast_fp16 = matmul(transpose_x = mh_w_19_transpose_x_0, transpose_y = mh_w_19_transpose_y_0, x = var_877_cast_fp16, y = var_881_cast_fp16)[name = string("mh_w_19_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_21_cast_fp16 = add(x = mh_w_19_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_21_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_889_cast_fp16 = softmax(axis = var_801, x = mh_w_21_cast_fp16)[name = string("op_889_cast_fp16")];
            tensor<int32, [4]> var_890 = const()[name = string("op_890"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_891_cast_fp16 = reshape(shape = var_890, x = value_13_cast_fp16)[name = string("op_891_cast_fp16")];
            bool attn_13_transpose_x_0 = const()[name = string("attn_13_transpose_x_0"), val = bool(false)];
            bool attn_13_transpose_y_0 = const()[name = string("attn_13_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_891_cast_fp16, y = var_889_cast_fp16)[name = string("attn_13_cast_fp16")];
            tensor<int32, [4]> var_894 = const()[name = string("op_894"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_31_cast_fp16 = reshape(shape = var_894, x = attn_13_cast_fp16)[name = string("input_31_cast_fp16")];
            string obj_49_pad_type_0 = const()[name = string("obj_49_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_49_strides_0 = const()[name = string("obj_49_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_49_pad_0 = const()[name = string("obj_49_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_49_dilations_0 = const()[name = string("obj_49_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_49_groups_0 = const()[name = string("obj_49_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147143936))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147930432))))[name = string("layers_3_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147938688)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_49_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_49_dilations_0, groups = obj_49_groups_0, pad = obj_49_pad_0, pad_type = obj_49_pad_type_0, strides = obj_49_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16_palettized, x = input_31_cast_fp16)[name = string("obj_49_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = obj_49_cast_fp16)[name = string("inputs_21_cast_fp16")];
            tensor<int32, [1]> out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_916_to_fp16 = const()[name = string("op_916_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_916_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")];
            tensor<fp16, [1024]> obj_51_gamma_0_to_fp16 = const()[name = string("obj_51_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147940800)))];
            tensor<fp16, [1024]> obj_51_beta_0_to_fp16 = const()[name = string("obj_51_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147942912)))];
            fp16 obj_51_epsilon_0_to_fp16 = const()[name = string("obj_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_51_cast_fp16")];
            string query_15_pad_type_0 = const()[name = string("query_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_15_strides_0 = const()[name = string("query_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_15_pad_0 = const()[name = string("query_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_15_dilations_0 = const()[name = string("query_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_15_groups_0 = const()[name = string("query_15_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147945024))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148731520))))[name = string("layers_3_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_3_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148739776)))];
            tensor<fp16, [1, 1024, 1, 1]> query_15_cast_fp16 = conv(bias = layers_3_encoder_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_3_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_51_cast_fp16)[name = string("query_15_cast_fp16")];
            string key_15_pad_type_0 = const()[name = string("key_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_15_strides_0 = const()[name = string("key_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_15_pad_0 = const()[name = string("key_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_15_dilations_0 = const()[name = string("key_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_15_groups_0 = const()[name = string("key_15_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148741888))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149528384))))[name = string("layers_3_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_15_cast_fp16 = conv(dilations = key_15_dilations_0, groups = key_15_groups_0, pad = key_15_pad_0, pad_type = key_15_pad_type_0, strides = key_15_strides_0, weight = layers_3_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_15_cast_fp16")];
            string value_15_pad_type_0 = const()[name = string("value_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_15_strides_0 = const()[name = string("value_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_15_pad_0 = const()[name = string("value_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_15_dilations_0 = const()[name = string("value_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_15_groups_0 = const()[name = string("value_15_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149536640))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150323136))))[name = string("layers_3_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150331392)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_15_cast_fp16 = conv(bias = layers_3_encoder_attn_v_proj_bias_to_fp16, dilations = value_15_dilations_0, groups = value_15_groups_0, pad = value_15_pad_0, pad_type = value_15_pad_type_0, strides = value_15_strides_0, weight = layers_3_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_15_cast_fp16")];
            tensor<int32, [4]> var_952 = const()[name = string("op_952"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_15_cast_fp16 = reshape(shape = var_952, x = query_15_cast_fp16)[name = string("mh_q_15_cast_fp16")];
            fp16 var_954_to_fp16 = const()[name = string("op_954_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_955_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_954_to_fp16)[name = string("op_955_cast_fp16")];
            tensor<int32, [4]> var_958 = const()[name = string("op_958"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_959_cast_fp16 = reshape(shape = var_958, x = key_15_cast_fp16)[name = string("op_959_cast_fp16")];
            bool mh_w_23_transpose_x_0 = const()[name = string("mh_w_23_transpose_x_0"), val = bool(true)];
            bool mh_w_23_transpose_y_0 = const()[name = string("mh_w_23_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_23_cast_fp16 = matmul(transpose_x = mh_w_23_transpose_x_0, transpose_y = mh_w_23_transpose_y_0, x = var_955_cast_fp16, y = var_959_cast_fp16)[name = string("mh_w_23_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_55_cast_fp16 = softmax(axis = var_801, x = mh_w_23_cast_fp16)[name = string("obj_55_cast_fp16")];
            tensor<int32, [4]> var_963 = const()[name = string("op_963"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_964_cast_fp16 = reshape(shape = var_963, x = value_15_cast_fp16)[name = string("op_964_cast_fp16")];
            bool attn_15_transpose_x_0 = const()[name = string("attn_15_transpose_x_0"), val = bool(false)];
            bool attn_15_transpose_y_0 = const()[name = string("attn_15_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_964_cast_fp16, y = obj_55_cast_fp16)[name = string("attn_15_cast_fp16")];
            tensor<int32, [4]> var_967 = const()[name = string("op_967"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_33_cast_fp16 = reshape(shape = var_967, x = attn_15_cast_fp16)[name = string("input_33_cast_fp16")];
            string obj_53_pad_type_0 = const()[name = string("obj_53_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_53_strides_0 = const()[name = string("obj_53_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_53_pad_0 = const()[name = string("obj_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_53_dilations_0 = const()[name = string("obj_53_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_53_groups_0 = const()[name = string("obj_53_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150333504))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151120000))))[name = string("layers_3_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_3_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151128256)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_53_cast_fp16 = conv(bias = layers_3_encoder_attn_o_proj_bias_to_fp16, dilations = obj_53_dilations_0, groups = obj_53_groups_0, pad = obj_53_pad_0, pad_type = obj_53_pad_type_0, strides = obj_53_strides_0, weight = layers_3_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_33_cast_fp16)[name = string("obj_53_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_53_cast_fp16)[name = string("inputs_23_cast_fp16")];
            tensor<int32, [1]> out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_985_to_fp16 = const()[name = string("op_985_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_985_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")];
            tensor<fp16, [1024]> input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151130368)))];
            tensor<fp16, [1024]> input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151132480)))];
            fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_35_cast_fp16")];
            string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_3_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151134592))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154280384))))[name = string("layers_3_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154313216)))];
            tensor<fp16, [1, 4096, 1, 1]> input_37_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_3_fc1_weight_to_fp16_palettized, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")];
            string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")];
            string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_3_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154321472))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157467264))))[name = string("layers_3_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157475520)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_9_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_3_fc2_weight_to_fp16_palettized, x = input_39_cast_fp16)[name = string("hidden_states_9_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_25_cast_fp16")];
            int32 var_1020 = const()[name = string("op_1020"), val = int32(3)];
            tensor<int32, [1]> out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1045_to_fp16 = const()[name = string("op_1045_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_1045_to_fp16, x = inputs_25_cast_fp16)[name = string("out_25_cast_fp16")];
            tensor<fp16, [1024]> obj_57_gamma_0_to_fp16 = const()[name = string("obj_57_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157477632)))];
            tensor<fp16, [1024]> obj_57_beta_0_to_fp16 = const()[name = string("obj_57_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157479744)))];
            fp16 obj_57_epsilon_0_to_fp16 = const()[name = string("obj_57_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_57_cast_fp16 = batch_norm(beta = obj_57_beta_0_to_fp16, epsilon = obj_57_epsilon_0_to_fp16, gamma = obj_57_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_25_cast_fp16)[name = string("obj_57_cast_fp16")];
            string query_17_pad_type_0 = const()[name = string("query_17_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_17_strides_0 = const()[name = string("query_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_17_pad_0 = const()[name = string("query_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_17_dilations_0 = const()[name = string("query_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_17_groups_0 = const()[name = string("query_17_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157481856))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158268352))))[name = string("layers_4_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158276608)))];
            tensor<fp16, [1, 1024, 1, 1]> query_17_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = obj_57_cast_fp16)[name = string("query_17_cast_fp16")];
            string current_key_9_pad_type_0 = const()[name = string("current_key_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_9_strides_0 = const()[name = string("current_key_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_9_pad_0 = const()[name = string("current_key_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_9_dilations_0 = const()[name = string("current_key_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_9_groups_0 = const()[name = string("current_key_9_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158278720))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159065216))))[name = string("layers_4_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_9_cast_fp16 = conv(dilations = current_key_9_dilations_0, groups = current_key_9_groups_0, pad = current_key_9_pad_0, pad_type = current_key_9_pad_type_0, strides = current_key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = obj_57_cast_fp16)[name = string("current_key_9_cast_fp16")];
            string current_value_9_pad_type_0 = const()[name = string("current_value_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_9_strides_0 = const()[name = string("current_value_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_9_pad_0 = const()[name = string("current_value_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_9_dilations_0 = const()[name = string("current_value_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_9_groups_0 = const()[name = string("current_value_9_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159073472))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159859968))))[name = string("layers_4_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159868224)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = current_value_9_dilations_0, groups = current_value_9_groups_0, pad = current_value_9_pad_0, pad_type = current_value_9_pad_type_0, strides = current_value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = obj_57_cast_fp16)[name = string("current_value_9_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1084_cast_fp16 = mul(x = var_87_cast_fp16_4, y = var_207_cast_fp16)[name = string("op_1084_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1085_cast_fp16 = mul(x = current_key_9_cast_fp16, y = var_205_cast_fp16)[name = string("op_1085_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_17_cast_fp16 = add(x = var_1084_cast_fp16, y = var_1085_cast_fp16)[name = string("key_17_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1088_cast_fp16 = mul(x = var_114_cast_fp16_4, y = var_207_cast_fp16)[name = string("op_1088_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1089_cast_fp16 = mul(x = current_value_9_cast_fp16, y = var_205_cast_fp16)[name = string("op_1089_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_17_cast_fp16 = add(x = var_1088_cast_fp16, y = var_1089_cast_fp16)[name = string("value_17_cast_fp16")];
            tensor<int32, [4]> var_1093 = const()[name = string("op_1093"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_17_cast_fp16 = reshape(shape = var_1093, x = query_17_cast_fp16)[name = string("mh_q_17_cast_fp16")];
            fp16 var_1095_to_fp16 = const()[name = string("op_1095_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_1096_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_1095_to_fp16)[name = string("op_1096_cast_fp16")];
            tensor<int32, [4]> var_1099 = const()[name = string("op_1099"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_1100_cast_fp16 = reshape(shape = var_1099, x = key_17_cast_fp16)[name = string("op_1100_cast_fp16")];
            bool mh_w_25_transpose_x_0 = const()[name = string("mh_w_25_transpose_x_0"), val = bool(true)];
            bool mh_w_25_transpose_y_0 = const()[name = string("mh_w_25_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_1096_cast_fp16, y = var_1100_cast_fp16)[name = string("mh_w_25_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_27_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_1108_cast_fp16 = softmax(axis = var_1020, x = mh_w_27_cast_fp16)[name = string("op_1108_cast_fp16")];
            tensor<int32, [4]> var_1109 = const()[name = string("op_1109"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_1110_cast_fp16 = reshape(shape = var_1109, x = value_17_cast_fp16)[name = string("op_1110_cast_fp16")];
            bool attn_17_transpose_x_0 = const()[name = string("attn_17_transpose_x_0"), val = bool(false)];
            bool attn_17_transpose_y_0 = const()[name = string("attn_17_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1110_cast_fp16, y = var_1108_cast_fp16)[name = string("attn_17_cast_fp16")];
            tensor<int32, [4]> var_1113 = const()[name = string("op_1113"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_41_cast_fp16 = reshape(shape = var_1113, x = attn_17_cast_fp16)[name = string("input_41_cast_fp16")];
            string obj_63_pad_type_0 = const()[name = string("obj_63_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_63_strides_0 = const()[name = string("obj_63_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_63_pad_0 = const()[name = string("obj_63_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_63_dilations_0 = const()[name = string("obj_63_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_63_groups_0 = const()[name = string("obj_63_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159870336))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160656832))))[name = string("layers_4_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160665088)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_63_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_63_dilations_0, groups = obj_63_groups_0, pad = obj_63_pad_0, pad_type = obj_63_pad_type_0, strides = obj_63_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = string("obj_63_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_63_cast_fp16)[name = string("inputs_27_cast_fp16")];
            tensor<int32, [1]> out_27_axes_0 = const()[name = string("out_27_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1135_to_fp16 = const()[name = string("op_1135_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1135_to_fp16, x = inputs_27_cast_fp16)[name = string("out_27_cast_fp16")];
            tensor<fp16, [1024]> obj_65_gamma_0_to_fp16 = const()[name = string("obj_65_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160667200)))];
            tensor<fp16, [1024]> obj_65_beta_0_to_fp16 = const()[name = string("obj_65_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160669312)))];
            fp16 obj_65_epsilon_0_to_fp16 = const()[name = string("obj_65_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_65_cast_fp16 = batch_norm(beta = obj_65_beta_0_to_fp16, epsilon = obj_65_epsilon_0_to_fp16, gamma = obj_65_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_27_cast_fp16)[name = string("obj_65_cast_fp16")];
            string query_19_pad_type_0 = const()[name = string("query_19_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_19_strides_0 = const()[name = string("query_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_19_pad_0 = const()[name = string("query_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_19_dilations_0 = const()[name = string("query_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_19_groups_0 = const()[name = string("query_19_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160671424))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161457920))))[name = string("layers_4_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_4_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161466176)))];
            tensor<fp16, [1, 1024, 1, 1]> query_19_cast_fp16 = conv(bias = layers_4_encoder_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_4_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_65_cast_fp16)[name = string("query_19_cast_fp16")];
            string key_19_pad_type_0 = const()[name = string("key_19_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_19_strides_0 = const()[name = string("key_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_19_pad_0 = const()[name = string("key_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_19_dilations_0 = const()[name = string("key_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_19_groups_0 = const()[name = string("key_19_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161468288))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162254784))))[name = string("layers_4_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_19_cast_fp16 = conv(dilations = key_19_dilations_0, groups = key_19_groups_0, pad = key_19_pad_0, pad_type = key_19_pad_type_0, strides = key_19_strides_0, weight = layers_4_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_19_cast_fp16")];
            string value_19_pad_type_0 = const()[name = string("value_19_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_19_strides_0 = const()[name = string("value_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_19_pad_0 = const()[name = string("value_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_19_dilations_0 = const()[name = string("value_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_19_groups_0 = const()[name = string("value_19_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162263040))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163049536))))[name = string("layers_4_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_4_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163057792)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_19_cast_fp16 = conv(bias = layers_4_encoder_attn_v_proj_bias_to_fp16, dilations = value_19_dilations_0, groups = value_19_groups_0, pad = value_19_pad_0, pad_type = value_19_pad_type_0, strides = value_19_strides_0, weight = layers_4_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_19_cast_fp16")];
            tensor<int32, [4]> var_1171 = const()[name = string("op_1171"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_19_cast_fp16 = reshape(shape = var_1171, x = query_19_cast_fp16)[name = string("mh_q_19_cast_fp16")];
            fp16 var_1173_to_fp16 = const()[name = string("op_1173_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_1174_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1173_to_fp16)[name = string("op_1174_cast_fp16")];
            tensor<int32, [4]> var_1177 = const()[name = string("op_1177"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1178_cast_fp16 = reshape(shape = var_1177, x = key_19_cast_fp16)[name = string("op_1178_cast_fp16")];
            bool mh_w_29_transpose_x_0 = const()[name = string("mh_w_29_transpose_x_0"), val = bool(true)];
            bool mh_w_29_transpose_y_0 = const()[name = string("mh_w_29_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_1174_cast_fp16, y = var_1178_cast_fp16)[name = string("mh_w_29_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_69_cast_fp16 = softmax(axis = var_1020, x = mh_w_29_cast_fp16)[name = string("obj_69_cast_fp16")];
            tensor<int32, [4]> var_1182 = const()[name = string("op_1182"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1183_cast_fp16 = reshape(shape = var_1182, x = value_19_cast_fp16)[name = string("op_1183_cast_fp16")];
            bool attn_19_transpose_x_0 = const()[name = string("attn_19_transpose_x_0"), val = bool(false)];
            bool attn_19_transpose_y_0 = const()[name = string("attn_19_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1183_cast_fp16, y = obj_69_cast_fp16)[name = string("attn_19_cast_fp16")];
            tensor<int32, [4]> var_1186 = const()[name = string("op_1186"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_43_cast_fp16 = reshape(shape = var_1186, x = attn_19_cast_fp16)[name = string("input_43_cast_fp16")];
            string obj_67_pad_type_0 = const()[name = string("obj_67_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_67_strides_0 = const()[name = string("obj_67_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_67_pad_0 = const()[name = string("obj_67_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_67_dilations_0 = const()[name = string("obj_67_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_67_groups_0 = const()[name = string("obj_67_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163059904))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163846400))))[name = string("layers_4_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_4_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163854656)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_67_cast_fp16 = conv(bias = layers_4_encoder_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_4_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_43_cast_fp16)[name = string("obj_67_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = obj_67_cast_fp16)[name = string("inputs_29_cast_fp16")];
            tensor<int32, [1]> out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1204_to_fp16 = const()[name = string("op_1204_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1204_to_fp16, x = inputs_29_cast_fp16)[name = string("out_29_cast_fp16")];
            tensor<fp16, [1024]> input_45_gamma_0_to_fp16 = const()[name = string("input_45_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163856768)))];
            tensor<fp16, [1024]> input_45_beta_0_to_fp16 = const()[name = string("input_45_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163858880)))];
            fp16 input_45_epsilon_0_to_fp16 = const()[name = string("input_45_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_45_cast_fp16 = batch_norm(beta = input_45_beta_0_to_fp16, epsilon = input_45_epsilon_0_to_fp16, gamma = input_45_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_29_cast_fp16)[name = string("input_45_cast_fp16")];
            string input_47_pad_type_0 = const()[name = string("input_47_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_47_strides_0 = const()[name = string("input_47_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_47_dilations_0 = const()[name = string("input_47_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_47_groups_0 = const()[name = string("input_47_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_4_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163860992))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167006784))))[name = string("layers_4_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167039616)))];
            tensor<fp16, [1, 4096, 1, 1]> input_47_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_47_dilations_0, groups = input_47_groups_0, pad = input_47_pad_0, pad_type = input_47_pad_type_0, strides = input_47_strides_0, weight = layers_4_fc1_weight_to_fp16_palettized, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")];
            string input_49_mode_0 = const()[name = string("input_49_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_49_cast_fp16 = gelu(mode = input_49_mode_0, x = input_47_cast_fp16)[name = string("input_49_cast_fp16")];
            string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_4_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167047872))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170193664))))[name = string("layers_4_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170201920)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_11_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_4_fc2_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = string("hidden_states_11_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_31_cast_fp16")];
            int32 var_1239 = const()[name = string("op_1239"), val = int32(3)];
            tensor<int32, [1]> out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1264_to_fp16 = const()[name = string("op_1264_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1264_to_fp16, x = inputs_31_cast_fp16)[name = string("out_31_cast_fp16")];
            tensor<fp16, [1024]> obj_71_gamma_0_to_fp16 = const()[name = string("obj_71_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170204032)))];
            tensor<fp16, [1024]> obj_71_beta_0_to_fp16 = const()[name = string("obj_71_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170206144)))];
            fp16 obj_71_epsilon_0_to_fp16 = const()[name = string("obj_71_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_71_cast_fp16 = batch_norm(beta = obj_71_beta_0_to_fp16, epsilon = obj_71_epsilon_0_to_fp16, gamma = obj_71_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_31_cast_fp16)[name = string("obj_71_cast_fp16")];
            string query_21_pad_type_0 = const()[name = string("query_21_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_21_strides_0 = const()[name = string("query_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_21_pad_0 = const()[name = string("query_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_21_dilations_0 = const()[name = string("query_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_21_groups_0 = const()[name = string("query_21_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170208256))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170994752))))[name = string("layers_5_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171003008)))];
            tensor<fp16, [1, 1024, 1, 1]> query_21_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16_palettized, x = obj_71_cast_fp16)[name = string("query_21_cast_fp16")];
            string current_key_11_pad_type_0 = const()[name = string("current_key_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_11_strides_0 = const()[name = string("current_key_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_11_pad_0 = const()[name = string("current_key_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_11_dilations_0 = const()[name = string("current_key_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_11_groups_0 = const()[name = string("current_key_11_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171005120))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171791616))))[name = string("layers_5_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_11_cast_fp16 = conv(dilations = current_key_11_dilations_0, groups = current_key_11_groups_0, pad = current_key_11_pad_0, pad_type = current_key_11_pad_type_0, strides = current_key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16_palettized, x = obj_71_cast_fp16)[name = string("current_key_11_cast_fp16")];
            string current_value_11_pad_type_0 = const()[name = string("current_value_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_11_strides_0 = const()[name = string("current_value_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_11_pad_0 = const()[name = string("current_value_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_11_dilations_0 = const()[name = string("current_value_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_11_groups_0 = const()[name = string("current_value_11_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171799872))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172586368))))[name = string("layers_5_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172594624)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = current_value_11_dilations_0, groups = current_value_11_groups_0, pad = current_value_11_pad_0, pad_type = current_value_11_pad_type_0, strides = current_value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16_palettized, x = obj_71_cast_fp16)[name = string("current_value_11_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1303_cast_fp16 = mul(x = var_87_cast_fp16_5, y = var_207_cast_fp16)[name = string("op_1303_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1304_cast_fp16 = mul(x = current_key_11_cast_fp16, y = var_205_cast_fp16)[name = string("op_1304_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_21_cast_fp16 = add(x = var_1303_cast_fp16, y = var_1304_cast_fp16)[name = string("key_21_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1307_cast_fp16 = mul(x = var_114_cast_fp16_5, y = var_207_cast_fp16)[name = string("op_1307_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1308_cast_fp16 = mul(x = current_value_11_cast_fp16, y = var_205_cast_fp16)[name = string("op_1308_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_21_cast_fp16 = add(x = var_1307_cast_fp16, y = var_1308_cast_fp16)[name = string("value_21_cast_fp16")];
            tensor<int32, [4]> var_1312 = const()[name = string("op_1312"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_21_cast_fp16 = reshape(shape = var_1312, x = query_21_cast_fp16)[name = string("mh_q_21_cast_fp16")];
            fp16 var_1314_to_fp16 = const()[name = string("op_1314_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_1315_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_1314_to_fp16)[name = string("op_1315_cast_fp16")];
            tensor<int32, [4]> var_1318 = const()[name = string("op_1318"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_1319_cast_fp16 = reshape(shape = var_1318, x = key_21_cast_fp16)[name = string("op_1319_cast_fp16")];
            bool mh_w_31_transpose_x_0 = const()[name = string("mh_w_31_transpose_x_0"), val = bool(true)];
            bool mh_w_31_transpose_y_0 = const()[name = string("mh_w_31_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_31_cast_fp16 = matmul(transpose_x = mh_w_31_transpose_x_0, transpose_y = mh_w_31_transpose_y_0, x = var_1315_cast_fp16, y = var_1319_cast_fp16)[name = string("mh_w_31_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_33_cast_fp16 = add(x = mh_w_31_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_33_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_1327_cast_fp16 = softmax(axis = var_1239, x = mh_w_33_cast_fp16)[name = string("op_1327_cast_fp16")];
            tensor<int32, [4]> var_1328 = const()[name = string("op_1328"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_1329_cast_fp16 = reshape(shape = var_1328, x = value_21_cast_fp16)[name = string("op_1329_cast_fp16")];
            bool attn_21_transpose_x_0 = const()[name = string("attn_21_transpose_x_0"), val = bool(false)];
            bool attn_21_transpose_y_0 = const()[name = string("attn_21_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_1329_cast_fp16, y = var_1327_cast_fp16)[name = string("attn_21_cast_fp16")];
            tensor<int32, [4]> var_1332 = const()[name = string("op_1332"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_51_cast_fp16 = reshape(shape = var_1332, x = attn_21_cast_fp16)[name = string("input_51_cast_fp16")];
            string obj_77_pad_type_0 = const()[name = string("obj_77_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_77_strides_0 = const()[name = string("obj_77_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_77_pad_0 = const()[name = string("obj_77_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_77_dilations_0 = const()[name = string("obj_77_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_77_groups_0 = const()[name = string("obj_77_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172596736))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173383232))))[name = string("layers_5_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173391488)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_77_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_77_dilations_0, groups = obj_77_groups_0, pad = obj_77_pad_0, pad_type = obj_77_pad_type_0, strides = obj_77_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16_palettized, x = input_51_cast_fp16)[name = string("obj_77_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = obj_77_cast_fp16)[name = string("inputs_33_cast_fp16")];
            tensor<int32, [1]> out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1354_to_fp16 = const()[name = string("op_1354_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1354_to_fp16, x = inputs_33_cast_fp16)[name = string("out_33_cast_fp16")];
            tensor<fp16, [1024]> obj_79_gamma_0_to_fp16 = const()[name = string("obj_79_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173393600)))];
            tensor<fp16, [1024]> obj_79_beta_0_to_fp16 = const()[name = string("obj_79_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173395712)))];
            fp16 obj_79_epsilon_0_to_fp16 = const()[name = string("obj_79_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_79_cast_fp16 = batch_norm(beta = obj_79_beta_0_to_fp16, epsilon = obj_79_epsilon_0_to_fp16, gamma = obj_79_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_33_cast_fp16)[name = string("obj_79_cast_fp16")];
            string query_23_pad_type_0 = const()[name = string("query_23_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_23_strides_0 = const()[name = string("query_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_23_pad_0 = const()[name = string("query_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_23_dilations_0 = const()[name = string("query_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_23_groups_0 = const()[name = string("query_23_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173397824))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174184320))))[name = string("layers_5_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_5_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174192576)))];
            tensor<fp16, [1, 1024, 1, 1]> query_23_cast_fp16 = conv(bias = layers_5_encoder_attn_q_proj_bias_to_fp16, dilations = query_23_dilations_0, groups = query_23_groups_0, pad = query_23_pad_0, pad_type = query_23_pad_type_0, strides = query_23_strides_0, weight = layers_5_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_79_cast_fp16)[name = string("query_23_cast_fp16")];
            string key_23_pad_type_0 = const()[name = string("key_23_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_23_strides_0 = const()[name = string("key_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_23_pad_0 = const()[name = string("key_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_23_dilations_0 = const()[name = string("key_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_23_groups_0 = const()[name = string("key_23_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174194688))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174981184))))[name = string("layers_5_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_23_cast_fp16 = conv(dilations = key_23_dilations_0, groups = key_23_groups_0, pad = key_23_pad_0, pad_type = key_23_pad_type_0, strides = key_23_strides_0, weight = layers_5_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_23_cast_fp16")];
            string value_23_pad_type_0 = const()[name = string("value_23_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_23_strides_0 = const()[name = string("value_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_23_pad_0 = const()[name = string("value_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_23_dilations_0 = const()[name = string("value_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_23_groups_0 = const()[name = string("value_23_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174989440))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175775936))))[name = string("layers_5_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_5_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175784192)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_23_cast_fp16 = conv(bias = layers_5_encoder_attn_v_proj_bias_to_fp16, dilations = value_23_dilations_0, groups = value_23_groups_0, pad = value_23_pad_0, pad_type = value_23_pad_type_0, strides = value_23_strides_0, weight = layers_5_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_23_cast_fp16")];
            tensor<int32, [4]> var_1390 = const()[name = string("op_1390"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_23_cast_fp16 = reshape(shape = var_1390, x = query_23_cast_fp16)[name = string("mh_q_23_cast_fp16")];
            fp16 var_1392_to_fp16 = const()[name = string("op_1392_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_1393_cast_fp16 = mul(x = mh_q_23_cast_fp16, y = var_1392_to_fp16)[name = string("op_1393_cast_fp16")];
            tensor<int32, [4]> var_1396 = const()[name = string("op_1396"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1397_cast_fp16 = reshape(shape = var_1396, x = key_23_cast_fp16)[name = string("op_1397_cast_fp16")];
            bool mh_w_35_transpose_x_0 = const()[name = string("mh_w_35_transpose_x_0"), val = bool(true)];
            bool mh_w_35_transpose_y_0 = const()[name = string("mh_w_35_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_35_cast_fp16 = matmul(transpose_x = mh_w_35_transpose_x_0, transpose_y = mh_w_35_transpose_y_0, x = var_1393_cast_fp16, y = var_1397_cast_fp16)[name = string("mh_w_35_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_83_cast_fp16 = softmax(axis = var_1239, x = mh_w_35_cast_fp16)[name = string("obj_83_cast_fp16")];
            tensor<int32, [4]> var_1401 = const()[name = string("op_1401"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1402_cast_fp16 = reshape(shape = var_1401, x = value_23_cast_fp16)[name = string("op_1402_cast_fp16")];
            bool attn_23_transpose_x_0 = const()[name = string("attn_23_transpose_x_0"), val = bool(false)];
            bool attn_23_transpose_y_0 = const()[name = string("attn_23_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_23_cast_fp16 = matmul(transpose_x = attn_23_transpose_x_0, transpose_y = attn_23_transpose_y_0, x = var_1402_cast_fp16, y = obj_83_cast_fp16)[name = string("attn_23_cast_fp16")];
            tensor<int32, [4]> var_1405 = const()[name = string("op_1405"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_53_cast_fp16 = reshape(shape = var_1405, x = attn_23_cast_fp16)[name = string("input_53_cast_fp16")];
            string obj_81_pad_type_0 = const()[name = string("obj_81_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_81_strides_0 = const()[name = string("obj_81_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_81_pad_0 = const()[name = string("obj_81_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_81_dilations_0 = const()[name = string("obj_81_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_81_groups_0 = const()[name = string("obj_81_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175786304))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176572800))))[name = string("layers_5_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_5_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176581056)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_81_cast_fp16 = conv(bias = layers_5_encoder_attn_o_proj_bias_to_fp16, dilations = obj_81_dilations_0, groups = obj_81_groups_0, pad = obj_81_pad_0, pad_type = obj_81_pad_type_0, strides = obj_81_strides_0, weight = layers_5_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_53_cast_fp16)[name = string("obj_81_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_81_cast_fp16)[name = string("inputs_35_cast_fp16")];
            tensor<int32, [1]> out_35_axes_0 = const()[name = string("out_35_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1423_to_fp16 = const()[name = string("op_1423_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1423_to_fp16, x = inputs_35_cast_fp16)[name = string("out_35_cast_fp16")];
            tensor<fp16, [1024]> input_55_gamma_0_to_fp16 = const()[name = string("input_55_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176583168)))];
            tensor<fp16, [1024]> input_55_beta_0_to_fp16 = const()[name = string("input_55_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176585280)))];
            fp16 input_55_epsilon_0_to_fp16 = const()[name = string("input_55_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_55_cast_fp16 = batch_norm(beta = input_55_beta_0_to_fp16, epsilon = input_55_epsilon_0_to_fp16, gamma = input_55_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_35_cast_fp16)[name = string("input_55_cast_fp16")];
            string input_57_pad_type_0 = const()[name = string("input_57_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_57_strides_0 = const()[name = string("input_57_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_57_pad_0 = const()[name = string("input_57_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_57_dilations_0 = const()[name = string("input_57_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_57_groups_0 = const()[name = string("input_57_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_5_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176587392))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179733184))))[name = string("layers_5_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179766016)))];
            tensor<fp16, [1, 4096, 1, 1]> input_57_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_57_dilations_0, groups = input_57_groups_0, pad = input_57_pad_0, pad_type = input_57_pad_type_0, strides = input_57_strides_0, weight = layers_5_fc1_weight_to_fp16_palettized, x = input_55_cast_fp16)[name = string("input_57_cast_fp16")];
            string input_59_mode_0 = const()[name = string("input_59_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_59_cast_fp16 = gelu(mode = input_59_mode_0, x = input_57_cast_fp16)[name = string("input_59_cast_fp16")];
            string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_5_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179774272))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182920064))))[name = string("layers_5_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182928320)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_13_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_5_fc2_weight_to_fp16_palettized, x = input_59_cast_fp16)[name = string("hidden_states_13_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_37_cast_fp16")];
            int32 var_1458 = const()[name = string("op_1458"), val = int32(3)];
            tensor<int32, [1]> out_37_axes_0 = const()[name = string("out_37_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1483_to_fp16 = const()[name = string("op_1483_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1483_to_fp16, x = inputs_37_cast_fp16)[name = string("out_37_cast_fp16")];
            tensor<fp16, [1024]> obj_85_gamma_0_to_fp16 = const()[name = string("obj_85_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182930432)))];
            tensor<fp16, [1024]> obj_85_beta_0_to_fp16 = const()[name = string("obj_85_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182932544)))];
            fp16 obj_85_epsilon_0_to_fp16 = const()[name = string("obj_85_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_85_cast_fp16 = batch_norm(beta = obj_85_beta_0_to_fp16, epsilon = obj_85_epsilon_0_to_fp16, gamma = obj_85_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_37_cast_fp16)[name = string("obj_85_cast_fp16")];
            string query_25_pad_type_0 = const()[name = string("query_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_25_strides_0 = const()[name = string("query_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_25_pad_0 = const()[name = string("query_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_25_dilations_0 = const()[name = string("query_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_25_groups_0 = const()[name = string("query_25_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182934656))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183721152))))[name = string("layers_6_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183729408)))];
            tensor<fp16, [1, 1024, 1, 1]> query_25_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_25_dilations_0, groups = query_25_groups_0, pad = query_25_pad_0, pad_type = query_25_pad_type_0, strides = query_25_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16_palettized, x = obj_85_cast_fp16)[name = string("query_25_cast_fp16")];
            string current_key_13_pad_type_0 = const()[name = string("current_key_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_13_strides_0 = const()[name = string("current_key_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_13_pad_0 = const()[name = string("current_key_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_13_dilations_0 = const()[name = string("current_key_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_13_groups_0 = const()[name = string("current_key_13_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183731520))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184518016))))[name = string("layers_6_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_13_cast_fp16 = conv(dilations = current_key_13_dilations_0, groups = current_key_13_groups_0, pad = current_key_13_pad_0, pad_type = current_key_13_pad_type_0, strides = current_key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16_palettized, x = obj_85_cast_fp16)[name = string("current_key_13_cast_fp16")];
            string current_value_13_pad_type_0 = const()[name = string("current_value_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_13_strides_0 = const()[name = string("current_value_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_13_pad_0 = const()[name = string("current_value_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_13_dilations_0 = const()[name = string("current_value_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_13_groups_0 = const()[name = string("current_value_13_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184526272))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185312768))))[name = string("layers_6_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185321024)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = current_value_13_dilations_0, groups = current_value_13_groups_0, pad = current_value_13_pad_0, pad_type = current_value_13_pad_type_0, strides = current_value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16_palettized, x = obj_85_cast_fp16)[name = string("current_value_13_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1522_cast_fp16 = mul(x = var_87_cast_fp16_6, y = var_207_cast_fp16)[name = string("op_1522_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1523_cast_fp16 = mul(x = current_key_13_cast_fp16, y = var_205_cast_fp16)[name = string("op_1523_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_25_cast_fp16 = add(x = var_1522_cast_fp16, y = var_1523_cast_fp16)[name = string("key_25_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1526_cast_fp16 = mul(x = var_114_cast_fp16_6, y = var_207_cast_fp16)[name = string("op_1526_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1527_cast_fp16 = mul(x = current_value_13_cast_fp16, y = var_205_cast_fp16)[name = string("op_1527_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_25_cast_fp16 = add(x = var_1526_cast_fp16, y = var_1527_cast_fp16)[name = string("value_25_cast_fp16")];
            tensor<int32, [4]> var_1531 = const()[name = string("op_1531"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_25_cast_fp16 = reshape(shape = var_1531, x = query_25_cast_fp16)[name = string("mh_q_25_cast_fp16")];
            fp16 var_1533_to_fp16 = const()[name = string("op_1533_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_1534_cast_fp16 = mul(x = mh_q_25_cast_fp16, y = var_1533_to_fp16)[name = string("op_1534_cast_fp16")];
            tensor<int32, [4]> var_1537 = const()[name = string("op_1537"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_1538_cast_fp16 = reshape(shape = var_1537, x = key_25_cast_fp16)[name = string("op_1538_cast_fp16")];
            bool mh_w_37_transpose_x_0 = const()[name = string("mh_w_37_transpose_x_0"), val = bool(true)];
            bool mh_w_37_transpose_y_0 = const()[name = string("mh_w_37_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_1534_cast_fp16, y = var_1538_cast_fp16)[name = string("mh_w_37_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_39_cast_fp16 = add(x = mh_w_37_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_39_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_1546_cast_fp16 = softmax(axis = var_1458, x = mh_w_39_cast_fp16)[name = string("op_1546_cast_fp16")];
            tensor<int32, [4]> var_1547 = const()[name = string("op_1547"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_1548_cast_fp16 = reshape(shape = var_1547, x = value_25_cast_fp16)[name = string("op_1548_cast_fp16")];
            bool attn_25_transpose_x_0 = const()[name = string("attn_25_transpose_x_0"), val = bool(false)];
            bool attn_25_transpose_y_0 = const()[name = string("attn_25_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_25_cast_fp16 = matmul(transpose_x = attn_25_transpose_x_0, transpose_y = attn_25_transpose_y_0, x = var_1548_cast_fp16, y = var_1546_cast_fp16)[name = string("attn_25_cast_fp16")];
            tensor<int32, [4]> var_1551 = const()[name = string("op_1551"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_61_cast_fp16 = reshape(shape = var_1551, x = attn_25_cast_fp16)[name = string("input_61_cast_fp16")];
            string obj_91_pad_type_0 = const()[name = string("obj_91_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_91_strides_0 = const()[name = string("obj_91_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_91_pad_0 = const()[name = string("obj_91_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_91_dilations_0 = const()[name = string("obj_91_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_91_groups_0 = const()[name = string("obj_91_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185323136))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186109632))))[name = string("layers_6_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186117888)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_91_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_91_dilations_0, groups = obj_91_groups_0, pad = obj_91_pad_0, pad_type = obj_91_pad_type_0, strides = obj_91_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16_palettized, x = input_61_cast_fp16)[name = string("obj_91_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_91_cast_fp16)[name = string("inputs_39_cast_fp16")];
            tensor<int32, [1]> out_39_axes_0 = const()[name = string("out_39_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1573_to_fp16 = const()[name = string("op_1573_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1573_to_fp16, x = inputs_39_cast_fp16)[name = string("out_39_cast_fp16")];
            tensor<fp16, [1024]> obj_93_gamma_0_to_fp16 = const()[name = string("obj_93_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186120000)))];
            tensor<fp16, [1024]> obj_93_beta_0_to_fp16 = const()[name = string("obj_93_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186122112)))];
            fp16 obj_93_epsilon_0_to_fp16 = const()[name = string("obj_93_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_93_cast_fp16 = batch_norm(beta = obj_93_beta_0_to_fp16, epsilon = obj_93_epsilon_0_to_fp16, gamma = obj_93_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_39_cast_fp16)[name = string("obj_93_cast_fp16")];
            string query_27_pad_type_0 = const()[name = string("query_27_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_27_strides_0 = const()[name = string("query_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_27_pad_0 = const()[name = string("query_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_27_dilations_0 = const()[name = string("query_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_27_groups_0 = const()[name = string("query_27_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186124224))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186910720))))[name = string("layers_6_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_6_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_6_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186918976)))];
            tensor<fp16, [1, 1024, 1, 1]> query_27_cast_fp16 = conv(bias = layers_6_encoder_attn_q_proj_bias_to_fp16, dilations = query_27_dilations_0, groups = query_27_groups_0, pad = query_27_pad_0, pad_type = query_27_pad_type_0, strides = query_27_strides_0, weight = layers_6_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_93_cast_fp16)[name = string("query_27_cast_fp16")];
            string key_27_pad_type_0 = const()[name = string("key_27_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_27_strides_0 = const()[name = string("key_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_27_pad_0 = const()[name = string("key_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_27_dilations_0 = const()[name = string("key_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_27_groups_0 = const()[name = string("key_27_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186921088))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187707584))))[name = string("layers_6_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_27_cast_fp16 = conv(dilations = key_27_dilations_0, groups = key_27_groups_0, pad = key_27_pad_0, pad_type = key_27_pad_type_0, strides = key_27_strides_0, weight = layers_6_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_27_cast_fp16")];
            string value_27_pad_type_0 = const()[name = string("value_27_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_27_strides_0 = const()[name = string("value_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_27_pad_0 = const()[name = string("value_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_27_dilations_0 = const()[name = string("value_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_27_groups_0 = const()[name = string("value_27_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187715840))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188502336))))[name = string("layers_6_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_6_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_6_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188510592)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_27_cast_fp16 = conv(bias = layers_6_encoder_attn_v_proj_bias_to_fp16, dilations = value_27_dilations_0, groups = value_27_groups_0, pad = value_27_pad_0, pad_type = value_27_pad_type_0, strides = value_27_strides_0, weight = layers_6_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_27_cast_fp16")];
            tensor<int32, [4]> var_1609 = const()[name = string("op_1609"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_27_cast_fp16 = reshape(shape = var_1609, x = query_27_cast_fp16)[name = string("mh_q_27_cast_fp16")];
            fp16 var_1611_to_fp16 = const()[name = string("op_1611_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_1612_cast_fp16 = mul(x = mh_q_27_cast_fp16, y = var_1611_to_fp16)[name = string("op_1612_cast_fp16")];
            tensor<int32, [4]> var_1615 = const()[name = string("op_1615"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1616_cast_fp16 = reshape(shape = var_1615, x = key_27_cast_fp16)[name = string("op_1616_cast_fp16")];
            bool mh_w_41_transpose_x_0 = const()[name = string("mh_w_41_transpose_x_0"), val = bool(true)];
            bool mh_w_41_transpose_y_0 = const()[name = string("mh_w_41_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_1612_cast_fp16, y = var_1616_cast_fp16)[name = string("mh_w_41_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_97_cast_fp16 = softmax(axis = var_1458, x = mh_w_41_cast_fp16)[name = string("obj_97_cast_fp16")];
            tensor<int32, [4]> var_1620 = const()[name = string("op_1620"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1621_cast_fp16 = reshape(shape = var_1620, x = value_27_cast_fp16)[name = string("op_1621_cast_fp16")];
            bool attn_27_transpose_x_0 = const()[name = string("attn_27_transpose_x_0"), val = bool(false)];
            bool attn_27_transpose_y_0 = const()[name = string("attn_27_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_27_cast_fp16 = matmul(transpose_x = attn_27_transpose_x_0, transpose_y = attn_27_transpose_y_0, x = var_1621_cast_fp16, y = obj_97_cast_fp16)[name = string("attn_27_cast_fp16")];
            tensor<int32, [4]> var_1624 = const()[name = string("op_1624"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_63_cast_fp16 = reshape(shape = var_1624, x = attn_27_cast_fp16)[name = string("input_63_cast_fp16")];
            string obj_95_pad_type_0 = const()[name = string("obj_95_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_95_strides_0 = const()[name = string("obj_95_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_95_pad_0 = const()[name = string("obj_95_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_95_dilations_0 = const()[name = string("obj_95_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_95_groups_0 = const()[name = string("obj_95_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188512704))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189299200))))[name = string("layers_6_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_6_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_6_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189307456)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_95_cast_fp16 = conv(bias = layers_6_encoder_attn_o_proj_bias_to_fp16, dilations = obj_95_dilations_0, groups = obj_95_groups_0, pad = obj_95_pad_0, pad_type = obj_95_pad_type_0, strides = obj_95_strides_0, weight = layers_6_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_63_cast_fp16)[name = string("obj_95_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = obj_95_cast_fp16)[name = string("inputs_41_cast_fp16")];
            tensor<int32, [1]> out_41_axes_0 = const()[name = string("out_41_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1642_to_fp16 = const()[name = string("op_1642_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1642_to_fp16, x = inputs_41_cast_fp16)[name = string("out_41_cast_fp16")];
            tensor<fp16, [1024]> input_65_gamma_0_to_fp16 = const()[name = string("input_65_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189309568)))];
            tensor<fp16, [1024]> input_65_beta_0_to_fp16 = const()[name = string("input_65_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189311680)))];
            fp16 input_65_epsilon_0_to_fp16 = const()[name = string("input_65_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_65_cast_fp16 = batch_norm(beta = input_65_beta_0_to_fp16, epsilon = input_65_epsilon_0_to_fp16, gamma = input_65_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_41_cast_fp16)[name = string("input_65_cast_fp16")];
            string input_67_pad_type_0 = const()[name = string("input_67_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_67_strides_0 = const()[name = string("input_67_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_67_pad_0 = const()[name = string("input_67_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_67_dilations_0 = const()[name = string("input_67_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_67_groups_0 = const()[name = string("input_67_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_6_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189313792))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192459584))))[name = string("layers_6_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_6_fc1_bias_to_fp16 = const()[name = string("layers_6_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192492416)))];
            tensor<fp16, [1, 4096, 1, 1]> input_67_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_67_dilations_0, groups = input_67_groups_0, pad = input_67_pad_0, pad_type = input_67_pad_type_0, strides = input_67_strides_0, weight = layers_6_fc1_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = string("input_67_cast_fp16")];
            string input_69_mode_0 = const()[name = string("input_69_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_69_cast_fp16 = gelu(mode = input_69_mode_0, x = input_67_cast_fp16)[name = string("input_69_cast_fp16")];
            string hidden_states_15_pad_type_0 = const()[name = string("hidden_states_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_15_strides_0 = const()[name = string("hidden_states_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_15_pad_0 = const()[name = string("hidden_states_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_15_dilations_0 = const()[name = string("hidden_states_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_15_groups_0 = const()[name = string("hidden_states_15_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_6_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192500672))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195646464))))[name = string("layers_6_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_6_fc2_bias_to_fp16 = const()[name = string("layers_6_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195654720)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_15_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = layers_6_fc2_weight_to_fp16_palettized, x = input_69_cast_fp16)[name = string("hidden_states_15_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = hidden_states_15_cast_fp16)[name = string("inputs_43_cast_fp16")];
            int32 var_1677 = const()[name = string("op_1677"), val = int32(3)];
            tensor<int32, [1]> out_43_axes_0 = const()[name = string("out_43_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1702_to_fp16 = const()[name = string("op_1702_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_1702_to_fp16, x = inputs_43_cast_fp16)[name = string("out_43_cast_fp16")];
            tensor<fp16, [1024]> obj_99_gamma_0_to_fp16 = const()[name = string("obj_99_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195656832)))];
            tensor<fp16, [1024]> obj_99_beta_0_to_fp16 = const()[name = string("obj_99_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195658944)))];
            fp16 obj_99_epsilon_0_to_fp16 = const()[name = string("obj_99_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_99_cast_fp16 = batch_norm(beta = obj_99_beta_0_to_fp16, epsilon = obj_99_epsilon_0_to_fp16, gamma = obj_99_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_43_cast_fp16)[name = string("obj_99_cast_fp16")];
            string query_29_pad_type_0 = const()[name = string("query_29_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_29_strides_0 = const()[name = string("query_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_29_pad_0 = const()[name = string("query_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_29_dilations_0 = const()[name = string("query_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_29_groups_0 = const()[name = string("query_29_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195661056))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196447552))))[name = string("layers_7_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196455808)))];
            tensor<fp16, [1, 1024, 1, 1]> query_29_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_29_dilations_0, groups = query_29_groups_0, pad = query_29_pad_0, pad_type = query_29_pad_type_0, strides = query_29_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16_palettized, x = obj_99_cast_fp16)[name = string("query_29_cast_fp16")];
            string current_key_15_pad_type_0 = const()[name = string("current_key_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_15_strides_0 = const()[name = string("current_key_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_15_pad_0 = const()[name = string("current_key_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_15_dilations_0 = const()[name = string("current_key_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_15_groups_0 = const()[name = string("current_key_15_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196457920))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197244416))))[name = string("layers_7_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_15_cast_fp16 = conv(dilations = current_key_15_dilations_0, groups = current_key_15_groups_0, pad = current_key_15_pad_0, pad_type = current_key_15_pad_type_0, strides = current_key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16_palettized, x = obj_99_cast_fp16)[name = string("current_key_15_cast_fp16")];
            string current_value_15_pad_type_0 = const()[name = string("current_value_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_15_strides_0 = const()[name = string("current_value_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_15_pad_0 = const()[name = string("current_value_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_15_dilations_0 = const()[name = string("current_value_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_15_groups_0 = const()[name = string("current_value_15_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197252672))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198039168))))[name = string("layers_7_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198047424)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = current_value_15_dilations_0, groups = current_value_15_groups_0, pad = current_value_15_pad_0, pad_type = current_value_15_pad_type_0, strides = current_value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16_palettized, x = obj_99_cast_fp16)[name = string("current_value_15_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1741_cast_fp16 = mul(x = var_87_cast_fp16_7, y = var_207_cast_fp16)[name = string("op_1741_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1742_cast_fp16 = mul(x = current_key_15_cast_fp16, y = var_205_cast_fp16)[name = string("op_1742_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_29_cast_fp16 = add(x = var_1741_cast_fp16, y = var_1742_cast_fp16)[name = string("key_29_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1745_cast_fp16 = mul(x = var_114_cast_fp16_7, y = var_207_cast_fp16)[name = string("op_1745_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1746_cast_fp16 = mul(x = current_value_15_cast_fp16, y = var_205_cast_fp16)[name = string("op_1746_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_29_cast_fp16 = add(x = var_1745_cast_fp16, y = var_1746_cast_fp16)[name = string("value_29_cast_fp16")];
            tensor<int32, [4]> var_1750 = const()[name = string("op_1750"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_29_cast_fp16 = reshape(shape = var_1750, x = query_29_cast_fp16)[name = string("mh_q_29_cast_fp16")];
            fp16 var_1752_to_fp16 = const()[name = string("op_1752_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_1753_cast_fp16 = mul(x = mh_q_29_cast_fp16, y = var_1752_to_fp16)[name = string("op_1753_cast_fp16")];
            tensor<int32, [4]> var_1756 = const()[name = string("op_1756"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_1757_cast_fp16 = reshape(shape = var_1756, x = key_29_cast_fp16)[name = string("op_1757_cast_fp16")];
            bool mh_w_43_transpose_x_0 = const()[name = string("mh_w_43_transpose_x_0"), val = bool(true)];
            bool mh_w_43_transpose_y_0 = const()[name = string("mh_w_43_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_43_cast_fp16 = matmul(transpose_x = mh_w_43_transpose_x_0, transpose_y = mh_w_43_transpose_y_0, x = var_1753_cast_fp16, y = var_1757_cast_fp16)[name = string("mh_w_43_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_45_cast_fp16 = add(x = mh_w_43_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_45_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_1765_cast_fp16 = softmax(axis = var_1677, x = mh_w_45_cast_fp16)[name = string("op_1765_cast_fp16")];
            tensor<int32, [4]> var_1766 = const()[name = string("op_1766"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_1767_cast_fp16 = reshape(shape = var_1766, x = value_29_cast_fp16)[name = string("op_1767_cast_fp16")];
            bool attn_29_transpose_x_0 = const()[name = string("attn_29_transpose_x_0"), val = bool(false)];
            bool attn_29_transpose_y_0 = const()[name = string("attn_29_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_29_cast_fp16 = matmul(transpose_x = attn_29_transpose_x_0, transpose_y = attn_29_transpose_y_0, x = var_1767_cast_fp16, y = var_1765_cast_fp16)[name = string("attn_29_cast_fp16")];
            tensor<int32, [4]> var_1770 = const()[name = string("op_1770"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_71_cast_fp16 = reshape(shape = var_1770, x = attn_29_cast_fp16)[name = string("input_71_cast_fp16")];
            string obj_105_pad_type_0 = const()[name = string("obj_105_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_105_strides_0 = const()[name = string("obj_105_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_105_pad_0 = const()[name = string("obj_105_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_105_dilations_0 = const()[name = string("obj_105_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_105_groups_0 = const()[name = string("obj_105_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198049536))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198836032))))[name = string("layers_7_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198844288)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_105_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_105_dilations_0, groups = obj_105_groups_0, pad = obj_105_pad_0, pad_type = obj_105_pad_type_0, strides = obj_105_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16_palettized, x = input_71_cast_fp16)[name = string("obj_105_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = obj_105_cast_fp16)[name = string("inputs_45_cast_fp16")];
            tensor<int32, [1]> out_45_axes_0 = const()[name = string("out_45_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1792_to_fp16 = const()[name = string("op_1792_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_1792_to_fp16, x = inputs_45_cast_fp16)[name = string("out_45_cast_fp16")];
            tensor<fp16, [1024]> obj_107_gamma_0_to_fp16 = const()[name = string("obj_107_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198846400)))];
            tensor<fp16, [1024]> obj_107_beta_0_to_fp16 = const()[name = string("obj_107_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198848512)))];
            fp16 obj_107_epsilon_0_to_fp16 = const()[name = string("obj_107_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_107_cast_fp16 = batch_norm(beta = obj_107_beta_0_to_fp16, epsilon = obj_107_epsilon_0_to_fp16, gamma = obj_107_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_45_cast_fp16)[name = string("obj_107_cast_fp16")];
            string query_31_pad_type_0 = const()[name = string("query_31_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_31_strides_0 = const()[name = string("query_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_31_pad_0 = const()[name = string("query_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_31_dilations_0 = const()[name = string("query_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_31_groups_0 = const()[name = string("query_31_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198850624))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199637120))))[name = string("layers_7_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_7_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_7_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199645376)))];
            tensor<fp16, [1, 1024, 1, 1]> query_31_cast_fp16 = conv(bias = layers_7_encoder_attn_q_proj_bias_to_fp16, dilations = query_31_dilations_0, groups = query_31_groups_0, pad = query_31_pad_0, pad_type = query_31_pad_type_0, strides = query_31_strides_0, weight = layers_7_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_107_cast_fp16)[name = string("query_31_cast_fp16")];
            string key_31_pad_type_0 = const()[name = string("key_31_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_31_strides_0 = const()[name = string("key_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_31_pad_0 = const()[name = string("key_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_31_dilations_0 = const()[name = string("key_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_31_groups_0 = const()[name = string("key_31_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199647488))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200433984))))[name = string("layers_7_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_31_cast_fp16 = conv(dilations = key_31_dilations_0, groups = key_31_groups_0, pad = key_31_pad_0, pad_type = key_31_pad_type_0, strides = key_31_strides_0, weight = layers_7_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_31_cast_fp16")];
            string value_31_pad_type_0 = const()[name = string("value_31_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_31_strides_0 = const()[name = string("value_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_31_pad_0 = const()[name = string("value_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_31_dilations_0 = const()[name = string("value_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_31_groups_0 = const()[name = string("value_31_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200442240))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201228736))))[name = string("layers_7_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_7_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_7_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201236992)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_31_cast_fp16 = conv(bias = layers_7_encoder_attn_v_proj_bias_to_fp16, dilations = value_31_dilations_0, groups = value_31_groups_0, pad = value_31_pad_0, pad_type = value_31_pad_type_0, strides = value_31_strides_0, weight = layers_7_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_31_cast_fp16")];
            tensor<int32, [4]> var_1828 = const()[name = string("op_1828"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_31_cast_fp16 = reshape(shape = var_1828, x = query_31_cast_fp16)[name = string("mh_q_31_cast_fp16")];
            fp16 var_1830_to_fp16 = const()[name = string("op_1830_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_1831_cast_fp16 = mul(x = mh_q_31_cast_fp16, y = var_1830_to_fp16)[name = string("op_1831_cast_fp16")];
            tensor<int32, [4]> var_1834 = const()[name = string("op_1834"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1835_cast_fp16 = reshape(shape = var_1834, x = key_31_cast_fp16)[name = string("op_1835_cast_fp16")];
            bool mh_w_47_transpose_x_0 = const()[name = string("mh_w_47_transpose_x_0"), val = bool(true)];
            bool mh_w_47_transpose_y_0 = const()[name = string("mh_w_47_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_47_cast_fp16 = matmul(transpose_x = mh_w_47_transpose_x_0, transpose_y = mh_w_47_transpose_y_0, x = var_1831_cast_fp16, y = var_1835_cast_fp16)[name = string("mh_w_47_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_111_cast_fp16 = softmax(axis = var_1677, x = mh_w_47_cast_fp16)[name = string("obj_111_cast_fp16")];
            tensor<int32, [4]> var_1839 = const()[name = string("op_1839"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_1840_cast_fp16 = reshape(shape = var_1839, x = value_31_cast_fp16)[name = string("op_1840_cast_fp16")];
            bool attn_31_transpose_x_0 = const()[name = string("attn_31_transpose_x_0"), val = bool(false)];
            bool attn_31_transpose_y_0 = const()[name = string("attn_31_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_31_cast_fp16 = matmul(transpose_x = attn_31_transpose_x_0, transpose_y = attn_31_transpose_y_0, x = var_1840_cast_fp16, y = obj_111_cast_fp16)[name = string("attn_31_cast_fp16")];
            tensor<int32, [4]> var_1843 = const()[name = string("op_1843"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_73_cast_fp16 = reshape(shape = var_1843, x = attn_31_cast_fp16)[name = string("input_73_cast_fp16")];
            string obj_109_pad_type_0 = const()[name = string("obj_109_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_109_strides_0 = const()[name = string("obj_109_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_109_pad_0 = const()[name = string("obj_109_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_109_dilations_0 = const()[name = string("obj_109_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_109_groups_0 = const()[name = string("obj_109_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201239104))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202025600))))[name = string("layers_7_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_7_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_7_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202033856)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_109_cast_fp16 = conv(bias = layers_7_encoder_attn_o_proj_bias_to_fp16, dilations = obj_109_dilations_0, groups = obj_109_groups_0, pad = obj_109_pad_0, pad_type = obj_109_pad_type_0, strides = obj_109_strides_0, weight = layers_7_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = string("obj_109_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_109_cast_fp16)[name = string("inputs_47_cast_fp16")];
            tensor<int32, [1]> out_47_axes_0 = const()[name = string("out_47_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1861_to_fp16 = const()[name = string("op_1861_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_1861_to_fp16, x = inputs_47_cast_fp16)[name = string("out_47_cast_fp16")];
            tensor<fp16, [1024]> input_75_gamma_0_to_fp16 = const()[name = string("input_75_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202035968)))];
            tensor<fp16, [1024]> input_75_beta_0_to_fp16 = const()[name = string("input_75_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202038080)))];
            fp16 input_75_epsilon_0_to_fp16 = const()[name = string("input_75_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_47_cast_fp16)[name = string("input_75_cast_fp16")];
            string input_77_pad_type_0 = const()[name = string("input_77_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_77_strides_0 = const()[name = string("input_77_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_77_pad_0 = const()[name = string("input_77_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_77_dilations_0 = const()[name = string("input_77_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_77_groups_0 = const()[name = string("input_77_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_7_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202040192))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205185984))))[name = string("layers_7_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_7_fc1_bias_to_fp16 = const()[name = string("layers_7_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205218816)))];
            tensor<fp16, [1, 4096, 1, 1]> input_77_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_7_fc1_weight_to_fp16_palettized, x = input_75_cast_fp16)[name = string("input_77_cast_fp16")];
            string input_79_mode_0 = const()[name = string("input_79_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = string("input_79_cast_fp16")];
            string hidden_states_17_pad_type_0 = const()[name = string("hidden_states_17_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_17_strides_0 = const()[name = string("hidden_states_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_17_pad_0 = const()[name = string("hidden_states_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_17_dilations_0 = const()[name = string("hidden_states_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_17_groups_0 = const()[name = string("hidden_states_17_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_7_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205227072))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208372864))))[name = string("layers_7_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_7_fc2_bias_to_fp16 = const()[name = string("layers_7_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208381120)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_17_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = layers_7_fc2_weight_to_fp16_palettized, x = input_79_cast_fp16)[name = string("hidden_states_17_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_17_cast_fp16)[name = string("inputs_49_cast_fp16")];
            int32 var_1896 = const()[name = string("op_1896"), val = int32(3)];
            tensor<int32, [1]> out_49_axes_0 = const()[name = string("out_49_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1921_to_fp16 = const()[name = string("op_1921_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_1921_to_fp16, x = inputs_49_cast_fp16)[name = string("out_49_cast_fp16")];
            tensor<fp16, [1024]> obj_113_gamma_0_to_fp16 = const()[name = string("obj_113_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208383232)))];
            tensor<fp16, [1024]> obj_113_beta_0_to_fp16 = const()[name = string("obj_113_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208385344)))];
            fp16 obj_113_epsilon_0_to_fp16 = const()[name = string("obj_113_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_113_cast_fp16 = batch_norm(beta = obj_113_beta_0_to_fp16, epsilon = obj_113_epsilon_0_to_fp16, gamma = obj_113_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_49_cast_fp16)[name = string("obj_113_cast_fp16")];
            string query_33_pad_type_0 = const()[name = string("query_33_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_33_strides_0 = const()[name = string("query_33_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_33_pad_0 = const()[name = string("query_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_33_dilations_0 = const()[name = string("query_33_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_33_groups_0 = const()[name = string("query_33_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208387456))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209173952))))[name = string("layers_8_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209182208)))];
            tensor<fp16, [1, 1024, 1, 1]> query_33_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_33_dilations_0, groups = query_33_groups_0, pad = query_33_pad_0, pad_type = query_33_pad_type_0, strides = query_33_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16_palettized, x = obj_113_cast_fp16)[name = string("query_33_cast_fp16")];
            string current_key_17_pad_type_0 = const()[name = string("current_key_17_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_17_strides_0 = const()[name = string("current_key_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_17_pad_0 = const()[name = string("current_key_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_17_dilations_0 = const()[name = string("current_key_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_17_groups_0 = const()[name = string("current_key_17_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209184320))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209970816))))[name = string("layers_8_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_17_cast_fp16 = conv(dilations = current_key_17_dilations_0, groups = current_key_17_groups_0, pad = current_key_17_pad_0, pad_type = current_key_17_pad_type_0, strides = current_key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16_palettized, x = obj_113_cast_fp16)[name = string("current_key_17_cast_fp16")];
            string current_value_17_pad_type_0 = const()[name = string("current_value_17_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_17_strides_0 = const()[name = string("current_value_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_17_pad_0 = const()[name = string("current_value_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_17_dilations_0 = const()[name = string("current_value_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_17_groups_0 = const()[name = string("current_value_17_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209979072))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210765568))))[name = string("layers_8_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210773824)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = current_value_17_dilations_0, groups = current_value_17_groups_0, pad = current_value_17_pad_0, pad_type = current_value_17_pad_type_0, strides = current_value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16_palettized, x = obj_113_cast_fp16)[name = string("current_value_17_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1960_cast_fp16 = mul(x = var_87_cast_fp16_8, y = var_207_cast_fp16)[name = string("op_1960_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1961_cast_fp16 = mul(x = current_key_17_cast_fp16, y = var_205_cast_fp16)[name = string("op_1961_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_33_cast_fp16 = add(x = var_1960_cast_fp16, y = var_1961_cast_fp16)[name = string("key_33_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1964_cast_fp16 = mul(x = var_114_cast_fp16_8, y = var_207_cast_fp16)[name = string("op_1964_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_1965_cast_fp16 = mul(x = current_value_17_cast_fp16, y = var_205_cast_fp16)[name = string("op_1965_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_33_cast_fp16 = add(x = var_1964_cast_fp16, y = var_1965_cast_fp16)[name = string("value_33_cast_fp16")];
            tensor<int32, [4]> var_1969 = const()[name = string("op_1969"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_33_cast_fp16 = reshape(shape = var_1969, x = query_33_cast_fp16)[name = string("mh_q_33_cast_fp16")];
            fp16 var_1971_to_fp16 = const()[name = string("op_1971_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_1972_cast_fp16 = mul(x = mh_q_33_cast_fp16, y = var_1971_to_fp16)[name = string("op_1972_cast_fp16")];
            tensor<int32, [4]> var_1975 = const()[name = string("op_1975"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_1976_cast_fp16 = reshape(shape = var_1975, x = key_33_cast_fp16)[name = string("op_1976_cast_fp16")];
            bool mh_w_49_transpose_x_0 = const()[name = string("mh_w_49_transpose_x_0"), val = bool(true)];
            bool mh_w_49_transpose_y_0 = const()[name = string("mh_w_49_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_49_cast_fp16 = matmul(transpose_x = mh_w_49_transpose_x_0, transpose_y = mh_w_49_transpose_y_0, x = var_1972_cast_fp16, y = var_1976_cast_fp16)[name = string("mh_w_49_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_51_cast_fp16 = add(x = mh_w_49_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_51_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_1984_cast_fp16 = softmax(axis = var_1896, x = mh_w_51_cast_fp16)[name = string("op_1984_cast_fp16")];
            tensor<int32, [4]> var_1985 = const()[name = string("op_1985"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_1986_cast_fp16 = reshape(shape = var_1985, x = value_33_cast_fp16)[name = string("op_1986_cast_fp16")];
            bool attn_33_transpose_x_0 = const()[name = string("attn_33_transpose_x_0"), val = bool(false)];
            bool attn_33_transpose_y_0 = const()[name = string("attn_33_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_33_cast_fp16 = matmul(transpose_x = attn_33_transpose_x_0, transpose_y = attn_33_transpose_y_0, x = var_1986_cast_fp16, y = var_1984_cast_fp16)[name = string("attn_33_cast_fp16")];
            tensor<int32, [4]> var_1989 = const()[name = string("op_1989"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_81_cast_fp16 = reshape(shape = var_1989, x = attn_33_cast_fp16)[name = string("input_81_cast_fp16")];
            string obj_119_pad_type_0 = const()[name = string("obj_119_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_119_strides_0 = const()[name = string("obj_119_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_119_pad_0 = const()[name = string("obj_119_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_119_dilations_0 = const()[name = string("obj_119_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_119_groups_0 = const()[name = string("obj_119_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210775936))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211562432))))[name = string("layers_8_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211570688)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_119_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_119_dilations_0, groups = obj_119_groups_0, pad = obj_119_pad_0, pad_type = obj_119_pad_type_0, strides = obj_119_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16_palettized, x = input_81_cast_fp16)[name = string("obj_119_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_51_cast_fp16 = add(x = inputs_49_cast_fp16, y = obj_119_cast_fp16)[name = string("inputs_51_cast_fp16")];
            tensor<int32, [1]> out_51_axes_0 = const()[name = string("out_51_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2011_to_fp16 = const()[name = string("op_2011_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_2011_to_fp16, x = inputs_51_cast_fp16)[name = string("out_51_cast_fp16")];
            tensor<fp16, [1024]> obj_121_gamma_0_to_fp16 = const()[name = string("obj_121_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211572800)))];
            tensor<fp16, [1024]> obj_121_beta_0_to_fp16 = const()[name = string("obj_121_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211574912)))];
            fp16 obj_121_epsilon_0_to_fp16 = const()[name = string("obj_121_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_121_cast_fp16 = batch_norm(beta = obj_121_beta_0_to_fp16, epsilon = obj_121_epsilon_0_to_fp16, gamma = obj_121_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_51_cast_fp16)[name = string("obj_121_cast_fp16")];
            string query_35_pad_type_0 = const()[name = string("query_35_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_35_strides_0 = const()[name = string("query_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_35_pad_0 = const()[name = string("query_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_35_dilations_0 = const()[name = string("query_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_35_groups_0 = const()[name = string("query_35_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211577024))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212363520))))[name = string("layers_8_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_8_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_8_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212371776)))];
            tensor<fp16, [1, 1024, 1, 1]> query_35_cast_fp16 = conv(bias = layers_8_encoder_attn_q_proj_bias_to_fp16, dilations = query_35_dilations_0, groups = query_35_groups_0, pad = query_35_pad_0, pad_type = query_35_pad_type_0, strides = query_35_strides_0, weight = layers_8_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_121_cast_fp16)[name = string("query_35_cast_fp16")];
            string key_35_pad_type_0 = const()[name = string("key_35_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_35_strides_0 = const()[name = string("key_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_35_pad_0 = const()[name = string("key_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_35_dilations_0 = const()[name = string("key_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_35_groups_0 = const()[name = string("key_35_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212373888))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213160384))))[name = string("layers_8_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_35_cast_fp16 = conv(dilations = key_35_dilations_0, groups = key_35_groups_0, pad = key_35_pad_0, pad_type = key_35_pad_type_0, strides = key_35_strides_0, weight = layers_8_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_35_cast_fp16")];
            string value_35_pad_type_0 = const()[name = string("value_35_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_35_strides_0 = const()[name = string("value_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_35_pad_0 = const()[name = string("value_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_35_dilations_0 = const()[name = string("value_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_35_groups_0 = const()[name = string("value_35_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213168640))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213955136))))[name = string("layers_8_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_8_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_8_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213963392)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_35_cast_fp16 = conv(bias = layers_8_encoder_attn_v_proj_bias_to_fp16, dilations = value_35_dilations_0, groups = value_35_groups_0, pad = value_35_pad_0, pad_type = value_35_pad_type_0, strides = value_35_strides_0, weight = layers_8_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_35_cast_fp16")];
            tensor<int32, [4]> var_2047 = const()[name = string("op_2047"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_35_cast_fp16 = reshape(shape = var_2047, x = query_35_cast_fp16)[name = string("mh_q_35_cast_fp16")];
            fp16 var_2049_to_fp16 = const()[name = string("op_2049_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_2050_cast_fp16 = mul(x = mh_q_35_cast_fp16, y = var_2049_to_fp16)[name = string("op_2050_cast_fp16")];
            tensor<int32, [4]> var_2053 = const()[name = string("op_2053"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2054_cast_fp16 = reshape(shape = var_2053, x = key_35_cast_fp16)[name = string("op_2054_cast_fp16")];
            bool mh_w_53_transpose_x_0 = const()[name = string("mh_w_53_transpose_x_0"), val = bool(true)];
            bool mh_w_53_transpose_y_0 = const()[name = string("mh_w_53_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_53_cast_fp16 = matmul(transpose_x = mh_w_53_transpose_x_0, transpose_y = mh_w_53_transpose_y_0, x = var_2050_cast_fp16, y = var_2054_cast_fp16)[name = string("mh_w_53_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_125_cast_fp16 = softmax(axis = var_1896, x = mh_w_53_cast_fp16)[name = string("obj_125_cast_fp16")];
            tensor<int32, [4]> var_2058 = const()[name = string("op_2058"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2059_cast_fp16 = reshape(shape = var_2058, x = value_35_cast_fp16)[name = string("op_2059_cast_fp16")];
            bool attn_35_transpose_x_0 = const()[name = string("attn_35_transpose_x_0"), val = bool(false)];
            bool attn_35_transpose_y_0 = const()[name = string("attn_35_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_35_cast_fp16 = matmul(transpose_x = attn_35_transpose_x_0, transpose_y = attn_35_transpose_y_0, x = var_2059_cast_fp16, y = obj_125_cast_fp16)[name = string("attn_35_cast_fp16")];
            tensor<int32, [4]> var_2062 = const()[name = string("op_2062"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_83_cast_fp16 = reshape(shape = var_2062, x = attn_35_cast_fp16)[name = string("input_83_cast_fp16")];
            string obj_123_pad_type_0 = const()[name = string("obj_123_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_123_strides_0 = const()[name = string("obj_123_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_123_pad_0 = const()[name = string("obj_123_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_123_dilations_0 = const()[name = string("obj_123_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_123_groups_0 = const()[name = string("obj_123_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213965504))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214752000))))[name = string("layers_8_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_8_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_8_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214760256)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_123_cast_fp16 = conv(bias = layers_8_encoder_attn_o_proj_bias_to_fp16, dilations = obj_123_dilations_0, groups = obj_123_groups_0, pad = obj_123_pad_0, pad_type = obj_123_pad_type_0, strides = obj_123_strides_0, weight = layers_8_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_83_cast_fp16)[name = string("obj_123_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = obj_123_cast_fp16)[name = string("inputs_53_cast_fp16")];
            tensor<int32, [1]> out_53_axes_0 = const()[name = string("out_53_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2080_to_fp16 = const()[name = string("op_2080_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_2080_to_fp16, x = inputs_53_cast_fp16)[name = string("out_53_cast_fp16")];
            tensor<fp16, [1024]> input_85_gamma_0_to_fp16 = const()[name = string("input_85_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214762368)))];
            tensor<fp16, [1024]> input_85_beta_0_to_fp16 = const()[name = string("input_85_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214764480)))];
            fp16 input_85_epsilon_0_to_fp16 = const()[name = string("input_85_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_85_cast_fp16 = batch_norm(beta = input_85_beta_0_to_fp16, epsilon = input_85_epsilon_0_to_fp16, gamma = input_85_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_53_cast_fp16)[name = string("input_85_cast_fp16")];
            string input_87_pad_type_0 = const()[name = string("input_87_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_87_strides_0 = const()[name = string("input_87_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_87_pad_0 = const()[name = string("input_87_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_87_dilations_0 = const()[name = string("input_87_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_87_groups_0 = const()[name = string("input_87_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_8_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214766592))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217912384))))[name = string("layers_8_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_8_fc1_bias_to_fp16 = const()[name = string("layers_8_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217945216)))];
            tensor<fp16, [1, 4096, 1, 1]> input_87_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_87_dilations_0, groups = input_87_groups_0, pad = input_87_pad_0, pad_type = input_87_pad_type_0, strides = input_87_strides_0, weight = layers_8_fc1_weight_to_fp16_palettized, x = input_85_cast_fp16)[name = string("input_87_cast_fp16")];
            string input_89_mode_0 = const()[name = string("input_89_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_89_cast_fp16 = gelu(mode = input_89_mode_0, x = input_87_cast_fp16)[name = string("input_89_cast_fp16")];
            string hidden_states_19_pad_type_0 = const()[name = string("hidden_states_19_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_19_strides_0 = const()[name = string("hidden_states_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_19_pad_0 = const()[name = string("hidden_states_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_19_dilations_0 = const()[name = string("hidden_states_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_19_groups_0 = const()[name = string("hidden_states_19_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_8_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217953472))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221099264))))[name = string("layers_8_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_8_fc2_bias_to_fp16 = const()[name = string("layers_8_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221107520)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_19_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = layers_8_fc2_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = string("hidden_states_19_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = hidden_states_19_cast_fp16)[name = string("inputs_55_cast_fp16")];
            int32 var_2115 = const()[name = string("op_2115"), val = int32(3)];
            tensor<int32, [1]> out_55_axes_0 = const()[name = string("out_55_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2140_to_fp16 = const()[name = string("op_2140_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_2140_to_fp16, x = inputs_55_cast_fp16)[name = string("out_55_cast_fp16")];
            tensor<fp16, [1024]> obj_127_gamma_0_to_fp16 = const()[name = string("obj_127_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221109632)))];
            tensor<fp16, [1024]> obj_127_beta_0_to_fp16 = const()[name = string("obj_127_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221111744)))];
            fp16 obj_127_epsilon_0_to_fp16 = const()[name = string("obj_127_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_127_cast_fp16 = batch_norm(beta = obj_127_beta_0_to_fp16, epsilon = obj_127_epsilon_0_to_fp16, gamma = obj_127_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_55_cast_fp16)[name = string("obj_127_cast_fp16")];
            string query_37_pad_type_0 = const()[name = string("query_37_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_37_strides_0 = const()[name = string("query_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_37_pad_0 = const()[name = string("query_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_37_dilations_0 = const()[name = string("query_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_37_groups_0 = const()[name = string("query_37_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221113856))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221900352))))[name = string("layers_9_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221908608)))];
            tensor<fp16, [1, 1024, 1, 1]> query_37_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_37_dilations_0, groups = query_37_groups_0, pad = query_37_pad_0, pad_type = query_37_pad_type_0, strides = query_37_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16_palettized, x = obj_127_cast_fp16)[name = string("query_37_cast_fp16")];
            string current_key_19_pad_type_0 = const()[name = string("current_key_19_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_19_strides_0 = const()[name = string("current_key_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_19_pad_0 = const()[name = string("current_key_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_19_dilations_0 = const()[name = string("current_key_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_19_groups_0 = const()[name = string("current_key_19_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221910720))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222697216))))[name = string("layers_9_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_19_cast_fp16 = conv(dilations = current_key_19_dilations_0, groups = current_key_19_groups_0, pad = current_key_19_pad_0, pad_type = current_key_19_pad_type_0, strides = current_key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16_palettized, x = obj_127_cast_fp16)[name = string("current_key_19_cast_fp16")];
            string current_value_19_pad_type_0 = const()[name = string("current_value_19_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_19_strides_0 = const()[name = string("current_value_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_19_pad_0 = const()[name = string("current_value_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_19_dilations_0 = const()[name = string("current_value_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_19_groups_0 = const()[name = string("current_value_19_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222705472))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223491968))))[name = string("layers_9_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223500224)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = current_value_19_dilations_0, groups = current_value_19_groups_0, pad = current_value_19_pad_0, pad_type = current_value_19_pad_type_0, strides = current_value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16_palettized, x = obj_127_cast_fp16)[name = string("current_value_19_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2179_cast_fp16 = mul(x = var_87_cast_fp16_9, y = var_207_cast_fp16)[name = string("op_2179_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2180_cast_fp16 = mul(x = current_key_19_cast_fp16, y = var_205_cast_fp16)[name = string("op_2180_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_37_cast_fp16 = add(x = var_2179_cast_fp16, y = var_2180_cast_fp16)[name = string("key_37_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2183_cast_fp16 = mul(x = var_114_cast_fp16_9, y = var_207_cast_fp16)[name = string("op_2183_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2184_cast_fp16 = mul(x = current_value_19_cast_fp16, y = var_205_cast_fp16)[name = string("op_2184_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_37_cast_fp16 = add(x = var_2183_cast_fp16, y = var_2184_cast_fp16)[name = string("value_37_cast_fp16")];
            tensor<int32, [4]> var_2188 = const()[name = string("op_2188"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_37_cast_fp16 = reshape(shape = var_2188, x = query_37_cast_fp16)[name = string("mh_q_37_cast_fp16")];
            fp16 var_2190_to_fp16 = const()[name = string("op_2190_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_2191_cast_fp16 = mul(x = mh_q_37_cast_fp16, y = var_2190_to_fp16)[name = string("op_2191_cast_fp16")];
            tensor<int32, [4]> var_2194 = const()[name = string("op_2194"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_2195_cast_fp16 = reshape(shape = var_2194, x = key_37_cast_fp16)[name = string("op_2195_cast_fp16")];
            bool mh_w_55_transpose_x_0 = const()[name = string("mh_w_55_transpose_x_0"), val = bool(true)];
            bool mh_w_55_transpose_y_0 = const()[name = string("mh_w_55_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_55_cast_fp16 = matmul(transpose_x = mh_w_55_transpose_x_0, transpose_y = mh_w_55_transpose_y_0, x = var_2191_cast_fp16, y = var_2195_cast_fp16)[name = string("mh_w_55_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_57_cast_fp16 = add(x = mh_w_55_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_57_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_2203_cast_fp16 = softmax(axis = var_2115, x = mh_w_57_cast_fp16)[name = string("op_2203_cast_fp16")];
            tensor<int32, [4]> var_2204 = const()[name = string("op_2204"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_2205_cast_fp16 = reshape(shape = var_2204, x = value_37_cast_fp16)[name = string("op_2205_cast_fp16")];
            bool attn_37_transpose_x_0 = const()[name = string("attn_37_transpose_x_0"), val = bool(false)];
            bool attn_37_transpose_y_0 = const()[name = string("attn_37_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_37_cast_fp16 = matmul(transpose_x = attn_37_transpose_x_0, transpose_y = attn_37_transpose_y_0, x = var_2205_cast_fp16, y = var_2203_cast_fp16)[name = string("attn_37_cast_fp16")];
            tensor<int32, [4]> var_2208 = const()[name = string("op_2208"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_91_cast_fp16 = reshape(shape = var_2208, x = attn_37_cast_fp16)[name = string("input_91_cast_fp16")];
            string obj_133_pad_type_0 = const()[name = string("obj_133_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_133_strides_0 = const()[name = string("obj_133_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_133_pad_0 = const()[name = string("obj_133_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_133_dilations_0 = const()[name = string("obj_133_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_133_groups_0 = const()[name = string("obj_133_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223502336))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224288832))))[name = string("layers_9_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224297088)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_133_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_133_dilations_0, groups = obj_133_groups_0, pad = obj_133_pad_0, pad_type = obj_133_pad_type_0, strides = obj_133_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16_palettized, x = input_91_cast_fp16)[name = string("obj_133_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = obj_133_cast_fp16)[name = string("inputs_57_cast_fp16")];
            tensor<int32, [1]> out_57_axes_0 = const()[name = string("out_57_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2230_to_fp16 = const()[name = string("op_2230_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_2230_to_fp16, x = inputs_57_cast_fp16)[name = string("out_57_cast_fp16")];
            tensor<fp16, [1024]> obj_135_gamma_0_to_fp16 = const()[name = string("obj_135_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224299200)))];
            tensor<fp16, [1024]> obj_135_beta_0_to_fp16 = const()[name = string("obj_135_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224301312)))];
            fp16 obj_135_epsilon_0_to_fp16 = const()[name = string("obj_135_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_135_cast_fp16 = batch_norm(beta = obj_135_beta_0_to_fp16, epsilon = obj_135_epsilon_0_to_fp16, gamma = obj_135_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_57_cast_fp16)[name = string("obj_135_cast_fp16")];
            string query_39_pad_type_0 = const()[name = string("query_39_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_39_strides_0 = const()[name = string("query_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_39_pad_0 = const()[name = string("query_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_39_dilations_0 = const()[name = string("query_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_39_groups_0 = const()[name = string("query_39_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224303424))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225089920))))[name = string("layers_9_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_9_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_9_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225098176)))];
            tensor<fp16, [1, 1024, 1, 1]> query_39_cast_fp16 = conv(bias = layers_9_encoder_attn_q_proj_bias_to_fp16, dilations = query_39_dilations_0, groups = query_39_groups_0, pad = query_39_pad_0, pad_type = query_39_pad_type_0, strides = query_39_strides_0, weight = layers_9_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_135_cast_fp16)[name = string("query_39_cast_fp16")];
            string key_39_pad_type_0 = const()[name = string("key_39_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_39_strides_0 = const()[name = string("key_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_39_pad_0 = const()[name = string("key_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_39_dilations_0 = const()[name = string("key_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_39_groups_0 = const()[name = string("key_39_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225100288))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225886784))))[name = string("layers_9_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_39_cast_fp16 = conv(dilations = key_39_dilations_0, groups = key_39_groups_0, pad = key_39_pad_0, pad_type = key_39_pad_type_0, strides = key_39_strides_0, weight = layers_9_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_39_cast_fp16")];
            string value_39_pad_type_0 = const()[name = string("value_39_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_39_strides_0 = const()[name = string("value_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_39_pad_0 = const()[name = string("value_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_39_dilations_0 = const()[name = string("value_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_39_groups_0 = const()[name = string("value_39_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225895040))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226681536))))[name = string("layers_9_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_9_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_9_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226689792)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_39_cast_fp16 = conv(bias = layers_9_encoder_attn_v_proj_bias_to_fp16, dilations = value_39_dilations_0, groups = value_39_groups_0, pad = value_39_pad_0, pad_type = value_39_pad_type_0, strides = value_39_strides_0, weight = layers_9_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_39_cast_fp16")];
            tensor<int32, [4]> var_2266 = const()[name = string("op_2266"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_39_cast_fp16 = reshape(shape = var_2266, x = query_39_cast_fp16)[name = string("mh_q_39_cast_fp16")];
            fp16 var_2268_to_fp16 = const()[name = string("op_2268_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_2269_cast_fp16 = mul(x = mh_q_39_cast_fp16, y = var_2268_to_fp16)[name = string("op_2269_cast_fp16")];
            tensor<int32, [4]> var_2272 = const()[name = string("op_2272"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2273_cast_fp16 = reshape(shape = var_2272, x = key_39_cast_fp16)[name = string("op_2273_cast_fp16")];
            bool mh_w_59_transpose_x_0 = const()[name = string("mh_w_59_transpose_x_0"), val = bool(true)];
            bool mh_w_59_transpose_y_0 = const()[name = string("mh_w_59_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_59_cast_fp16 = matmul(transpose_x = mh_w_59_transpose_x_0, transpose_y = mh_w_59_transpose_y_0, x = var_2269_cast_fp16, y = var_2273_cast_fp16)[name = string("mh_w_59_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_139_cast_fp16 = softmax(axis = var_2115, x = mh_w_59_cast_fp16)[name = string("obj_139_cast_fp16")];
            tensor<int32, [4]> var_2277 = const()[name = string("op_2277"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2278_cast_fp16 = reshape(shape = var_2277, x = value_39_cast_fp16)[name = string("op_2278_cast_fp16")];
            bool attn_39_transpose_x_0 = const()[name = string("attn_39_transpose_x_0"), val = bool(false)];
            bool attn_39_transpose_y_0 = const()[name = string("attn_39_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_39_cast_fp16 = matmul(transpose_x = attn_39_transpose_x_0, transpose_y = attn_39_transpose_y_0, x = var_2278_cast_fp16, y = obj_139_cast_fp16)[name = string("attn_39_cast_fp16")];
            tensor<int32, [4]> var_2281 = const()[name = string("op_2281"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_93_cast_fp16 = reshape(shape = var_2281, x = attn_39_cast_fp16)[name = string("input_93_cast_fp16")];
            string obj_137_pad_type_0 = const()[name = string("obj_137_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_137_strides_0 = const()[name = string("obj_137_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_137_pad_0 = const()[name = string("obj_137_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_137_dilations_0 = const()[name = string("obj_137_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_137_groups_0 = const()[name = string("obj_137_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226691904))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227478400))))[name = string("layers_9_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_9_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_9_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227486656)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_137_cast_fp16 = conv(bias = layers_9_encoder_attn_o_proj_bias_to_fp16, dilations = obj_137_dilations_0, groups = obj_137_groups_0, pad = obj_137_pad_0, pad_type = obj_137_pad_type_0, strides = obj_137_strides_0, weight = layers_9_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_93_cast_fp16)[name = string("obj_137_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = obj_137_cast_fp16)[name = string("inputs_59_cast_fp16")];
            tensor<int32, [1]> out_59_axes_0 = const()[name = string("out_59_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2299_to_fp16 = const()[name = string("op_2299_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_2299_to_fp16, x = inputs_59_cast_fp16)[name = string("out_59_cast_fp16")];
            tensor<fp16, [1024]> input_95_gamma_0_to_fp16 = const()[name = string("input_95_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227488768)))];
            tensor<fp16, [1024]> input_95_beta_0_to_fp16 = const()[name = string("input_95_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227490880)))];
            fp16 input_95_epsilon_0_to_fp16 = const()[name = string("input_95_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_95_cast_fp16 = batch_norm(beta = input_95_beta_0_to_fp16, epsilon = input_95_epsilon_0_to_fp16, gamma = input_95_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_59_cast_fp16)[name = string("input_95_cast_fp16")];
            string input_97_pad_type_0 = const()[name = string("input_97_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_97_strides_0 = const()[name = string("input_97_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_97_pad_0 = const()[name = string("input_97_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_97_dilations_0 = const()[name = string("input_97_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_97_groups_0 = const()[name = string("input_97_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_9_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227492992))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230638784))))[name = string("layers_9_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_9_fc1_bias_to_fp16 = const()[name = string("layers_9_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230671616)))];
            tensor<fp16, [1, 4096, 1, 1]> input_97_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_97_dilations_0, groups = input_97_groups_0, pad = input_97_pad_0, pad_type = input_97_pad_type_0, strides = input_97_strides_0, weight = layers_9_fc1_weight_to_fp16_palettized, x = input_95_cast_fp16)[name = string("input_97_cast_fp16")];
            string input_99_mode_0 = const()[name = string("input_99_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_99_cast_fp16 = gelu(mode = input_99_mode_0, x = input_97_cast_fp16)[name = string("input_99_cast_fp16")];
            string hidden_states_21_pad_type_0 = const()[name = string("hidden_states_21_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_21_strides_0 = const()[name = string("hidden_states_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_21_pad_0 = const()[name = string("hidden_states_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_21_dilations_0 = const()[name = string("hidden_states_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_21_groups_0 = const()[name = string("hidden_states_21_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_9_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230679872))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233825664))))[name = string("layers_9_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_9_fc2_bias_to_fp16 = const()[name = string("layers_9_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233833920)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_21_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = layers_9_fc2_weight_to_fp16_palettized, x = input_99_cast_fp16)[name = string("hidden_states_21_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_61_cast_fp16 = add(x = inputs_59_cast_fp16, y = hidden_states_21_cast_fp16)[name = string("inputs_61_cast_fp16")];
            int32 var_2334 = const()[name = string("op_2334"), val = int32(3)];
            tensor<int32, [1]> out_61_axes_0 = const()[name = string("out_61_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2359_to_fp16 = const()[name = string("op_2359_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_2359_to_fp16, x = inputs_61_cast_fp16)[name = string("out_61_cast_fp16")];
            tensor<fp16, [1024]> obj_141_gamma_0_to_fp16 = const()[name = string("obj_141_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233836032)))];
            tensor<fp16, [1024]> obj_141_beta_0_to_fp16 = const()[name = string("obj_141_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233838144)))];
            fp16 obj_141_epsilon_0_to_fp16 = const()[name = string("obj_141_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_141_cast_fp16 = batch_norm(beta = obj_141_beta_0_to_fp16, epsilon = obj_141_epsilon_0_to_fp16, gamma = obj_141_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_61_cast_fp16)[name = string("obj_141_cast_fp16")];
            string query_41_pad_type_0 = const()[name = string("query_41_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_41_strides_0 = const()[name = string("query_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_41_pad_0 = const()[name = string("query_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_41_dilations_0 = const()[name = string("query_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_41_groups_0 = const()[name = string("query_41_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233840256))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234626752))))[name = string("layers_10_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234635008)))];
            tensor<fp16, [1, 1024, 1, 1]> query_41_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_41_dilations_0, groups = query_41_groups_0, pad = query_41_pad_0, pad_type = query_41_pad_type_0, strides = query_41_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16_palettized, x = obj_141_cast_fp16)[name = string("query_41_cast_fp16")];
            string current_key_21_pad_type_0 = const()[name = string("current_key_21_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_21_strides_0 = const()[name = string("current_key_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_21_pad_0 = const()[name = string("current_key_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_21_dilations_0 = const()[name = string("current_key_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_21_groups_0 = const()[name = string("current_key_21_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234637120))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235423616))))[name = string("layers_10_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_21_cast_fp16 = conv(dilations = current_key_21_dilations_0, groups = current_key_21_groups_0, pad = current_key_21_pad_0, pad_type = current_key_21_pad_type_0, strides = current_key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16_palettized, x = obj_141_cast_fp16)[name = string("current_key_21_cast_fp16")];
            string current_value_21_pad_type_0 = const()[name = string("current_value_21_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_21_strides_0 = const()[name = string("current_value_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_21_pad_0 = const()[name = string("current_value_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_21_dilations_0 = const()[name = string("current_value_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_21_groups_0 = const()[name = string("current_value_21_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235431872))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236218368))))[name = string("layers_10_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236226624)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = current_value_21_dilations_0, groups = current_value_21_groups_0, pad = current_value_21_pad_0, pad_type = current_value_21_pad_type_0, strides = current_value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16_palettized, x = obj_141_cast_fp16)[name = string("current_value_21_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2398_cast_fp16 = mul(x = var_87_cast_fp16_10, y = var_207_cast_fp16)[name = string("op_2398_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2399_cast_fp16 = mul(x = current_key_21_cast_fp16, y = var_205_cast_fp16)[name = string("op_2399_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_41_cast_fp16 = add(x = var_2398_cast_fp16, y = var_2399_cast_fp16)[name = string("key_41_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2402_cast_fp16 = mul(x = var_114_cast_fp16_10, y = var_207_cast_fp16)[name = string("op_2402_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2403_cast_fp16 = mul(x = current_value_21_cast_fp16, y = var_205_cast_fp16)[name = string("op_2403_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_41_cast_fp16 = add(x = var_2402_cast_fp16, y = var_2403_cast_fp16)[name = string("value_41_cast_fp16")];
            tensor<int32, [4]> var_2407 = const()[name = string("op_2407"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_41_cast_fp16 = reshape(shape = var_2407, x = query_41_cast_fp16)[name = string("mh_q_41_cast_fp16")];
            fp16 var_2409_to_fp16 = const()[name = string("op_2409_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_2410_cast_fp16 = mul(x = mh_q_41_cast_fp16, y = var_2409_to_fp16)[name = string("op_2410_cast_fp16")];
            tensor<int32, [4]> var_2413 = const()[name = string("op_2413"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_2414_cast_fp16 = reshape(shape = var_2413, x = key_41_cast_fp16)[name = string("op_2414_cast_fp16")];
            bool mh_w_61_transpose_x_0 = const()[name = string("mh_w_61_transpose_x_0"), val = bool(true)];
            bool mh_w_61_transpose_y_0 = const()[name = string("mh_w_61_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_61_cast_fp16 = matmul(transpose_x = mh_w_61_transpose_x_0, transpose_y = mh_w_61_transpose_y_0, x = var_2410_cast_fp16, y = var_2414_cast_fp16)[name = string("mh_w_61_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_63_cast_fp16 = add(x = mh_w_61_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_63_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_2422_cast_fp16 = softmax(axis = var_2334, x = mh_w_63_cast_fp16)[name = string("op_2422_cast_fp16")];
            tensor<int32, [4]> var_2423 = const()[name = string("op_2423"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_2424_cast_fp16 = reshape(shape = var_2423, x = value_41_cast_fp16)[name = string("op_2424_cast_fp16")];
            bool attn_41_transpose_x_0 = const()[name = string("attn_41_transpose_x_0"), val = bool(false)];
            bool attn_41_transpose_y_0 = const()[name = string("attn_41_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_41_cast_fp16 = matmul(transpose_x = attn_41_transpose_x_0, transpose_y = attn_41_transpose_y_0, x = var_2424_cast_fp16, y = var_2422_cast_fp16)[name = string("attn_41_cast_fp16")];
            tensor<int32, [4]> var_2427 = const()[name = string("op_2427"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_101_cast_fp16 = reshape(shape = var_2427, x = attn_41_cast_fp16)[name = string("input_101_cast_fp16")];
            string obj_147_pad_type_0 = const()[name = string("obj_147_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_147_strides_0 = const()[name = string("obj_147_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_147_pad_0 = const()[name = string("obj_147_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_147_dilations_0 = const()[name = string("obj_147_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_147_groups_0 = const()[name = string("obj_147_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236228736))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237015232))))[name = string("layers_10_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237023488)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_147_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_147_dilations_0, groups = obj_147_groups_0, pad = obj_147_pad_0, pad_type = obj_147_pad_type_0, strides = obj_147_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16_palettized, x = input_101_cast_fp16)[name = string("obj_147_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = obj_147_cast_fp16)[name = string("inputs_63_cast_fp16")];
            tensor<int32, [1]> out_63_axes_0 = const()[name = string("out_63_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2449_to_fp16 = const()[name = string("op_2449_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_2449_to_fp16, x = inputs_63_cast_fp16)[name = string("out_63_cast_fp16")];
            tensor<fp16, [1024]> obj_149_gamma_0_to_fp16 = const()[name = string("obj_149_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237025600)))];
            tensor<fp16, [1024]> obj_149_beta_0_to_fp16 = const()[name = string("obj_149_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237027712)))];
            fp16 obj_149_epsilon_0_to_fp16 = const()[name = string("obj_149_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_149_cast_fp16 = batch_norm(beta = obj_149_beta_0_to_fp16, epsilon = obj_149_epsilon_0_to_fp16, gamma = obj_149_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_63_cast_fp16)[name = string("obj_149_cast_fp16")];
            string query_43_pad_type_0 = const()[name = string("query_43_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_43_strides_0 = const()[name = string("query_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_43_pad_0 = const()[name = string("query_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_43_dilations_0 = const()[name = string("query_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_43_groups_0 = const()[name = string("query_43_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237029824))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237816320))))[name = string("layers_10_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_10_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_10_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237824576)))];
            tensor<fp16, [1, 1024, 1, 1]> query_43_cast_fp16 = conv(bias = layers_10_encoder_attn_q_proj_bias_to_fp16, dilations = query_43_dilations_0, groups = query_43_groups_0, pad = query_43_pad_0, pad_type = query_43_pad_type_0, strides = query_43_strides_0, weight = layers_10_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_149_cast_fp16)[name = string("query_43_cast_fp16")];
            string key_43_pad_type_0 = const()[name = string("key_43_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_43_strides_0 = const()[name = string("key_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_43_pad_0 = const()[name = string("key_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_43_dilations_0 = const()[name = string("key_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_43_groups_0 = const()[name = string("key_43_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237826688))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238613184))))[name = string("layers_10_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_43_cast_fp16 = conv(dilations = key_43_dilations_0, groups = key_43_groups_0, pad = key_43_pad_0, pad_type = key_43_pad_type_0, strides = key_43_strides_0, weight = layers_10_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_43_cast_fp16")];
            string value_43_pad_type_0 = const()[name = string("value_43_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_43_strides_0 = const()[name = string("value_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_43_pad_0 = const()[name = string("value_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_43_dilations_0 = const()[name = string("value_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_43_groups_0 = const()[name = string("value_43_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238621440))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239407936))))[name = string("layers_10_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_10_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_10_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239416192)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_43_cast_fp16 = conv(bias = layers_10_encoder_attn_v_proj_bias_to_fp16, dilations = value_43_dilations_0, groups = value_43_groups_0, pad = value_43_pad_0, pad_type = value_43_pad_type_0, strides = value_43_strides_0, weight = layers_10_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_43_cast_fp16")];
            tensor<int32, [4]> var_2485 = const()[name = string("op_2485"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_43_cast_fp16 = reshape(shape = var_2485, x = query_43_cast_fp16)[name = string("mh_q_43_cast_fp16")];
            fp16 var_2487_to_fp16 = const()[name = string("op_2487_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_2488_cast_fp16 = mul(x = mh_q_43_cast_fp16, y = var_2487_to_fp16)[name = string("op_2488_cast_fp16")];
            tensor<int32, [4]> var_2491 = const()[name = string("op_2491"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2492_cast_fp16 = reshape(shape = var_2491, x = key_43_cast_fp16)[name = string("op_2492_cast_fp16")];
            bool mh_w_65_transpose_x_0 = const()[name = string("mh_w_65_transpose_x_0"), val = bool(true)];
            bool mh_w_65_transpose_y_0 = const()[name = string("mh_w_65_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_65_cast_fp16 = matmul(transpose_x = mh_w_65_transpose_x_0, transpose_y = mh_w_65_transpose_y_0, x = var_2488_cast_fp16, y = var_2492_cast_fp16)[name = string("mh_w_65_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_153_cast_fp16 = softmax(axis = var_2334, x = mh_w_65_cast_fp16)[name = string("obj_153_cast_fp16")];
            tensor<int32, [4]> var_2496 = const()[name = string("op_2496"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2497_cast_fp16 = reshape(shape = var_2496, x = value_43_cast_fp16)[name = string("op_2497_cast_fp16")];
            bool attn_43_transpose_x_0 = const()[name = string("attn_43_transpose_x_0"), val = bool(false)];
            bool attn_43_transpose_y_0 = const()[name = string("attn_43_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_43_cast_fp16 = matmul(transpose_x = attn_43_transpose_x_0, transpose_y = attn_43_transpose_y_0, x = var_2497_cast_fp16, y = obj_153_cast_fp16)[name = string("attn_43_cast_fp16")];
            tensor<int32, [4]> var_2500 = const()[name = string("op_2500"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_103_cast_fp16 = reshape(shape = var_2500, x = attn_43_cast_fp16)[name = string("input_103_cast_fp16")];
            string obj_151_pad_type_0 = const()[name = string("obj_151_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_151_strides_0 = const()[name = string("obj_151_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_151_pad_0 = const()[name = string("obj_151_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_151_dilations_0 = const()[name = string("obj_151_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_151_groups_0 = const()[name = string("obj_151_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239418304))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240204800))))[name = string("layers_10_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_10_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_10_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240213056)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_151_cast_fp16 = conv(bias = layers_10_encoder_attn_o_proj_bias_to_fp16, dilations = obj_151_dilations_0, groups = obj_151_groups_0, pad = obj_151_pad_0, pad_type = obj_151_pad_type_0, strides = obj_151_strides_0, weight = layers_10_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_103_cast_fp16)[name = string("obj_151_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = obj_151_cast_fp16)[name = string("inputs_65_cast_fp16")];
            tensor<int32, [1]> out_65_axes_0 = const()[name = string("out_65_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2518_to_fp16 = const()[name = string("op_2518_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_2518_to_fp16, x = inputs_65_cast_fp16)[name = string("out_65_cast_fp16")];
            tensor<fp16, [1024]> input_105_gamma_0_to_fp16 = const()[name = string("input_105_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240215168)))];
            tensor<fp16, [1024]> input_105_beta_0_to_fp16 = const()[name = string("input_105_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240217280)))];
            fp16 input_105_epsilon_0_to_fp16 = const()[name = string("input_105_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_105_cast_fp16 = batch_norm(beta = input_105_beta_0_to_fp16, epsilon = input_105_epsilon_0_to_fp16, gamma = input_105_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_65_cast_fp16)[name = string("input_105_cast_fp16")];
            string input_107_pad_type_0 = const()[name = string("input_107_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_107_strides_0 = const()[name = string("input_107_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_107_pad_0 = const()[name = string("input_107_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_107_dilations_0 = const()[name = string("input_107_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_107_groups_0 = const()[name = string("input_107_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_10_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240219392))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243365184))))[name = string("layers_10_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_10_fc1_bias_to_fp16 = const()[name = string("layers_10_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243398016)))];
            tensor<fp16, [1, 4096, 1, 1]> input_107_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_107_dilations_0, groups = input_107_groups_0, pad = input_107_pad_0, pad_type = input_107_pad_type_0, strides = input_107_strides_0, weight = layers_10_fc1_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = string("input_107_cast_fp16")];
            string input_109_mode_0 = const()[name = string("input_109_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_109_cast_fp16 = gelu(mode = input_109_mode_0, x = input_107_cast_fp16)[name = string("input_109_cast_fp16")];
            string hidden_states_23_pad_type_0 = const()[name = string("hidden_states_23_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_23_strides_0 = const()[name = string("hidden_states_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_23_pad_0 = const()[name = string("hidden_states_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_23_dilations_0 = const()[name = string("hidden_states_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_23_groups_0 = const()[name = string("hidden_states_23_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_10_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243406272))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246552064))))[name = string("layers_10_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_10_fc2_bias_to_fp16 = const()[name = string("layers_10_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246560320)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_23_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = layers_10_fc2_weight_to_fp16_palettized, x = input_109_cast_fp16)[name = string("hidden_states_23_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = hidden_states_23_cast_fp16)[name = string("inputs_67_cast_fp16")];
            int32 var_2553 = const()[name = string("op_2553"), val = int32(3)];
            tensor<int32, [1]> out_67_axes_0 = const()[name = string("out_67_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2578_to_fp16 = const()[name = string("op_2578_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_2578_to_fp16, x = inputs_67_cast_fp16)[name = string("out_67_cast_fp16")];
            tensor<fp16, [1024]> obj_155_gamma_0_to_fp16 = const()[name = string("obj_155_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246562432)))];
            tensor<fp16, [1024]> obj_155_beta_0_to_fp16 = const()[name = string("obj_155_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246564544)))];
            fp16 obj_155_epsilon_0_to_fp16 = const()[name = string("obj_155_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_155_cast_fp16 = batch_norm(beta = obj_155_beta_0_to_fp16, epsilon = obj_155_epsilon_0_to_fp16, gamma = obj_155_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_67_cast_fp16)[name = string("obj_155_cast_fp16")];
            string query_45_pad_type_0 = const()[name = string("query_45_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_45_strides_0 = const()[name = string("query_45_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_45_pad_0 = const()[name = string("query_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_45_dilations_0 = const()[name = string("query_45_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_45_groups_0 = const()[name = string("query_45_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246566656))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247353152))))[name = string("layers_11_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247361408)))];
            tensor<fp16, [1, 1024, 1, 1]> query_45_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_45_dilations_0, groups = query_45_groups_0, pad = query_45_pad_0, pad_type = query_45_pad_type_0, strides = query_45_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16_palettized, x = obj_155_cast_fp16)[name = string("query_45_cast_fp16")];
            string current_key_23_pad_type_0 = const()[name = string("current_key_23_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_23_strides_0 = const()[name = string("current_key_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_23_pad_0 = const()[name = string("current_key_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_23_dilations_0 = const()[name = string("current_key_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_23_groups_0 = const()[name = string("current_key_23_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247363520))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248150016))))[name = string("layers_11_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_23_cast_fp16 = conv(dilations = current_key_23_dilations_0, groups = current_key_23_groups_0, pad = current_key_23_pad_0, pad_type = current_key_23_pad_type_0, strides = current_key_23_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16_palettized, x = obj_155_cast_fp16)[name = string("current_key_23_cast_fp16")];
            string current_value_23_pad_type_0 = const()[name = string("current_value_23_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_23_strides_0 = const()[name = string("current_value_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_23_pad_0 = const()[name = string("current_value_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_23_dilations_0 = const()[name = string("current_value_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_23_groups_0 = const()[name = string("current_value_23_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248158272))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248944768))))[name = string("layers_11_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248953024)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_23_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = current_value_23_dilations_0, groups = current_value_23_groups_0, pad = current_value_23_pad_0, pad_type = current_value_23_pad_type_0, strides = current_value_23_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16_palettized, x = obj_155_cast_fp16)[name = string("current_value_23_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2617_cast_fp16 = mul(x = var_87_cast_fp16_11, y = var_207_cast_fp16)[name = string("op_2617_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2618_cast_fp16 = mul(x = current_key_23_cast_fp16, y = var_205_cast_fp16)[name = string("op_2618_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_45_cast_fp16 = add(x = var_2617_cast_fp16, y = var_2618_cast_fp16)[name = string("key_45_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2621_cast_fp16 = mul(x = var_114_cast_fp16_11, y = var_207_cast_fp16)[name = string("op_2621_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2622_cast_fp16 = mul(x = current_value_23_cast_fp16, y = var_205_cast_fp16)[name = string("op_2622_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_45_cast_fp16 = add(x = var_2621_cast_fp16, y = var_2622_cast_fp16)[name = string("value_45_cast_fp16")];
            tensor<int32, [4]> var_2626 = const()[name = string("op_2626"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_45_cast_fp16 = reshape(shape = var_2626, x = query_45_cast_fp16)[name = string("mh_q_45_cast_fp16")];
            fp16 var_2628_to_fp16 = const()[name = string("op_2628_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_2629_cast_fp16 = mul(x = mh_q_45_cast_fp16, y = var_2628_to_fp16)[name = string("op_2629_cast_fp16")];
            tensor<int32, [4]> var_2632 = const()[name = string("op_2632"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_2633_cast_fp16 = reshape(shape = var_2632, x = key_45_cast_fp16)[name = string("op_2633_cast_fp16")];
            bool mh_w_67_transpose_x_0 = const()[name = string("mh_w_67_transpose_x_0"), val = bool(true)];
            bool mh_w_67_transpose_y_0 = const()[name = string("mh_w_67_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_67_cast_fp16 = matmul(transpose_x = mh_w_67_transpose_x_0, transpose_y = mh_w_67_transpose_y_0, x = var_2629_cast_fp16, y = var_2633_cast_fp16)[name = string("mh_w_67_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_69_cast_fp16 = add(x = mh_w_67_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_69_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_2641_cast_fp16 = softmax(axis = var_2553, x = mh_w_69_cast_fp16)[name = string("op_2641_cast_fp16")];
            tensor<int32, [4]> var_2642 = const()[name = string("op_2642"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_2643_cast_fp16 = reshape(shape = var_2642, x = value_45_cast_fp16)[name = string("op_2643_cast_fp16")];
            bool attn_45_transpose_x_0 = const()[name = string("attn_45_transpose_x_0"), val = bool(false)];
            bool attn_45_transpose_y_0 = const()[name = string("attn_45_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_45_cast_fp16 = matmul(transpose_x = attn_45_transpose_x_0, transpose_y = attn_45_transpose_y_0, x = var_2643_cast_fp16, y = var_2641_cast_fp16)[name = string("attn_45_cast_fp16")];
            tensor<int32, [4]> var_2646 = const()[name = string("op_2646"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_111_cast_fp16 = reshape(shape = var_2646, x = attn_45_cast_fp16)[name = string("input_111_cast_fp16")];
            string obj_161_pad_type_0 = const()[name = string("obj_161_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_161_strides_0 = const()[name = string("obj_161_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_161_pad_0 = const()[name = string("obj_161_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_161_dilations_0 = const()[name = string("obj_161_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_161_groups_0 = const()[name = string("obj_161_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248955136))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249741632))))[name = string("layers_11_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249749888)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_161_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_161_dilations_0, groups = obj_161_groups_0, pad = obj_161_pad_0, pad_type = obj_161_pad_type_0, strides = obj_161_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16_palettized, x = input_111_cast_fp16)[name = string("obj_161_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = obj_161_cast_fp16)[name = string("inputs_69_cast_fp16")];
            tensor<int32, [1]> out_69_axes_0 = const()[name = string("out_69_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2668_to_fp16 = const()[name = string("op_2668_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_2668_to_fp16, x = inputs_69_cast_fp16)[name = string("out_69_cast_fp16")];
            tensor<fp16, [1024]> obj_163_gamma_0_to_fp16 = const()[name = string("obj_163_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249752000)))];
            tensor<fp16, [1024]> obj_163_beta_0_to_fp16 = const()[name = string("obj_163_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249754112)))];
            fp16 obj_163_epsilon_0_to_fp16 = const()[name = string("obj_163_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_163_cast_fp16 = batch_norm(beta = obj_163_beta_0_to_fp16, epsilon = obj_163_epsilon_0_to_fp16, gamma = obj_163_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_69_cast_fp16)[name = string("obj_163_cast_fp16")];
            string query_47_pad_type_0 = const()[name = string("query_47_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_47_strides_0 = const()[name = string("query_47_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_47_pad_0 = const()[name = string("query_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_47_dilations_0 = const()[name = string("query_47_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_47_groups_0 = const()[name = string("query_47_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249756224))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250542720))))[name = string("layers_11_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_11_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_11_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250550976)))];
            tensor<fp16, [1, 1024, 1, 1]> query_47_cast_fp16 = conv(bias = layers_11_encoder_attn_q_proj_bias_to_fp16, dilations = query_47_dilations_0, groups = query_47_groups_0, pad = query_47_pad_0, pad_type = query_47_pad_type_0, strides = query_47_strides_0, weight = layers_11_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_163_cast_fp16)[name = string("query_47_cast_fp16")];
            string key_47_pad_type_0 = const()[name = string("key_47_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_47_strides_0 = const()[name = string("key_47_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_47_pad_0 = const()[name = string("key_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_47_dilations_0 = const()[name = string("key_47_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_47_groups_0 = const()[name = string("key_47_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250553088))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251339584))))[name = string("layers_11_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_47_cast_fp16 = conv(dilations = key_47_dilations_0, groups = key_47_groups_0, pad = key_47_pad_0, pad_type = key_47_pad_type_0, strides = key_47_strides_0, weight = layers_11_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_47_cast_fp16")];
            string value_47_pad_type_0 = const()[name = string("value_47_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_47_strides_0 = const()[name = string("value_47_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_47_pad_0 = const()[name = string("value_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_47_dilations_0 = const()[name = string("value_47_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_47_groups_0 = const()[name = string("value_47_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251347840))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252134336))))[name = string("layers_11_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_11_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_11_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252142592)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_47_cast_fp16 = conv(bias = layers_11_encoder_attn_v_proj_bias_to_fp16, dilations = value_47_dilations_0, groups = value_47_groups_0, pad = value_47_pad_0, pad_type = value_47_pad_type_0, strides = value_47_strides_0, weight = layers_11_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_47_cast_fp16")];
            tensor<int32, [4]> var_2704 = const()[name = string("op_2704"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_47_cast_fp16 = reshape(shape = var_2704, x = query_47_cast_fp16)[name = string("mh_q_47_cast_fp16")];
            fp16 var_2706_to_fp16 = const()[name = string("op_2706_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_2707_cast_fp16 = mul(x = mh_q_47_cast_fp16, y = var_2706_to_fp16)[name = string("op_2707_cast_fp16")];
            tensor<int32, [4]> var_2710 = const()[name = string("op_2710"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2711_cast_fp16 = reshape(shape = var_2710, x = key_47_cast_fp16)[name = string("op_2711_cast_fp16")];
            bool mh_w_71_transpose_x_0 = const()[name = string("mh_w_71_transpose_x_0"), val = bool(true)];
            bool mh_w_71_transpose_y_0 = const()[name = string("mh_w_71_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_71_cast_fp16 = matmul(transpose_x = mh_w_71_transpose_x_0, transpose_y = mh_w_71_transpose_y_0, x = var_2707_cast_fp16, y = var_2711_cast_fp16)[name = string("mh_w_71_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_167_cast_fp16 = softmax(axis = var_2553, x = mh_w_71_cast_fp16)[name = string("obj_167_cast_fp16")];
            tensor<int32, [4]> var_2715 = const()[name = string("op_2715"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2716_cast_fp16 = reshape(shape = var_2715, x = value_47_cast_fp16)[name = string("op_2716_cast_fp16")];
            bool attn_47_transpose_x_0 = const()[name = string("attn_47_transpose_x_0"), val = bool(false)];
            bool attn_47_transpose_y_0 = const()[name = string("attn_47_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_47_cast_fp16 = matmul(transpose_x = attn_47_transpose_x_0, transpose_y = attn_47_transpose_y_0, x = var_2716_cast_fp16, y = obj_167_cast_fp16)[name = string("attn_47_cast_fp16")];
            tensor<int32, [4]> var_2719 = const()[name = string("op_2719"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_113_cast_fp16 = reshape(shape = var_2719, x = attn_47_cast_fp16)[name = string("input_113_cast_fp16")];
            string obj_165_pad_type_0 = const()[name = string("obj_165_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_165_strides_0 = const()[name = string("obj_165_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_165_pad_0 = const()[name = string("obj_165_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_165_dilations_0 = const()[name = string("obj_165_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_165_groups_0 = const()[name = string("obj_165_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252144704))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252931200))))[name = string("layers_11_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_11_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_11_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252939456)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_165_cast_fp16 = conv(bias = layers_11_encoder_attn_o_proj_bias_to_fp16, dilations = obj_165_dilations_0, groups = obj_165_groups_0, pad = obj_165_pad_0, pad_type = obj_165_pad_type_0, strides = obj_165_strides_0, weight = layers_11_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = string("obj_165_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_71_cast_fp16 = add(x = inputs_69_cast_fp16, y = obj_165_cast_fp16)[name = string("inputs_71_cast_fp16")];
            tensor<int32, [1]> out_71_axes_0 = const()[name = string("out_71_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2737_to_fp16 = const()[name = string("op_2737_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_2737_to_fp16, x = inputs_71_cast_fp16)[name = string("out_71_cast_fp16")];
            tensor<fp16, [1024]> input_115_gamma_0_to_fp16 = const()[name = string("input_115_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252941568)))];
            tensor<fp16, [1024]> input_115_beta_0_to_fp16 = const()[name = string("input_115_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252943680)))];
            fp16 input_115_epsilon_0_to_fp16 = const()[name = string("input_115_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_115_cast_fp16 = batch_norm(beta = input_115_beta_0_to_fp16, epsilon = input_115_epsilon_0_to_fp16, gamma = input_115_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_71_cast_fp16)[name = string("input_115_cast_fp16")];
            string input_117_pad_type_0 = const()[name = string("input_117_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_117_strides_0 = const()[name = string("input_117_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_117_pad_0 = const()[name = string("input_117_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_117_dilations_0 = const()[name = string("input_117_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_117_groups_0 = const()[name = string("input_117_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_11_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252945792))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256091584))))[name = string("layers_11_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_11_fc1_bias_to_fp16 = const()[name = string("layers_11_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256124416)))];
            tensor<fp16, [1, 4096, 1, 1]> input_117_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_117_dilations_0, groups = input_117_groups_0, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = input_117_strides_0, weight = layers_11_fc1_weight_to_fp16_palettized, x = input_115_cast_fp16)[name = string("input_117_cast_fp16")];
            string input_119_mode_0 = const()[name = string("input_119_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_119_cast_fp16 = gelu(mode = input_119_mode_0, x = input_117_cast_fp16)[name = string("input_119_cast_fp16")];
            string hidden_states_25_pad_type_0 = const()[name = string("hidden_states_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_25_strides_0 = const()[name = string("hidden_states_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_25_pad_0 = const()[name = string("hidden_states_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_25_dilations_0 = const()[name = string("hidden_states_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_25_groups_0 = const()[name = string("hidden_states_25_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_11_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256132672))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259278464))))[name = string("layers_11_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_11_fc2_bias_to_fp16 = const()[name = string("layers_11_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259286720)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_25_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = layers_11_fc2_weight_to_fp16_palettized, x = input_119_cast_fp16)[name = string("hidden_states_25_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_73_cast_fp16 = add(x = inputs_71_cast_fp16, y = hidden_states_25_cast_fp16)[name = string("inputs_73_cast_fp16")];
            int32 var_2772 = const()[name = string("op_2772"), val = int32(3)];
            tensor<int32, [1]> out_73_axes_0 = const()[name = string("out_73_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2797_to_fp16 = const()[name = string("op_2797_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_73_cast_fp16 = layer_norm(axes = out_73_axes_0, epsilon = var_2797_to_fp16, x = inputs_73_cast_fp16)[name = string("out_73_cast_fp16")];
            tensor<fp16, [1024]> obj_169_gamma_0_to_fp16 = const()[name = string("obj_169_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259288832)))];
            tensor<fp16, [1024]> obj_169_beta_0_to_fp16 = const()[name = string("obj_169_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259290944)))];
            fp16 obj_169_epsilon_0_to_fp16 = const()[name = string("obj_169_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_169_cast_fp16 = batch_norm(beta = obj_169_beta_0_to_fp16, epsilon = obj_169_epsilon_0_to_fp16, gamma = obj_169_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_73_cast_fp16)[name = string("obj_169_cast_fp16")];
            string query_49_pad_type_0 = const()[name = string("query_49_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_49_strides_0 = const()[name = string("query_49_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_49_pad_0 = const()[name = string("query_49_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_49_dilations_0 = const()[name = string("query_49_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_49_groups_0 = const()[name = string("query_49_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259293056))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260079552))))[name = string("layers_12_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260087808)))];
            tensor<fp16, [1, 1024, 1, 1]> query_49_cast_fp16 = conv(bias = layers_12_self_attn_q_proj_bias_to_fp16, dilations = query_49_dilations_0, groups = query_49_groups_0, pad = query_49_pad_0, pad_type = query_49_pad_type_0, strides = query_49_strides_0, weight = layers_12_self_attn_q_proj_weight_to_fp16_palettized, x = obj_169_cast_fp16)[name = string("query_49_cast_fp16")];
            string current_key_25_pad_type_0 = const()[name = string("current_key_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_25_strides_0 = const()[name = string("current_key_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_25_pad_0 = const()[name = string("current_key_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_25_dilations_0 = const()[name = string("current_key_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_25_groups_0 = const()[name = string("current_key_25_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260089920))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260876416))))[name = string("layers_12_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_25_cast_fp16 = conv(dilations = current_key_25_dilations_0, groups = current_key_25_groups_0, pad = current_key_25_pad_0, pad_type = current_key_25_pad_type_0, strides = current_key_25_strides_0, weight = layers_12_self_attn_k_proj_weight_to_fp16_palettized, x = obj_169_cast_fp16)[name = string("current_key_25_cast_fp16")];
            string current_value_25_pad_type_0 = const()[name = string("current_value_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_25_strides_0 = const()[name = string("current_value_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_25_pad_0 = const()[name = string("current_value_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_25_dilations_0 = const()[name = string("current_value_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_25_groups_0 = const()[name = string("current_value_25_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260884672))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261671168))))[name = string("layers_12_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261679424)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_25_cast_fp16 = conv(bias = layers_12_self_attn_v_proj_bias_to_fp16, dilations = current_value_25_dilations_0, groups = current_value_25_groups_0, pad = current_value_25_pad_0, pad_type = current_value_25_pad_type_0, strides = current_value_25_strides_0, weight = layers_12_self_attn_v_proj_weight_to_fp16_palettized, x = obj_169_cast_fp16)[name = string("current_value_25_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2836_cast_fp16 = mul(x = var_87_cast_fp16_12, y = var_207_cast_fp16)[name = string("op_2836_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2837_cast_fp16 = mul(x = current_key_25_cast_fp16, y = var_205_cast_fp16)[name = string("op_2837_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_49_cast_fp16 = add(x = var_2836_cast_fp16, y = var_2837_cast_fp16)[name = string("key_49_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2840_cast_fp16 = mul(x = var_114_cast_fp16_12, y = var_207_cast_fp16)[name = string("op_2840_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_2841_cast_fp16 = mul(x = current_value_25_cast_fp16, y = var_205_cast_fp16)[name = string("op_2841_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_49_cast_fp16 = add(x = var_2840_cast_fp16, y = var_2841_cast_fp16)[name = string("value_49_cast_fp16")];
            tensor<int32, [4]> var_2845 = const()[name = string("op_2845"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_49_cast_fp16 = reshape(shape = var_2845, x = query_49_cast_fp16)[name = string("mh_q_49_cast_fp16")];
            fp16 var_2847_to_fp16 = const()[name = string("op_2847_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_2848_cast_fp16 = mul(x = mh_q_49_cast_fp16, y = var_2847_to_fp16)[name = string("op_2848_cast_fp16")];
            tensor<int32, [4]> var_2851 = const()[name = string("op_2851"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_2852_cast_fp16 = reshape(shape = var_2851, x = key_49_cast_fp16)[name = string("op_2852_cast_fp16")];
            bool mh_w_73_transpose_x_0 = const()[name = string("mh_w_73_transpose_x_0"), val = bool(true)];
            bool mh_w_73_transpose_y_0 = const()[name = string("mh_w_73_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_73_cast_fp16 = matmul(transpose_x = mh_w_73_transpose_x_0, transpose_y = mh_w_73_transpose_y_0, x = var_2848_cast_fp16, y = var_2852_cast_fp16)[name = string("mh_w_73_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_75_cast_fp16 = add(x = mh_w_73_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_75_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_2860_cast_fp16 = softmax(axis = var_2772, x = mh_w_75_cast_fp16)[name = string("op_2860_cast_fp16")];
            tensor<int32, [4]> var_2861 = const()[name = string("op_2861"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_2862_cast_fp16 = reshape(shape = var_2861, x = value_49_cast_fp16)[name = string("op_2862_cast_fp16")];
            bool attn_49_transpose_x_0 = const()[name = string("attn_49_transpose_x_0"), val = bool(false)];
            bool attn_49_transpose_y_0 = const()[name = string("attn_49_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_49_cast_fp16 = matmul(transpose_x = attn_49_transpose_x_0, transpose_y = attn_49_transpose_y_0, x = var_2862_cast_fp16, y = var_2860_cast_fp16)[name = string("attn_49_cast_fp16")];
            tensor<int32, [4]> var_2865 = const()[name = string("op_2865"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_121_cast_fp16 = reshape(shape = var_2865, x = attn_49_cast_fp16)[name = string("input_121_cast_fp16")];
            string obj_175_pad_type_0 = const()[name = string("obj_175_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_175_strides_0 = const()[name = string("obj_175_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_175_pad_0 = const()[name = string("obj_175_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_175_dilations_0 = const()[name = string("obj_175_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_175_groups_0 = const()[name = string("obj_175_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261681536))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262468032))))[name = string("layers_12_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_12_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262476288)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_175_cast_fp16 = conv(bias = layers_12_self_attn_o_proj_bias_to_fp16, dilations = obj_175_dilations_0, groups = obj_175_groups_0, pad = obj_175_pad_0, pad_type = obj_175_pad_type_0, strides = obj_175_strides_0, weight = layers_12_self_attn_o_proj_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = string("obj_175_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_75_cast_fp16 = add(x = inputs_73_cast_fp16, y = obj_175_cast_fp16)[name = string("inputs_75_cast_fp16")];
            tensor<int32, [1]> out_75_axes_0 = const()[name = string("out_75_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2887_to_fp16 = const()[name = string("op_2887_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_75_cast_fp16 = layer_norm(axes = out_75_axes_0, epsilon = var_2887_to_fp16, x = inputs_75_cast_fp16)[name = string("out_75_cast_fp16")];
            tensor<fp16, [1024]> obj_177_gamma_0_to_fp16 = const()[name = string("obj_177_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262478400)))];
            tensor<fp16, [1024]> obj_177_beta_0_to_fp16 = const()[name = string("obj_177_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262480512)))];
            fp16 obj_177_epsilon_0_to_fp16 = const()[name = string("obj_177_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_177_cast_fp16 = batch_norm(beta = obj_177_beta_0_to_fp16, epsilon = obj_177_epsilon_0_to_fp16, gamma = obj_177_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_75_cast_fp16)[name = string("obj_177_cast_fp16")];
            string query_51_pad_type_0 = const()[name = string("query_51_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_51_strides_0 = const()[name = string("query_51_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_51_pad_0 = const()[name = string("query_51_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_51_dilations_0 = const()[name = string("query_51_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_51_groups_0 = const()[name = string("query_51_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262482624))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263269120))))[name = string("layers_12_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_12_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_12_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263277376)))];
            tensor<fp16, [1, 1024, 1, 1]> query_51_cast_fp16 = conv(bias = layers_12_encoder_attn_q_proj_bias_to_fp16, dilations = query_51_dilations_0, groups = query_51_groups_0, pad = query_51_pad_0, pad_type = query_51_pad_type_0, strides = query_51_strides_0, weight = layers_12_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_177_cast_fp16)[name = string("query_51_cast_fp16")];
            string key_51_pad_type_0 = const()[name = string("key_51_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_51_strides_0 = const()[name = string("key_51_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_51_pad_0 = const()[name = string("key_51_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_51_dilations_0 = const()[name = string("key_51_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_51_groups_0 = const()[name = string("key_51_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263279488))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264065984))))[name = string("layers_12_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_51_cast_fp16 = conv(dilations = key_51_dilations_0, groups = key_51_groups_0, pad = key_51_pad_0, pad_type = key_51_pad_type_0, strides = key_51_strides_0, weight = layers_12_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_51_cast_fp16")];
            string value_51_pad_type_0 = const()[name = string("value_51_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_51_strides_0 = const()[name = string("value_51_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_51_pad_0 = const()[name = string("value_51_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_51_dilations_0 = const()[name = string("value_51_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_51_groups_0 = const()[name = string("value_51_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264074240))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264860736))))[name = string("layers_12_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_12_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_12_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264868992)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_51_cast_fp16 = conv(bias = layers_12_encoder_attn_v_proj_bias_to_fp16, dilations = value_51_dilations_0, groups = value_51_groups_0, pad = value_51_pad_0, pad_type = value_51_pad_type_0, strides = value_51_strides_0, weight = layers_12_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_51_cast_fp16")];
            tensor<int32, [4]> var_2923 = const()[name = string("op_2923"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_51_cast_fp16 = reshape(shape = var_2923, x = query_51_cast_fp16)[name = string("mh_q_51_cast_fp16")];
            fp16 var_2925_to_fp16 = const()[name = string("op_2925_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_2926_cast_fp16 = mul(x = mh_q_51_cast_fp16, y = var_2925_to_fp16)[name = string("op_2926_cast_fp16")];
            tensor<int32, [4]> var_2929 = const()[name = string("op_2929"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2930_cast_fp16 = reshape(shape = var_2929, x = key_51_cast_fp16)[name = string("op_2930_cast_fp16")];
            bool mh_w_77_transpose_x_0 = const()[name = string("mh_w_77_transpose_x_0"), val = bool(true)];
            bool mh_w_77_transpose_y_0 = const()[name = string("mh_w_77_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_77_cast_fp16 = matmul(transpose_x = mh_w_77_transpose_x_0, transpose_y = mh_w_77_transpose_y_0, x = var_2926_cast_fp16, y = var_2930_cast_fp16)[name = string("mh_w_77_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_181_cast_fp16 = softmax(axis = var_2772, x = mh_w_77_cast_fp16)[name = string("obj_181_cast_fp16")];
            tensor<int32, [4]> var_2934 = const()[name = string("op_2934"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_2935_cast_fp16 = reshape(shape = var_2934, x = value_51_cast_fp16)[name = string("op_2935_cast_fp16")];
            bool attn_51_transpose_x_0 = const()[name = string("attn_51_transpose_x_0"), val = bool(false)];
            bool attn_51_transpose_y_0 = const()[name = string("attn_51_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_51_cast_fp16 = matmul(transpose_x = attn_51_transpose_x_0, transpose_y = attn_51_transpose_y_0, x = var_2935_cast_fp16, y = obj_181_cast_fp16)[name = string("attn_51_cast_fp16")];
            tensor<int32, [4]> var_2938 = const()[name = string("op_2938"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_123_cast_fp16 = reshape(shape = var_2938, x = attn_51_cast_fp16)[name = string("input_123_cast_fp16")];
            string obj_179_pad_type_0 = const()[name = string("obj_179_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_179_strides_0 = const()[name = string("obj_179_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_179_pad_0 = const()[name = string("obj_179_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_179_dilations_0 = const()[name = string("obj_179_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_179_groups_0 = const()[name = string("obj_179_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264871104))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265657600))))[name = string("layers_12_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_12_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_12_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265665856)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_179_cast_fp16 = conv(bias = layers_12_encoder_attn_o_proj_bias_to_fp16, dilations = obj_179_dilations_0, groups = obj_179_groups_0, pad = obj_179_pad_0, pad_type = obj_179_pad_type_0, strides = obj_179_strides_0, weight = layers_12_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_123_cast_fp16)[name = string("obj_179_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_77_cast_fp16 = add(x = inputs_75_cast_fp16, y = obj_179_cast_fp16)[name = string("inputs_77_cast_fp16")];
            tensor<int32, [1]> out_77_axes_0 = const()[name = string("out_77_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2956_to_fp16 = const()[name = string("op_2956_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_77_cast_fp16 = layer_norm(axes = out_77_axes_0, epsilon = var_2956_to_fp16, x = inputs_77_cast_fp16)[name = string("out_77_cast_fp16")];
            tensor<fp16, [1024]> input_125_gamma_0_to_fp16 = const()[name = string("input_125_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265667968)))];
            tensor<fp16, [1024]> input_125_beta_0_to_fp16 = const()[name = string("input_125_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265670080)))];
            fp16 input_125_epsilon_0_to_fp16 = const()[name = string("input_125_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_125_cast_fp16 = batch_norm(beta = input_125_beta_0_to_fp16, epsilon = input_125_epsilon_0_to_fp16, gamma = input_125_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_77_cast_fp16)[name = string("input_125_cast_fp16")];
            string input_127_pad_type_0 = const()[name = string("input_127_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_127_strides_0 = const()[name = string("input_127_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_127_pad_0 = const()[name = string("input_127_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_127_dilations_0 = const()[name = string("input_127_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_127_groups_0 = const()[name = string("input_127_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_12_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265672192))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268817984))))[name = string("layers_12_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_12_fc1_bias_to_fp16 = const()[name = string("layers_12_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268850816)))];
            tensor<fp16, [1, 4096, 1, 1]> input_127_cast_fp16 = conv(bias = layers_12_fc1_bias_to_fp16, dilations = input_127_dilations_0, groups = input_127_groups_0, pad = input_127_pad_0, pad_type = input_127_pad_type_0, strides = input_127_strides_0, weight = layers_12_fc1_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = string("input_127_cast_fp16")];
            string input_129_mode_0 = const()[name = string("input_129_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_129_cast_fp16 = gelu(mode = input_129_mode_0, x = input_127_cast_fp16)[name = string("input_129_cast_fp16")];
            string hidden_states_27_pad_type_0 = const()[name = string("hidden_states_27_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_27_strides_0 = const()[name = string("hidden_states_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_27_pad_0 = const()[name = string("hidden_states_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_27_dilations_0 = const()[name = string("hidden_states_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_27_groups_0 = const()[name = string("hidden_states_27_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_12_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268859072))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272004864))))[name = string("layers_12_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_12_fc2_bias_to_fp16 = const()[name = string("layers_12_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272013120)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_27_cast_fp16 = conv(bias = layers_12_fc2_bias_to_fp16, dilations = hidden_states_27_dilations_0, groups = hidden_states_27_groups_0, pad = hidden_states_27_pad_0, pad_type = hidden_states_27_pad_type_0, strides = hidden_states_27_strides_0, weight = layers_12_fc2_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = string("hidden_states_27_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_79_cast_fp16 = add(x = inputs_77_cast_fp16, y = hidden_states_27_cast_fp16)[name = string("inputs_79_cast_fp16")];
            int32 var_2991 = const()[name = string("op_2991"), val = int32(3)];
            tensor<int32, [1]> out_79_axes_0 = const()[name = string("out_79_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_3016_to_fp16 = const()[name = string("op_3016_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_79_cast_fp16 = layer_norm(axes = out_79_axes_0, epsilon = var_3016_to_fp16, x = inputs_79_cast_fp16)[name = string("out_79_cast_fp16")];
            tensor<fp16, [1024]> obj_183_gamma_0_to_fp16 = const()[name = string("obj_183_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272015232)))];
            tensor<fp16, [1024]> obj_183_beta_0_to_fp16 = const()[name = string("obj_183_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272017344)))];
            fp16 obj_183_epsilon_0_to_fp16 = const()[name = string("obj_183_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_183_cast_fp16 = batch_norm(beta = obj_183_beta_0_to_fp16, epsilon = obj_183_epsilon_0_to_fp16, gamma = obj_183_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_79_cast_fp16)[name = string("obj_183_cast_fp16")];
            string query_53_pad_type_0 = const()[name = string("query_53_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_53_strides_0 = const()[name = string("query_53_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_53_pad_0 = const()[name = string("query_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_53_dilations_0 = const()[name = string("query_53_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_53_groups_0 = const()[name = string("query_53_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272019456))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272805952))))[name = string("layers_13_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272814208)))];
            tensor<fp16, [1, 1024, 1, 1]> query_53_cast_fp16 = conv(bias = layers_13_self_attn_q_proj_bias_to_fp16, dilations = query_53_dilations_0, groups = query_53_groups_0, pad = query_53_pad_0, pad_type = query_53_pad_type_0, strides = query_53_strides_0, weight = layers_13_self_attn_q_proj_weight_to_fp16_palettized, x = obj_183_cast_fp16)[name = string("query_53_cast_fp16")];
            string current_key_27_pad_type_0 = const()[name = string("current_key_27_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_27_strides_0 = const()[name = string("current_key_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_27_pad_0 = const()[name = string("current_key_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_27_dilations_0 = const()[name = string("current_key_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_27_groups_0 = const()[name = string("current_key_27_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272816320))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273602816))))[name = string("layers_13_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_27_cast_fp16 = conv(dilations = current_key_27_dilations_0, groups = current_key_27_groups_0, pad = current_key_27_pad_0, pad_type = current_key_27_pad_type_0, strides = current_key_27_strides_0, weight = layers_13_self_attn_k_proj_weight_to_fp16_palettized, x = obj_183_cast_fp16)[name = string("current_key_27_cast_fp16")];
            string current_value_27_pad_type_0 = const()[name = string("current_value_27_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_27_strides_0 = const()[name = string("current_value_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_27_pad_0 = const()[name = string("current_value_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_27_dilations_0 = const()[name = string("current_value_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_27_groups_0 = const()[name = string("current_value_27_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273611072))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274397568))))[name = string("layers_13_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274405824)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_27_cast_fp16 = conv(bias = layers_13_self_attn_v_proj_bias_to_fp16, dilations = current_value_27_dilations_0, groups = current_value_27_groups_0, pad = current_value_27_pad_0, pad_type = current_value_27_pad_type_0, strides = current_value_27_strides_0, weight = layers_13_self_attn_v_proj_weight_to_fp16_palettized, x = obj_183_cast_fp16)[name = string("current_value_27_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3055_cast_fp16 = mul(x = var_87_cast_fp16_13, y = var_207_cast_fp16)[name = string("op_3055_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3056_cast_fp16 = mul(x = current_key_27_cast_fp16, y = var_205_cast_fp16)[name = string("op_3056_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_53_cast_fp16 = add(x = var_3055_cast_fp16, y = var_3056_cast_fp16)[name = string("key_53_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3059_cast_fp16 = mul(x = var_114_cast_fp16_13, y = var_207_cast_fp16)[name = string("op_3059_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3060_cast_fp16 = mul(x = current_value_27_cast_fp16, y = var_205_cast_fp16)[name = string("op_3060_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_53_cast_fp16 = add(x = var_3059_cast_fp16, y = var_3060_cast_fp16)[name = string("value_53_cast_fp16")];
            tensor<int32, [4]> var_3064 = const()[name = string("op_3064"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_53_cast_fp16 = reshape(shape = var_3064, x = query_53_cast_fp16)[name = string("mh_q_53_cast_fp16")];
            fp16 var_3066_to_fp16 = const()[name = string("op_3066_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_3067_cast_fp16 = mul(x = mh_q_53_cast_fp16, y = var_3066_to_fp16)[name = string("op_3067_cast_fp16")];
            tensor<int32, [4]> var_3070 = const()[name = string("op_3070"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_3071_cast_fp16 = reshape(shape = var_3070, x = key_53_cast_fp16)[name = string("op_3071_cast_fp16")];
            bool mh_w_79_transpose_x_0 = const()[name = string("mh_w_79_transpose_x_0"), val = bool(true)];
            bool mh_w_79_transpose_y_0 = const()[name = string("mh_w_79_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_79_cast_fp16 = matmul(transpose_x = mh_w_79_transpose_x_0, transpose_y = mh_w_79_transpose_y_0, x = var_3067_cast_fp16, y = var_3071_cast_fp16)[name = string("mh_w_79_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_81_cast_fp16 = add(x = mh_w_79_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_81_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_3079_cast_fp16 = softmax(axis = var_2991, x = mh_w_81_cast_fp16)[name = string("op_3079_cast_fp16")];
            tensor<int32, [4]> var_3080 = const()[name = string("op_3080"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_3081_cast_fp16 = reshape(shape = var_3080, x = value_53_cast_fp16)[name = string("op_3081_cast_fp16")];
            bool attn_53_transpose_x_0 = const()[name = string("attn_53_transpose_x_0"), val = bool(false)];
            bool attn_53_transpose_y_0 = const()[name = string("attn_53_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_53_cast_fp16 = matmul(transpose_x = attn_53_transpose_x_0, transpose_y = attn_53_transpose_y_0, x = var_3081_cast_fp16, y = var_3079_cast_fp16)[name = string("attn_53_cast_fp16")];
            tensor<int32, [4]> var_3084 = const()[name = string("op_3084"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_131_cast_fp16 = reshape(shape = var_3084, x = attn_53_cast_fp16)[name = string("input_131_cast_fp16")];
            string obj_189_pad_type_0 = const()[name = string("obj_189_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_189_strides_0 = const()[name = string("obj_189_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_189_pad_0 = const()[name = string("obj_189_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_189_dilations_0 = const()[name = string("obj_189_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_189_groups_0 = const()[name = string("obj_189_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274407936))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275194432))))[name = string("layers_13_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_13_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275202688)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_189_cast_fp16 = conv(bias = layers_13_self_attn_o_proj_bias_to_fp16, dilations = obj_189_dilations_0, groups = obj_189_groups_0, pad = obj_189_pad_0, pad_type = obj_189_pad_type_0, strides = obj_189_strides_0, weight = layers_13_self_attn_o_proj_weight_to_fp16_palettized, x = input_131_cast_fp16)[name = string("obj_189_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_81_cast_fp16 = add(x = inputs_79_cast_fp16, y = obj_189_cast_fp16)[name = string("inputs_81_cast_fp16")];
            tensor<int32, [1]> out_81_axes_0 = const()[name = string("out_81_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_3106_to_fp16 = const()[name = string("op_3106_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_81_cast_fp16 = layer_norm(axes = out_81_axes_0, epsilon = var_3106_to_fp16, x = inputs_81_cast_fp16)[name = string("out_81_cast_fp16")];
            tensor<fp16, [1024]> obj_191_gamma_0_to_fp16 = const()[name = string("obj_191_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275204800)))];
            tensor<fp16, [1024]> obj_191_beta_0_to_fp16 = const()[name = string("obj_191_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275206912)))];
            fp16 obj_191_epsilon_0_to_fp16 = const()[name = string("obj_191_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_191_cast_fp16 = batch_norm(beta = obj_191_beta_0_to_fp16, epsilon = obj_191_epsilon_0_to_fp16, gamma = obj_191_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_81_cast_fp16)[name = string("obj_191_cast_fp16")];
            string query_55_pad_type_0 = const()[name = string("query_55_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_55_strides_0 = const()[name = string("query_55_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_55_pad_0 = const()[name = string("query_55_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_55_dilations_0 = const()[name = string("query_55_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_55_groups_0 = const()[name = string("query_55_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275209024))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275995520))))[name = string("layers_13_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_13_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_13_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276003776)))];
            tensor<fp16, [1, 1024, 1, 1]> query_55_cast_fp16 = conv(bias = layers_13_encoder_attn_q_proj_bias_to_fp16, dilations = query_55_dilations_0, groups = query_55_groups_0, pad = query_55_pad_0, pad_type = query_55_pad_type_0, strides = query_55_strides_0, weight = layers_13_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_191_cast_fp16)[name = string("query_55_cast_fp16")];
            string key_55_pad_type_0 = const()[name = string("key_55_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_55_strides_0 = const()[name = string("key_55_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_55_pad_0 = const()[name = string("key_55_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_55_dilations_0 = const()[name = string("key_55_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_55_groups_0 = const()[name = string("key_55_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276005888))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276792384))))[name = string("layers_13_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_55_cast_fp16 = conv(dilations = key_55_dilations_0, groups = key_55_groups_0, pad = key_55_pad_0, pad_type = key_55_pad_type_0, strides = key_55_strides_0, weight = layers_13_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_55_cast_fp16")];
            string value_55_pad_type_0 = const()[name = string("value_55_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_55_strides_0 = const()[name = string("value_55_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_55_pad_0 = const()[name = string("value_55_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_55_dilations_0 = const()[name = string("value_55_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_55_groups_0 = const()[name = string("value_55_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276800640))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277587136))))[name = string("layers_13_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_13_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_13_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277595392)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_55_cast_fp16 = conv(bias = layers_13_encoder_attn_v_proj_bias_to_fp16, dilations = value_55_dilations_0, groups = value_55_groups_0, pad = value_55_pad_0, pad_type = value_55_pad_type_0, strides = value_55_strides_0, weight = layers_13_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_55_cast_fp16")];
            tensor<int32, [4]> var_3142 = const()[name = string("op_3142"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_55_cast_fp16 = reshape(shape = var_3142, x = query_55_cast_fp16)[name = string("mh_q_55_cast_fp16")];
            fp16 var_3144_to_fp16 = const()[name = string("op_3144_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_3145_cast_fp16 = mul(x = mh_q_55_cast_fp16, y = var_3144_to_fp16)[name = string("op_3145_cast_fp16")];
            tensor<int32, [4]> var_3148 = const()[name = string("op_3148"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_3149_cast_fp16 = reshape(shape = var_3148, x = key_55_cast_fp16)[name = string("op_3149_cast_fp16")];
            bool mh_w_83_transpose_x_0 = const()[name = string("mh_w_83_transpose_x_0"), val = bool(true)];
            bool mh_w_83_transpose_y_0 = const()[name = string("mh_w_83_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_83_cast_fp16 = matmul(transpose_x = mh_w_83_transpose_x_0, transpose_y = mh_w_83_transpose_y_0, x = var_3145_cast_fp16, y = var_3149_cast_fp16)[name = string("mh_w_83_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_195_cast_fp16 = softmax(axis = var_2991, x = mh_w_83_cast_fp16)[name = string("obj_195_cast_fp16")];
            tensor<int32, [4]> var_3153 = const()[name = string("op_3153"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_3154_cast_fp16 = reshape(shape = var_3153, x = value_55_cast_fp16)[name = string("op_3154_cast_fp16")];
            bool attn_55_transpose_x_0 = const()[name = string("attn_55_transpose_x_0"), val = bool(false)];
            bool attn_55_transpose_y_0 = const()[name = string("attn_55_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_55_cast_fp16 = matmul(transpose_x = attn_55_transpose_x_0, transpose_y = attn_55_transpose_y_0, x = var_3154_cast_fp16, y = obj_195_cast_fp16)[name = string("attn_55_cast_fp16")];
            tensor<int32, [4]> var_3157 = const()[name = string("op_3157"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_133_cast_fp16 = reshape(shape = var_3157, x = attn_55_cast_fp16)[name = string("input_133_cast_fp16")];
            string obj_193_pad_type_0 = const()[name = string("obj_193_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_193_strides_0 = const()[name = string("obj_193_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_193_pad_0 = const()[name = string("obj_193_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_193_dilations_0 = const()[name = string("obj_193_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_193_groups_0 = const()[name = string("obj_193_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277597504))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278384000))))[name = string("layers_13_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_13_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_13_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278392256)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_193_cast_fp16 = conv(bias = layers_13_encoder_attn_o_proj_bias_to_fp16, dilations = obj_193_dilations_0, groups = obj_193_groups_0, pad = obj_193_pad_0, pad_type = obj_193_pad_type_0, strides = obj_193_strides_0, weight = layers_13_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_133_cast_fp16)[name = string("obj_193_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_83_cast_fp16 = add(x = inputs_81_cast_fp16, y = obj_193_cast_fp16)[name = string("inputs_83_cast_fp16")];
            tensor<int32, [1]> out_83_axes_0 = const()[name = string("out_83_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_3178_to_fp16 = const()[name = string("op_3178_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_83_cast_fp16 = layer_norm(axes = out_83_axes_0, epsilon = var_3178_to_fp16, x = inputs_83_cast_fp16)[name = string("out_83_cast_fp16")];
            tensor<fp16, [1024]> input_135_gamma_0_to_fp16 = const()[name = string("input_135_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278394368)))];
            tensor<fp16, [1024]> input_135_beta_0_to_fp16 = const()[name = string("input_135_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278396480)))];
            fp16 input_135_epsilon_0_to_fp16 = const()[name = string("input_135_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_135_cast_fp16 = batch_norm(beta = input_135_beta_0_to_fp16, epsilon = input_135_epsilon_0_to_fp16, gamma = input_135_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_83_cast_fp16)[name = string("input_135_cast_fp16")];
            string input_137_pad_type_0 = const()[name = string("input_137_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_137_strides_0 = const()[name = string("input_137_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_137_pad_0 = const()[name = string("input_137_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_137_dilations_0 = const()[name = string("input_137_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_137_groups_0 = const()[name = string("input_137_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_13_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278398592))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281544384))))[name = string("layers_13_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_13_fc1_bias_to_fp16 = const()[name = string("layers_13_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281577216)))];
            tensor<fp16, [1, 4096, 1, 1]> input_137_cast_fp16 = conv(bias = layers_13_fc1_bias_to_fp16, dilations = input_137_dilations_0, groups = input_137_groups_0, pad = input_137_pad_0, pad_type = input_137_pad_type_0, strides = input_137_strides_0, weight = layers_13_fc1_weight_to_fp16_palettized, x = input_135_cast_fp16)[name = string("input_137_cast_fp16")];
            string input_139_mode_0 = const()[name = string("input_139_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_139_cast_fp16 = gelu(mode = input_139_mode_0, x = input_137_cast_fp16)[name = string("input_139_cast_fp16")];
            string hidden_states_29_pad_type_0 = const()[name = string("hidden_states_29_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_29_strides_0 = const()[name = string("hidden_states_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_29_pad_0 = const()[name = string("hidden_states_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_29_dilations_0 = const()[name = string("hidden_states_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_29_groups_0 = const()[name = string("hidden_states_29_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_13_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281585472))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284731264))))[name = string("layers_13_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_13_fc2_bias_to_fp16 = const()[name = string("layers_13_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284739520)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_29_cast_fp16 = conv(bias = layers_13_fc2_bias_to_fp16, dilations = hidden_states_29_dilations_0, groups = hidden_states_29_groups_0, pad = hidden_states_29_pad_0, pad_type = hidden_states_29_pad_type_0, strides = hidden_states_29_strides_0, weight = layers_13_fc2_weight_to_fp16_palettized, x = input_139_cast_fp16)[name = string("hidden_states_29_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_85_cast_fp16 = add(x = inputs_83_cast_fp16, y = hidden_states_29_cast_fp16)[name = string("inputs_85_cast_fp16")];
            int32 var_3214 = const()[name = string("op_3214"), val = int32(3)];
            tensor<int32, [1]> out_85_axes_0 = const()[name = string("out_85_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_3239_to_fp16 = const()[name = string("op_3239_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_85_cast_fp16 = layer_norm(axes = out_85_axes_0, epsilon = var_3239_to_fp16, x = inputs_85_cast_fp16)[name = string("out_85_cast_fp16")];
            tensor<fp16, [1024]> obj_197_gamma_0_to_fp16 = const()[name = string("obj_197_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284741632)))];
            tensor<fp16, [1024]> obj_197_beta_0_to_fp16 = const()[name = string("obj_197_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284743744)))];
            fp16 obj_197_epsilon_0_to_fp16 = const()[name = string("obj_197_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_197_cast_fp16 = batch_norm(beta = obj_197_beta_0_to_fp16, epsilon = obj_197_epsilon_0_to_fp16, gamma = obj_197_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_85_cast_fp16)[name = string("obj_197_cast_fp16")];
            string query_57_pad_type_0 = const()[name = string("query_57_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_57_strides_0 = const()[name = string("query_57_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_57_pad_0 = const()[name = string("query_57_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_57_dilations_0 = const()[name = string("query_57_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_57_groups_0 = const()[name = string("query_57_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284745856))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285532352))))[name = string("layers_14_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285540608)))];
            tensor<fp16, [1, 1024, 1, 1]> query_57_cast_fp16 = conv(bias = layers_14_self_attn_q_proj_bias_to_fp16, dilations = query_57_dilations_0, groups = query_57_groups_0, pad = query_57_pad_0, pad_type = query_57_pad_type_0, strides = query_57_strides_0, weight = layers_14_self_attn_q_proj_weight_to_fp16_palettized, x = obj_197_cast_fp16)[name = string("query_57_cast_fp16")];
            string current_key_29_pad_type_0 = const()[name = string("current_key_29_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_29_strides_0 = const()[name = string("current_key_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_29_pad_0 = const()[name = string("current_key_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_29_dilations_0 = const()[name = string("current_key_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_29_groups_0 = const()[name = string("current_key_29_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285542720))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286329216))))[name = string("layers_14_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_29_cast_fp16 = conv(dilations = current_key_29_dilations_0, groups = current_key_29_groups_0, pad = current_key_29_pad_0, pad_type = current_key_29_pad_type_0, strides = current_key_29_strides_0, weight = layers_14_self_attn_k_proj_weight_to_fp16_palettized, x = obj_197_cast_fp16)[name = string("current_key_29_cast_fp16")];
            string current_value_29_pad_type_0 = const()[name = string("current_value_29_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_29_strides_0 = const()[name = string("current_value_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_29_pad_0 = const()[name = string("current_value_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_29_dilations_0 = const()[name = string("current_value_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_29_groups_0 = const()[name = string("current_value_29_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286337472))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287123968))))[name = string("layers_14_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287132224)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_29_cast_fp16 = conv(bias = layers_14_self_attn_v_proj_bias_to_fp16, dilations = current_value_29_dilations_0, groups = current_value_29_groups_0, pad = current_value_29_pad_0, pad_type = current_value_29_pad_type_0, strides = current_value_29_strides_0, weight = layers_14_self_attn_v_proj_weight_to_fp16_palettized, x = obj_197_cast_fp16)[name = string("current_value_29_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3278_cast_fp16 = mul(x = var_87_cast_fp16_14, y = var_207_cast_fp16)[name = string("op_3278_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3279_cast_fp16 = mul(x = current_key_29_cast_fp16, y = var_205_cast_fp16)[name = string("op_3279_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_57_cast_fp16 = add(x = var_3278_cast_fp16, y = var_3279_cast_fp16)[name = string("key_57_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3282_cast_fp16 = mul(x = var_114_cast_fp16_14, y = var_207_cast_fp16)[name = string("op_3282_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3283_cast_fp16 = mul(x = current_value_29_cast_fp16, y = var_205_cast_fp16)[name = string("op_3283_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_57_cast_fp16 = add(x = var_3282_cast_fp16, y = var_3283_cast_fp16)[name = string("value_57_cast_fp16")];
            tensor<int32, [4]> var_3287 = const()[name = string("op_3287"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_57_cast_fp16 = reshape(shape = var_3287, x = query_57_cast_fp16)[name = string("mh_q_57_cast_fp16")];
            fp16 var_3289_to_fp16 = const()[name = string("op_3289_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_3290_cast_fp16 = mul(x = mh_q_57_cast_fp16, y = var_3289_to_fp16)[name = string("op_3290_cast_fp16")];
            tensor<int32, [4]> var_3293 = const()[name = string("op_3293"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_3294_cast_fp16 = reshape(shape = var_3293, x = key_57_cast_fp16)[name = string("op_3294_cast_fp16")];
            bool mh_w_85_transpose_x_0 = const()[name = string("mh_w_85_transpose_x_0"), val = bool(true)];
            bool mh_w_85_transpose_y_0 = const()[name = string("mh_w_85_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_85_cast_fp16 = matmul(transpose_x = mh_w_85_transpose_x_0, transpose_y = mh_w_85_transpose_y_0, x = var_3290_cast_fp16, y = var_3294_cast_fp16)[name = string("mh_w_85_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_87_cast_fp16 = add(x = mh_w_85_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_87_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_3302_cast_fp16 = softmax(axis = var_3214, x = mh_w_87_cast_fp16)[name = string("op_3302_cast_fp16")];
            tensor<int32, [4]> var_3303 = const()[name = string("op_3303"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_3304_cast_fp16 = reshape(shape = var_3303, x = value_57_cast_fp16)[name = string("op_3304_cast_fp16")];
            bool attn_57_transpose_x_0 = const()[name = string("attn_57_transpose_x_0"), val = bool(false)];
            bool attn_57_transpose_y_0 = const()[name = string("attn_57_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_57_cast_fp16 = matmul(transpose_x = attn_57_transpose_x_0, transpose_y = attn_57_transpose_y_0, x = var_3304_cast_fp16, y = var_3302_cast_fp16)[name = string("attn_57_cast_fp16")];
            tensor<int32, [4]> var_3307 = const()[name = string("op_3307"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_141_cast_fp16 = reshape(shape = var_3307, x = attn_57_cast_fp16)[name = string("input_141_cast_fp16")];
            string obj_203_pad_type_0 = const()[name = string("obj_203_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_203_strides_0 = const()[name = string("obj_203_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_203_pad_0 = const()[name = string("obj_203_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_203_dilations_0 = const()[name = string("obj_203_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_203_groups_0 = const()[name = string("obj_203_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287134336))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287920832))))[name = string("layers_14_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_14_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287929088)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_203_cast_fp16 = conv(bias = layers_14_self_attn_o_proj_bias_to_fp16, dilations = obj_203_dilations_0, groups = obj_203_groups_0, pad = obj_203_pad_0, pad_type = obj_203_pad_type_0, strides = obj_203_strides_0, weight = layers_14_self_attn_o_proj_weight_to_fp16_palettized, x = input_141_cast_fp16)[name = string("obj_203_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_87_cast_fp16 = add(x = inputs_85_cast_fp16, y = obj_203_cast_fp16)[name = string("inputs_87_cast_fp16")];
            tensor<int32, [1]> out_87_axes_0 = const()[name = string("out_87_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_3329_to_fp16 = const()[name = string("op_3329_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_87_cast_fp16 = layer_norm(axes = out_87_axes_0, epsilon = var_3329_to_fp16, x = inputs_87_cast_fp16)[name = string("out_87_cast_fp16")];
            tensor<fp16, [1024]> obj_205_gamma_0_to_fp16 = const()[name = string("obj_205_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287931200)))];
            tensor<fp16, [1024]> obj_205_beta_0_to_fp16 = const()[name = string("obj_205_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287933312)))];
            fp16 obj_205_epsilon_0_to_fp16 = const()[name = string("obj_205_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_205_cast_fp16 = batch_norm(beta = obj_205_beta_0_to_fp16, epsilon = obj_205_epsilon_0_to_fp16, gamma = obj_205_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_87_cast_fp16)[name = string("obj_205_cast_fp16")];
            string query_59_pad_type_0 = const()[name = string("query_59_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_59_strides_0 = const()[name = string("query_59_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_59_pad_0 = const()[name = string("query_59_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_59_dilations_0 = const()[name = string("query_59_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_59_groups_0 = const()[name = string("query_59_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287935424))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288721920))))[name = string("layers_14_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_14_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_14_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288730176)))];
            tensor<fp16, [1, 1024, 1, 1]> query_59_cast_fp16 = conv(bias = layers_14_encoder_attn_q_proj_bias_to_fp16, dilations = query_59_dilations_0, groups = query_59_groups_0, pad = query_59_pad_0, pad_type = query_59_pad_type_0, strides = query_59_strides_0, weight = layers_14_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_205_cast_fp16)[name = string("query_59_cast_fp16")];
            string key_59_pad_type_0 = const()[name = string("key_59_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_59_strides_0 = const()[name = string("key_59_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_59_pad_0 = const()[name = string("key_59_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_59_dilations_0 = const()[name = string("key_59_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_59_groups_0 = const()[name = string("key_59_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288732288))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289518784))))[name = string("layers_14_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_59_cast_fp16 = conv(dilations = key_59_dilations_0, groups = key_59_groups_0, pad = key_59_pad_0, pad_type = key_59_pad_type_0, strides = key_59_strides_0, weight = layers_14_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_59_cast_fp16")];
            string value_59_pad_type_0 = const()[name = string("value_59_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_59_strides_0 = const()[name = string("value_59_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_59_pad_0 = const()[name = string("value_59_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_59_dilations_0 = const()[name = string("value_59_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_59_groups_0 = const()[name = string("value_59_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289527040))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290313536))))[name = string("layers_14_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_14_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_14_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290321792)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_59_cast_fp16 = conv(bias = layers_14_encoder_attn_v_proj_bias_to_fp16, dilations = value_59_dilations_0, groups = value_59_groups_0, pad = value_59_pad_0, pad_type = value_59_pad_type_0, strides = value_59_strides_0, weight = layers_14_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_59_cast_fp16")];
            tensor<int32, [4]> var_3365 = const()[name = string("op_3365"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_59_cast_fp16 = reshape(shape = var_3365, x = query_59_cast_fp16)[name = string("mh_q_59_cast_fp16")];
            fp16 var_3367_to_fp16 = const()[name = string("op_3367_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_3368_cast_fp16 = mul(x = mh_q_59_cast_fp16, y = var_3367_to_fp16)[name = string("op_3368_cast_fp16")];
            tensor<int32, [4]> var_3371 = const()[name = string("op_3371"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_3372_cast_fp16 = reshape(shape = var_3371, x = key_59_cast_fp16)[name = string("op_3372_cast_fp16")];
            bool mh_w_89_transpose_x_0 = const()[name = string("mh_w_89_transpose_x_0"), val = bool(true)];
            bool mh_w_89_transpose_y_0 = const()[name = string("mh_w_89_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_89_cast_fp16 = matmul(transpose_x = mh_w_89_transpose_x_0, transpose_y = mh_w_89_transpose_y_0, x = var_3368_cast_fp16, y = var_3372_cast_fp16)[name = string("mh_w_89_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_209_cast_fp16 = softmax(axis = var_3214, x = mh_w_89_cast_fp16)[name = string("obj_209_cast_fp16")];
            tensor<int32, [4]> var_3376 = const()[name = string("op_3376"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_3377_cast_fp16 = reshape(shape = var_3376, x = value_59_cast_fp16)[name = string("op_3377_cast_fp16")];
            bool attn_59_transpose_x_0 = const()[name = string("attn_59_transpose_x_0"), val = bool(false)];
            bool attn_59_transpose_y_0 = const()[name = string("attn_59_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_59_cast_fp16 = matmul(transpose_x = attn_59_transpose_x_0, transpose_y = attn_59_transpose_y_0, x = var_3377_cast_fp16, y = obj_209_cast_fp16)[name = string("attn_59_cast_fp16")];
            tensor<int32, [4]> var_3380 = const()[name = string("op_3380"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_143_cast_fp16 = reshape(shape = var_3380, x = attn_59_cast_fp16)[name = string("input_143_cast_fp16")];
            string obj_207_pad_type_0 = const()[name = string("obj_207_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_207_strides_0 = const()[name = string("obj_207_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_207_pad_0 = const()[name = string("obj_207_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_207_dilations_0 = const()[name = string("obj_207_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_207_groups_0 = const()[name = string("obj_207_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290323904))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291110400))))[name = string("layers_14_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_14_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_14_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291118656)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_207_cast_fp16 = conv(bias = layers_14_encoder_attn_o_proj_bias_to_fp16, dilations = obj_207_dilations_0, groups = obj_207_groups_0, pad = obj_207_pad_0, pad_type = obj_207_pad_type_0, strides = obj_207_strides_0, weight = layers_14_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_143_cast_fp16)[name = string("obj_207_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_89_cast_fp16 = add(x = inputs_87_cast_fp16, y = obj_207_cast_fp16)[name = string("inputs_89_cast_fp16")];
            tensor<int32, [1]> out_89_axes_0 = const()[name = string("out_89_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_3398_to_fp16 = const()[name = string("op_3398_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_89_cast_fp16 = layer_norm(axes = out_89_axes_0, epsilon = var_3398_to_fp16, x = inputs_89_cast_fp16)[name = string("out_89_cast_fp16")];
            tensor<fp16, [1024]> input_145_gamma_0_to_fp16 = const()[name = string("input_145_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291120768)))];
            tensor<fp16, [1024]> input_145_beta_0_to_fp16 = const()[name = string("input_145_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291122880)))];
            fp16 input_145_epsilon_0_to_fp16 = const()[name = string("input_145_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_145_cast_fp16 = batch_norm(beta = input_145_beta_0_to_fp16, epsilon = input_145_epsilon_0_to_fp16, gamma = input_145_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_89_cast_fp16)[name = string("input_145_cast_fp16")];
            string input_147_pad_type_0 = const()[name = string("input_147_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_147_strides_0 = const()[name = string("input_147_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_147_pad_0 = const()[name = string("input_147_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_147_dilations_0 = const()[name = string("input_147_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_147_groups_0 = const()[name = string("input_147_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_14_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291124992))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294270784))))[name = string("layers_14_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_14_fc1_bias_to_fp16 = const()[name = string("layers_14_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294303616)))];
            tensor<fp16, [1, 4096, 1, 1]> input_147_cast_fp16 = conv(bias = layers_14_fc1_bias_to_fp16, dilations = input_147_dilations_0, groups = input_147_groups_0, pad = input_147_pad_0, pad_type = input_147_pad_type_0, strides = input_147_strides_0, weight = layers_14_fc1_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = string("input_147_cast_fp16")];
            string input_149_mode_0 = const()[name = string("input_149_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_149_cast_fp16 = gelu(mode = input_149_mode_0, x = input_147_cast_fp16)[name = string("input_149_cast_fp16")];
            string hidden_states_31_pad_type_0 = const()[name = string("hidden_states_31_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_31_strides_0 = const()[name = string("hidden_states_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_31_pad_0 = const()[name = string("hidden_states_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_31_dilations_0 = const()[name = string("hidden_states_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_31_groups_0 = const()[name = string("hidden_states_31_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_14_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294311872))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297457664))))[name = string("layers_14_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_14_fc2_bias_to_fp16 = const()[name = string("layers_14_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297465920)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_31_cast_fp16 = conv(bias = layers_14_fc2_bias_to_fp16, dilations = hidden_states_31_dilations_0, groups = hidden_states_31_groups_0, pad = hidden_states_31_pad_0, pad_type = hidden_states_31_pad_type_0, strides = hidden_states_31_strides_0, weight = layers_14_fc2_weight_to_fp16_palettized, x = input_149_cast_fp16)[name = string("hidden_states_31_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_91_cast_fp16 = add(x = inputs_89_cast_fp16, y = hidden_states_31_cast_fp16)[name = string("inputs_91_cast_fp16")];
            int32 var_3433 = const()[name = string("op_3433"), val = int32(3)];
            tensor<int32, [1]> out_91_axes_0 = const()[name = string("out_91_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_3458_to_fp16 = const()[name = string("op_3458_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_91_cast_fp16 = layer_norm(axes = out_91_axes_0, epsilon = var_3458_to_fp16, x = inputs_91_cast_fp16)[name = string("out_91_cast_fp16")];
            tensor<fp16, [1024]> obj_211_gamma_0_to_fp16 = const()[name = string("obj_211_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297468032)))];
            tensor<fp16, [1024]> obj_211_beta_0_to_fp16 = const()[name = string("obj_211_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297470144)))];
            fp16 obj_211_epsilon_0_to_fp16 = const()[name = string("obj_211_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_211_cast_fp16 = batch_norm(beta = obj_211_beta_0_to_fp16, epsilon = obj_211_epsilon_0_to_fp16, gamma = obj_211_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_91_cast_fp16)[name = string("obj_211_cast_fp16")];
            string query_61_pad_type_0 = const()[name = string("query_61_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_61_strides_0 = const()[name = string("query_61_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_61_pad_0 = const()[name = string("query_61_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_61_dilations_0 = const()[name = string("query_61_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_61_groups_0 = const()[name = string("query_61_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297472256))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298258752))))[name = string("layers_15_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298267008)))];
            tensor<fp16, [1, 1024, 1, 1]> query_61_cast_fp16 = conv(bias = layers_15_self_attn_q_proj_bias_to_fp16, dilations = query_61_dilations_0, groups = query_61_groups_0, pad = query_61_pad_0, pad_type = query_61_pad_type_0, strides = query_61_strides_0, weight = layers_15_self_attn_q_proj_weight_to_fp16_palettized, x = obj_211_cast_fp16)[name = string("query_61_cast_fp16")];
            string current_key_31_pad_type_0 = const()[name = string("current_key_31_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_31_strides_0 = const()[name = string("current_key_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_31_pad_0 = const()[name = string("current_key_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_31_dilations_0 = const()[name = string("current_key_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_31_groups_0 = const()[name = string("current_key_31_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298269120))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299055616))))[name = string("layers_15_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_31_cast_fp16 = conv(dilations = current_key_31_dilations_0, groups = current_key_31_groups_0, pad = current_key_31_pad_0, pad_type = current_key_31_pad_type_0, strides = current_key_31_strides_0, weight = layers_15_self_attn_k_proj_weight_to_fp16_palettized, x = obj_211_cast_fp16)[name = string("current_key_31_cast_fp16")];
            string current_value_31_pad_type_0 = const()[name = string("current_value_31_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_31_strides_0 = const()[name = string("current_value_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_31_pad_0 = const()[name = string("current_value_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_31_dilations_0 = const()[name = string("current_value_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_31_groups_0 = const()[name = string("current_value_31_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299063872))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299850368))))[name = string("layers_15_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299858624)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_31_cast_fp16 = conv(bias = layers_15_self_attn_v_proj_bias_to_fp16, dilations = current_value_31_dilations_0, groups = current_value_31_groups_0, pad = current_value_31_pad_0, pad_type = current_value_31_pad_type_0, strides = current_value_31_strides_0, weight = layers_15_self_attn_v_proj_weight_to_fp16_palettized, x = obj_211_cast_fp16)[name = string("current_value_31_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3497_cast_fp16 = mul(x = var_87_cast_fp16_15, y = var_207_cast_fp16)[name = string("op_3497_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3498_cast_fp16 = mul(x = current_key_31_cast_fp16, y = var_205_cast_fp16)[name = string("op_3498_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_61_cast_fp16 = add(x = var_3497_cast_fp16, y = var_3498_cast_fp16)[name = string("key_61_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3501_cast_fp16 = mul(x = var_114_cast_fp16_15, y = var_207_cast_fp16)[name = string("op_3501_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3502_cast_fp16 = mul(x = current_value_31_cast_fp16, y = var_205_cast_fp16)[name = string("op_3502_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_61_cast_fp16 = add(x = var_3501_cast_fp16, y = var_3502_cast_fp16)[name = string("value_61_cast_fp16")];
            tensor<int32, [4]> var_3506 = const()[name = string("op_3506"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_61_cast_fp16 = reshape(shape = var_3506, x = query_61_cast_fp16)[name = string("mh_q_61_cast_fp16")];
            fp16 var_3508_to_fp16 = const()[name = string("op_3508_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_3509_cast_fp16 = mul(x = mh_q_61_cast_fp16, y = var_3508_to_fp16)[name = string("op_3509_cast_fp16")];
            tensor<int32, [4]> var_3512 = const()[name = string("op_3512"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_3513_cast_fp16 = reshape(shape = var_3512, x = key_61_cast_fp16)[name = string("op_3513_cast_fp16")];
            bool mh_w_91_transpose_x_0 = const()[name = string("mh_w_91_transpose_x_0"), val = bool(true)];
            bool mh_w_91_transpose_y_0 = const()[name = string("mh_w_91_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_91_cast_fp16 = matmul(transpose_x = mh_w_91_transpose_x_0, transpose_y = mh_w_91_transpose_y_0, x = var_3509_cast_fp16, y = var_3513_cast_fp16)[name = string("mh_w_91_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_93_cast_fp16 = add(x = mh_w_91_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_93_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_3521_cast_fp16 = softmax(axis = var_3433, x = mh_w_93_cast_fp16)[name = string("op_3521_cast_fp16")];
            tensor<int32, [4]> var_3522 = const()[name = string("op_3522"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_3523_cast_fp16 = reshape(shape = var_3522, x = value_61_cast_fp16)[name = string("op_3523_cast_fp16")];
            bool attn_61_transpose_x_0 = const()[name = string("attn_61_transpose_x_0"), val = bool(false)];
            bool attn_61_transpose_y_0 = const()[name = string("attn_61_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_61_cast_fp16 = matmul(transpose_x = attn_61_transpose_x_0, transpose_y = attn_61_transpose_y_0, x = var_3523_cast_fp16, y = var_3521_cast_fp16)[name = string("attn_61_cast_fp16")];
            tensor<int32, [4]> var_3526 = const()[name = string("op_3526"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_151_cast_fp16 = reshape(shape = var_3526, x = attn_61_cast_fp16)[name = string("input_151_cast_fp16")];
            string obj_217_pad_type_0 = const()[name = string("obj_217_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_217_strides_0 = const()[name = string("obj_217_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_217_pad_0 = const()[name = string("obj_217_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_217_dilations_0 = const()[name = string("obj_217_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_217_groups_0 = const()[name = string("obj_217_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299860736))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300647232))))[name = string("layers_15_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_15_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300655488)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_217_cast_fp16 = conv(bias = layers_15_self_attn_o_proj_bias_to_fp16, dilations = obj_217_dilations_0, groups = obj_217_groups_0, pad = obj_217_pad_0, pad_type = obj_217_pad_type_0, strides = obj_217_strides_0, weight = layers_15_self_attn_o_proj_weight_to_fp16_palettized, x = input_151_cast_fp16)[name = string("obj_217_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_93_cast_fp16 = add(x = inputs_91_cast_fp16, y = obj_217_cast_fp16)[name = string("inputs_93_cast_fp16")];
            tensor<int32, [1]> out_93_axes_0 = const()[name = string("out_93_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_3548_to_fp16 = const()[name = string("op_3548_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_93_cast_fp16 = layer_norm(axes = out_93_axes_0, epsilon = var_3548_to_fp16, x = inputs_93_cast_fp16)[name = string("out_93_cast_fp16")];
            tensor<fp16, [1024]> obj_219_gamma_0_to_fp16 = const()[name = string("obj_219_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300657600)))];
            tensor<fp16, [1024]> obj_219_beta_0_to_fp16 = const()[name = string("obj_219_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300659712)))];
            fp16 obj_219_epsilon_0_to_fp16 = const()[name = string("obj_219_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_219_cast_fp16 = batch_norm(beta = obj_219_beta_0_to_fp16, epsilon = obj_219_epsilon_0_to_fp16, gamma = obj_219_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_93_cast_fp16)[name = string("obj_219_cast_fp16")];
            string query_63_pad_type_0 = const()[name = string("query_63_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_63_strides_0 = const()[name = string("query_63_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_63_pad_0 = const()[name = string("query_63_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_63_dilations_0 = const()[name = string("query_63_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_63_groups_0 = const()[name = string("query_63_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300661824))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301448320))))[name = string("layers_15_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_15_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_15_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301456576)))];
            tensor<fp16, [1, 1024, 1, 1]> query_63_cast_fp16 = conv(bias = layers_15_encoder_attn_q_proj_bias_to_fp16, dilations = query_63_dilations_0, groups = query_63_groups_0, pad = query_63_pad_0, pad_type = query_63_pad_type_0, strides = query_63_strides_0, weight = layers_15_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_219_cast_fp16)[name = string("query_63_cast_fp16")];
            string key_63_pad_type_0 = const()[name = string("key_63_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_63_strides_0 = const()[name = string("key_63_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_63_pad_0 = const()[name = string("key_63_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_63_dilations_0 = const()[name = string("key_63_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_63_groups_0 = const()[name = string("key_63_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301458688))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302245184))))[name = string("layers_15_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_63_cast_fp16 = conv(dilations = key_63_dilations_0, groups = key_63_groups_0, pad = key_63_pad_0, pad_type = key_63_pad_type_0, strides = key_63_strides_0, weight = layers_15_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_63_cast_fp16")];
            string value_63_pad_type_0 = const()[name = string("value_63_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_63_strides_0 = const()[name = string("value_63_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_63_pad_0 = const()[name = string("value_63_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_63_dilations_0 = const()[name = string("value_63_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_63_groups_0 = const()[name = string("value_63_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302253440))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303039936))))[name = string("layers_15_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_15_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_15_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303048192)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_63_cast_fp16 = conv(bias = layers_15_encoder_attn_v_proj_bias_to_fp16, dilations = value_63_dilations_0, groups = value_63_groups_0, pad = value_63_pad_0, pad_type = value_63_pad_type_0, strides = value_63_strides_0, weight = layers_15_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_63_cast_fp16")];
            tensor<int32, [4]> var_3584 = const()[name = string("op_3584"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_63_cast_fp16 = reshape(shape = var_3584, x = query_63_cast_fp16)[name = string("mh_q_63_cast_fp16")];
            fp16 var_3586_to_fp16 = const()[name = string("op_3586_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_3587_cast_fp16 = mul(x = mh_q_63_cast_fp16, y = var_3586_to_fp16)[name = string("op_3587_cast_fp16")];
            tensor<int32, [4]> var_3590 = const()[name = string("op_3590"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_3591_cast_fp16 = reshape(shape = var_3590, x = key_63_cast_fp16)[name = string("op_3591_cast_fp16")];
            bool mh_w_95_transpose_x_0 = const()[name = string("mh_w_95_transpose_x_0"), val = bool(true)];
            bool mh_w_95_transpose_y_0 = const()[name = string("mh_w_95_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_95_cast_fp16 = matmul(transpose_x = mh_w_95_transpose_x_0, transpose_y = mh_w_95_transpose_y_0, x = var_3587_cast_fp16, y = var_3591_cast_fp16)[name = string("mh_w_95_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_223_cast_fp16 = softmax(axis = var_3433, x = mh_w_95_cast_fp16)[name = string("obj_223_cast_fp16")];
            tensor<int32, [4]> var_3595 = const()[name = string("op_3595"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_3596_cast_fp16 = reshape(shape = var_3595, x = value_63_cast_fp16)[name = string("op_3596_cast_fp16")];
            bool attn_63_transpose_x_0 = const()[name = string("attn_63_transpose_x_0"), val = bool(false)];
            bool attn_63_transpose_y_0 = const()[name = string("attn_63_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_63_cast_fp16 = matmul(transpose_x = attn_63_transpose_x_0, transpose_y = attn_63_transpose_y_0, x = var_3596_cast_fp16, y = obj_223_cast_fp16)[name = string("attn_63_cast_fp16")];
            tensor<int32, [4]> var_3599 = const()[name = string("op_3599"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_153_cast_fp16 = reshape(shape = var_3599, x = attn_63_cast_fp16)[name = string("input_153_cast_fp16")];
            string obj_221_pad_type_0 = const()[name = string("obj_221_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_221_strides_0 = const()[name = string("obj_221_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_221_pad_0 = const()[name = string("obj_221_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_221_dilations_0 = const()[name = string("obj_221_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_221_groups_0 = const()[name = string("obj_221_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303050304))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303836800))))[name = string("layers_15_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_15_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_15_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303845056)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_221_cast_fp16 = conv(bias = layers_15_encoder_attn_o_proj_bias_to_fp16, dilations = obj_221_dilations_0, groups = obj_221_groups_0, pad = obj_221_pad_0, pad_type = obj_221_pad_type_0, strides = obj_221_strides_0, weight = layers_15_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = string("obj_221_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_95_cast_fp16 = add(x = inputs_93_cast_fp16, y = obj_221_cast_fp16)[name = string("inputs_95_cast_fp16")];
            tensor<int32, [1]> out_95_axes_0 = const()[name = string("out_95_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_3620_to_fp16 = const()[name = string("op_3620_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_95_cast_fp16 = layer_norm(axes = out_95_axes_0, epsilon = var_3620_to_fp16, x = inputs_95_cast_fp16)[name = string("out_95_cast_fp16")];
            tensor<fp16, [1024]> input_155_gamma_0_to_fp16 = const()[name = string("input_155_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303847168)))];
            tensor<fp16, [1024]> input_155_beta_0_to_fp16 = const()[name = string("input_155_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303849280)))];
            fp16 input_155_epsilon_0_to_fp16 = const()[name = string("input_155_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_155_cast_fp16 = batch_norm(beta = input_155_beta_0_to_fp16, epsilon = input_155_epsilon_0_to_fp16, gamma = input_155_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_95_cast_fp16)[name = string("input_155_cast_fp16")];
            string input_157_pad_type_0 = const()[name = string("input_157_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_157_strides_0 = const()[name = string("input_157_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_157_pad_0 = const()[name = string("input_157_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_157_dilations_0 = const()[name = string("input_157_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_157_groups_0 = const()[name = string("input_157_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_15_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303851392))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306997184))))[name = string("layers_15_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_15_fc1_bias_to_fp16 = const()[name = string("layers_15_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307030016)))];
            tensor<fp16, [1, 4096, 1, 1]> input_157_cast_fp16 = conv(bias = layers_15_fc1_bias_to_fp16, dilations = input_157_dilations_0, groups = input_157_groups_0, pad = input_157_pad_0, pad_type = input_157_pad_type_0, strides = input_157_strides_0, weight = layers_15_fc1_weight_to_fp16_palettized, x = input_155_cast_fp16)[name = string("input_157_cast_fp16")];
            string input_159_mode_0 = const()[name = string("input_159_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_159_cast_fp16 = gelu(mode = input_159_mode_0, x = input_157_cast_fp16)[name = string("input_159_cast_fp16")];
            string hidden_states_33_pad_type_0 = const()[name = string("hidden_states_33_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_33_strides_0 = const()[name = string("hidden_states_33_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_33_pad_0 = const()[name = string("hidden_states_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_33_dilations_0 = const()[name = string("hidden_states_33_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_33_groups_0 = const()[name = string("hidden_states_33_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_15_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307038272))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310184064))))[name = string("layers_15_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_15_fc2_bias_to_fp16 = const()[name = string("layers_15_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310192320)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_33_cast_fp16 = conv(bias = layers_15_fc2_bias_to_fp16, dilations = hidden_states_33_dilations_0, groups = hidden_states_33_groups_0, pad = hidden_states_33_pad_0, pad_type = hidden_states_33_pad_type_0, strides = hidden_states_33_strides_0, weight = layers_15_fc2_weight_to_fp16_palettized, x = input_159_cast_fp16)[name = string("hidden_states_33_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_97_cast_fp16 = add(x = inputs_95_cast_fp16, y = hidden_states_33_cast_fp16)[name = string("inputs_97_cast_fp16")];
            int32 var_3656 = const()[name = string("op_3656"), val = int32(3)];
            tensor<int32, [1]> out_97_axes_0 = const()[name = string("out_97_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_3681_to_fp16 = const()[name = string("op_3681_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_97_cast_fp16 = layer_norm(axes = out_97_axes_0, epsilon = var_3681_to_fp16, x = inputs_97_cast_fp16)[name = string("out_97_cast_fp16")];
            tensor<fp16, [1024]> obj_225_gamma_0_to_fp16 = const()[name = string("obj_225_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310194432)))];
            tensor<fp16, [1024]> obj_225_beta_0_to_fp16 = const()[name = string("obj_225_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310196544)))];
            fp16 obj_225_epsilon_0_to_fp16 = const()[name = string("obj_225_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_225_cast_fp16 = batch_norm(beta = obj_225_beta_0_to_fp16, epsilon = obj_225_epsilon_0_to_fp16, gamma = obj_225_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_97_cast_fp16)[name = string("obj_225_cast_fp16")];
            string query_65_pad_type_0 = const()[name = string("query_65_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_65_strides_0 = const()[name = string("query_65_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_65_pad_0 = const()[name = string("query_65_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_65_dilations_0 = const()[name = string("query_65_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_65_groups_0 = const()[name = string("query_65_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310198656))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310985152))))[name = string("layers_16_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310993408)))];
            tensor<fp16, [1, 1024, 1, 1]> query_65_cast_fp16 = conv(bias = layers_16_self_attn_q_proj_bias_to_fp16, dilations = query_65_dilations_0, groups = query_65_groups_0, pad = query_65_pad_0, pad_type = query_65_pad_type_0, strides = query_65_strides_0, weight = layers_16_self_attn_q_proj_weight_to_fp16_palettized, x = obj_225_cast_fp16)[name = string("query_65_cast_fp16")];
            string current_key_33_pad_type_0 = const()[name = string("current_key_33_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_33_strides_0 = const()[name = string("current_key_33_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_33_pad_0 = const()[name = string("current_key_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_33_dilations_0 = const()[name = string("current_key_33_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_33_groups_0 = const()[name = string("current_key_33_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310995520))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311782016))))[name = string("layers_16_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_33_cast_fp16 = conv(dilations = current_key_33_dilations_0, groups = current_key_33_groups_0, pad = current_key_33_pad_0, pad_type = current_key_33_pad_type_0, strides = current_key_33_strides_0, weight = layers_16_self_attn_k_proj_weight_to_fp16_palettized, x = obj_225_cast_fp16)[name = string("current_key_33_cast_fp16")];
            string current_value_33_pad_type_0 = const()[name = string("current_value_33_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_33_strides_0 = const()[name = string("current_value_33_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_33_pad_0 = const()[name = string("current_value_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_33_dilations_0 = const()[name = string("current_value_33_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_33_groups_0 = const()[name = string("current_value_33_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311790272))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312576768))))[name = string("layers_16_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312585024)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_33_cast_fp16 = conv(bias = layers_16_self_attn_v_proj_bias_to_fp16, dilations = current_value_33_dilations_0, groups = current_value_33_groups_0, pad = current_value_33_pad_0, pad_type = current_value_33_pad_type_0, strides = current_value_33_strides_0, weight = layers_16_self_attn_v_proj_weight_to_fp16_palettized, x = obj_225_cast_fp16)[name = string("current_value_33_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3720_cast_fp16 = mul(x = var_87_cast_fp16_16, y = var_207_cast_fp16)[name = string("op_3720_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3721_cast_fp16 = mul(x = current_key_33_cast_fp16, y = var_205_cast_fp16)[name = string("op_3721_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_65_cast_fp16 = add(x = var_3720_cast_fp16, y = var_3721_cast_fp16)[name = string("key_65_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3724_cast_fp16 = mul(x = var_114_cast_fp16_16, y = var_207_cast_fp16)[name = string("op_3724_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3725_cast_fp16 = mul(x = current_value_33_cast_fp16, y = var_205_cast_fp16)[name = string("op_3725_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_65_cast_fp16 = add(x = var_3724_cast_fp16, y = var_3725_cast_fp16)[name = string("value_65_cast_fp16")];
            tensor<int32, [4]> var_3729 = const()[name = string("op_3729"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_65_cast_fp16 = reshape(shape = var_3729, x = query_65_cast_fp16)[name = string("mh_q_65_cast_fp16")];
            fp16 var_3731_to_fp16 = const()[name = string("op_3731_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_3732_cast_fp16 = mul(x = mh_q_65_cast_fp16, y = var_3731_to_fp16)[name = string("op_3732_cast_fp16")];
            tensor<int32, [4]> var_3735 = const()[name = string("op_3735"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_3736_cast_fp16 = reshape(shape = var_3735, x = key_65_cast_fp16)[name = string("op_3736_cast_fp16")];
            bool mh_w_97_transpose_x_0 = const()[name = string("mh_w_97_transpose_x_0"), val = bool(true)];
            bool mh_w_97_transpose_y_0 = const()[name = string("mh_w_97_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_97_cast_fp16 = matmul(transpose_x = mh_w_97_transpose_x_0, transpose_y = mh_w_97_transpose_y_0, x = var_3732_cast_fp16, y = var_3736_cast_fp16)[name = string("mh_w_97_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_99_cast_fp16 = add(x = mh_w_97_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_99_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_3744_cast_fp16 = softmax(axis = var_3656, x = mh_w_99_cast_fp16)[name = string("op_3744_cast_fp16")];
            tensor<int32, [4]> var_3745 = const()[name = string("op_3745"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_3746_cast_fp16 = reshape(shape = var_3745, x = value_65_cast_fp16)[name = string("op_3746_cast_fp16")];
            bool attn_65_transpose_x_0 = const()[name = string("attn_65_transpose_x_0"), val = bool(false)];
            bool attn_65_transpose_y_0 = const()[name = string("attn_65_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_65_cast_fp16 = matmul(transpose_x = attn_65_transpose_x_0, transpose_y = attn_65_transpose_y_0, x = var_3746_cast_fp16, y = var_3744_cast_fp16)[name = string("attn_65_cast_fp16")];
            tensor<int32, [4]> var_3749 = const()[name = string("op_3749"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_161_cast_fp16 = reshape(shape = var_3749, x = attn_65_cast_fp16)[name = string("input_161_cast_fp16")];
            string obj_231_pad_type_0 = const()[name = string("obj_231_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_231_strides_0 = const()[name = string("obj_231_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_231_pad_0 = const()[name = string("obj_231_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_231_dilations_0 = const()[name = string("obj_231_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_231_groups_0 = const()[name = string("obj_231_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312587136))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313373632))))[name = string("layers_16_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_16_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313381888)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_231_cast_fp16 = conv(bias = layers_16_self_attn_o_proj_bias_to_fp16, dilations = obj_231_dilations_0, groups = obj_231_groups_0, pad = obj_231_pad_0, pad_type = obj_231_pad_type_0, strides = obj_231_strides_0, weight = layers_16_self_attn_o_proj_weight_to_fp16_palettized, x = input_161_cast_fp16)[name = string("obj_231_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_99_cast_fp16 = add(x = inputs_97_cast_fp16, y = obj_231_cast_fp16)[name = string("inputs_99_cast_fp16")];
            tensor<int32, [1]> out_99_axes_0 = const()[name = string("out_99_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_3771_to_fp16 = const()[name = string("op_3771_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_99_cast_fp16 = layer_norm(axes = out_99_axes_0, epsilon = var_3771_to_fp16, x = inputs_99_cast_fp16)[name = string("out_99_cast_fp16")];
            tensor<fp16, [1024]> obj_233_gamma_0_to_fp16 = const()[name = string("obj_233_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313384000)))];
            tensor<fp16, [1024]> obj_233_beta_0_to_fp16 = const()[name = string("obj_233_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313386112)))];
            fp16 obj_233_epsilon_0_to_fp16 = const()[name = string("obj_233_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_233_cast_fp16 = batch_norm(beta = obj_233_beta_0_to_fp16, epsilon = obj_233_epsilon_0_to_fp16, gamma = obj_233_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_99_cast_fp16)[name = string("obj_233_cast_fp16")];
            string query_67_pad_type_0 = const()[name = string("query_67_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_67_strides_0 = const()[name = string("query_67_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_67_pad_0 = const()[name = string("query_67_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_67_dilations_0 = const()[name = string("query_67_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_67_groups_0 = const()[name = string("query_67_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313388224))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314174720))))[name = string("layers_16_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_16_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_16_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314182976)))];
            tensor<fp16, [1, 1024, 1, 1]> query_67_cast_fp16 = conv(bias = layers_16_encoder_attn_q_proj_bias_to_fp16, dilations = query_67_dilations_0, groups = query_67_groups_0, pad = query_67_pad_0, pad_type = query_67_pad_type_0, strides = query_67_strides_0, weight = layers_16_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_233_cast_fp16)[name = string("query_67_cast_fp16")];
            string key_67_pad_type_0 = const()[name = string("key_67_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_67_strides_0 = const()[name = string("key_67_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_67_pad_0 = const()[name = string("key_67_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_67_dilations_0 = const()[name = string("key_67_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_67_groups_0 = const()[name = string("key_67_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314185088))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314971584))))[name = string("layers_16_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_67_cast_fp16 = conv(dilations = key_67_dilations_0, groups = key_67_groups_0, pad = key_67_pad_0, pad_type = key_67_pad_type_0, strides = key_67_strides_0, weight = layers_16_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_67_cast_fp16")];
            string value_67_pad_type_0 = const()[name = string("value_67_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_67_strides_0 = const()[name = string("value_67_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_67_pad_0 = const()[name = string("value_67_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_67_dilations_0 = const()[name = string("value_67_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_67_groups_0 = const()[name = string("value_67_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314979840))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315766336))))[name = string("layers_16_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_16_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_16_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315774592)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_67_cast_fp16 = conv(bias = layers_16_encoder_attn_v_proj_bias_to_fp16, dilations = value_67_dilations_0, groups = value_67_groups_0, pad = value_67_pad_0, pad_type = value_67_pad_type_0, strides = value_67_strides_0, weight = layers_16_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_67_cast_fp16")];
            tensor<int32, [4]> var_3807 = const()[name = string("op_3807"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_67_cast_fp16 = reshape(shape = var_3807, x = query_67_cast_fp16)[name = string("mh_q_67_cast_fp16")];
            fp16 var_3809_to_fp16 = const()[name = string("op_3809_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_3810_cast_fp16 = mul(x = mh_q_67_cast_fp16, y = var_3809_to_fp16)[name = string("op_3810_cast_fp16")];
            tensor<int32, [4]> var_3813 = const()[name = string("op_3813"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_3814_cast_fp16 = reshape(shape = var_3813, x = key_67_cast_fp16)[name = string("op_3814_cast_fp16")];
            bool mh_w_101_transpose_x_0 = const()[name = string("mh_w_101_transpose_x_0"), val = bool(true)];
            bool mh_w_101_transpose_y_0 = const()[name = string("mh_w_101_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_101_cast_fp16 = matmul(transpose_x = mh_w_101_transpose_x_0, transpose_y = mh_w_101_transpose_y_0, x = var_3810_cast_fp16, y = var_3814_cast_fp16)[name = string("mh_w_101_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_237_cast_fp16 = softmax(axis = var_3656, x = mh_w_101_cast_fp16)[name = string("obj_237_cast_fp16")];
            tensor<int32, [4]> var_3818 = const()[name = string("op_3818"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_3819_cast_fp16 = reshape(shape = var_3818, x = value_67_cast_fp16)[name = string("op_3819_cast_fp16")];
            bool attn_67_transpose_x_0 = const()[name = string("attn_67_transpose_x_0"), val = bool(false)];
            bool attn_67_transpose_y_0 = const()[name = string("attn_67_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_67_cast_fp16 = matmul(transpose_x = attn_67_transpose_x_0, transpose_y = attn_67_transpose_y_0, x = var_3819_cast_fp16, y = obj_237_cast_fp16)[name = string("attn_67_cast_fp16")];
            tensor<int32, [4]> var_3822 = const()[name = string("op_3822"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_163_cast_fp16 = reshape(shape = var_3822, x = attn_67_cast_fp16)[name = string("input_163_cast_fp16")];
            string obj_235_pad_type_0 = const()[name = string("obj_235_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_235_strides_0 = const()[name = string("obj_235_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_235_pad_0 = const()[name = string("obj_235_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_235_dilations_0 = const()[name = string("obj_235_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_235_groups_0 = const()[name = string("obj_235_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315776704))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316563200))))[name = string("layers_16_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_16_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_16_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316571456)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_235_cast_fp16 = conv(bias = layers_16_encoder_attn_o_proj_bias_to_fp16, dilations = obj_235_dilations_0, groups = obj_235_groups_0, pad = obj_235_pad_0, pad_type = obj_235_pad_type_0, strides = obj_235_strides_0, weight = layers_16_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_163_cast_fp16)[name = string("obj_235_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_101_cast_fp16 = add(x = inputs_99_cast_fp16, y = obj_235_cast_fp16)[name = string("inputs_101_cast_fp16")];
            tensor<int32, [1]> out_101_axes_0 = const()[name = string("out_101_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_3843_to_fp16 = const()[name = string("op_3843_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_101_cast_fp16 = layer_norm(axes = out_101_axes_0, epsilon = var_3843_to_fp16, x = inputs_101_cast_fp16)[name = string("out_101_cast_fp16")];
            tensor<fp16, [1024]> input_165_gamma_0_to_fp16 = const()[name = string("input_165_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316573568)))];
            tensor<fp16, [1024]> input_165_beta_0_to_fp16 = const()[name = string("input_165_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316575680)))];
            fp16 input_165_epsilon_0_to_fp16 = const()[name = string("input_165_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_165_cast_fp16 = batch_norm(beta = input_165_beta_0_to_fp16, epsilon = input_165_epsilon_0_to_fp16, gamma = input_165_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_101_cast_fp16)[name = string("input_165_cast_fp16")];
            string input_167_pad_type_0 = const()[name = string("input_167_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_167_strides_0 = const()[name = string("input_167_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_167_pad_0 = const()[name = string("input_167_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_167_dilations_0 = const()[name = string("input_167_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_167_groups_0 = const()[name = string("input_167_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_16_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316577792))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319723584))))[name = string("layers_16_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_16_fc1_bias_to_fp16 = const()[name = string("layers_16_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319756416)))];
            tensor<fp16, [1, 4096, 1, 1]> input_167_cast_fp16 = conv(bias = layers_16_fc1_bias_to_fp16, dilations = input_167_dilations_0, groups = input_167_groups_0, pad = input_167_pad_0, pad_type = input_167_pad_type_0, strides = input_167_strides_0, weight = layers_16_fc1_weight_to_fp16_palettized, x = input_165_cast_fp16)[name = string("input_167_cast_fp16")];
            string input_169_mode_0 = const()[name = string("input_169_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_169_cast_fp16 = gelu(mode = input_169_mode_0, x = input_167_cast_fp16)[name = string("input_169_cast_fp16")];
            string hidden_states_35_pad_type_0 = const()[name = string("hidden_states_35_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_35_strides_0 = const()[name = string("hidden_states_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_35_pad_0 = const()[name = string("hidden_states_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_35_dilations_0 = const()[name = string("hidden_states_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_35_groups_0 = const()[name = string("hidden_states_35_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_16_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319764672))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(322910464))))[name = string("layers_16_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_16_fc2_bias_to_fp16 = const()[name = string("layers_16_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(322918720)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_35_cast_fp16 = conv(bias = layers_16_fc2_bias_to_fp16, dilations = hidden_states_35_dilations_0, groups = hidden_states_35_groups_0, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = hidden_states_35_strides_0, weight = layers_16_fc2_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = string("hidden_states_35_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_103_cast_fp16 = add(x = inputs_101_cast_fp16, y = hidden_states_35_cast_fp16)[name = string("inputs_103_cast_fp16")];
            int32 var_3879 = const()[name = string("op_3879"), val = int32(3)];
            tensor<int32, [1]> out_103_axes_0 = const()[name = string("out_103_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_3904_to_fp16 = const()[name = string("op_3904_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_103_cast_fp16 = layer_norm(axes = out_103_axes_0, epsilon = var_3904_to_fp16, x = inputs_103_cast_fp16)[name = string("out_103_cast_fp16")];
            tensor<fp16, [1024]> obj_239_gamma_0_to_fp16 = const()[name = string("obj_239_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(322920832)))];
            tensor<fp16, [1024]> obj_239_beta_0_to_fp16 = const()[name = string("obj_239_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(322922944)))];
            fp16 obj_239_epsilon_0_to_fp16 = const()[name = string("obj_239_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_239_cast_fp16 = batch_norm(beta = obj_239_beta_0_to_fp16, epsilon = obj_239_epsilon_0_to_fp16, gamma = obj_239_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_103_cast_fp16)[name = string("obj_239_cast_fp16")];
            string query_69_pad_type_0 = const()[name = string("query_69_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_69_strides_0 = const()[name = string("query_69_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_69_pad_0 = const()[name = string("query_69_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_69_dilations_0 = const()[name = string("query_69_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_69_groups_0 = const()[name = string("query_69_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(322925056))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(323711552))))[name = string("layers_17_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(323719808)))];
            tensor<fp16, [1, 1024, 1, 1]> query_69_cast_fp16 = conv(bias = layers_17_self_attn_q_proj_bias_to_fp16, dilations = query_69_dilations_0, groups = query_69_groups_0, pad = query_69_pad_0, pad_type = query_69_pad_type_0, strides = query_69_strides_0, weight = layers_17_self_attn_q_proj_weight_to_fp16_palettized, x = obj_239_cast_fp16)[name = string("query_69_cast_fp16")];
            string current_key_35_pad_type_0 = const()[name = string("current_key_35_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_35_strides_0 = const()[name = string("current_key_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_35_pad_0 = const()[name = string("current_key_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_35_dilations_0 = const()[name = string("current_key_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_35_groups_0 = const()[name = string("current_key_35_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(323721920))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(324508416))))[name = string("layers_17_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_35_cast_fp16 = conv(dilations = current_key_35_dilations_0, groups = current_key_35_groups_0, pad = current_key_35_pad_0, pad_type = current_key_35_pad_type_0, strides = current_key_35_strides_0, weight = layers_17_self_attn_k_proj_weight_to_fp16_palettized, x = obj_239_cast_fp16)[name = string("current_key_35_cast_fp16")];
            string current_value_35_pad_type_0 = const()[name = string("current_value_35_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_35_strides_0 = const()[name = string("current_value_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_35_pad_0 = const()[name = string("current_value_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_35_dilations_0 = const()[name = string("current_value_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_35_groups_0 = const()[name = string("current_value_35_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(324516672))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(325303168))))[name = string("layers_17_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(325311424)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_35_cast_fp16 = conv(bias = layers_17_self_attn_v_proj_bias_to_fp16, dilations = current_value_35_dilations_0, groups = current_value_35_groups_0, pad = current_value_35_pad_0, pad_type = current_value_35_pad_type_0, strides = current_value_35_strides_0, weight = layers_17_self_attn_v_proj_weight_to_fp16_palettized, x = obj_239_cast_fp16)[name = string("current_value_35_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3943_cast_fp16 = mul(x = var_87_cast_fp16_17, y = var_207_cast_fp16)[name = string("op_3943_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3944_cast_fp16 = mul(x = current_key_35_cast_fp16, y = var_205_cast_fp16)[name = string("op_3944_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_69_cast_fp16 = add(x = var_3943_cast_fp16, y = var_3944_cast_fp16)[name = string("key_69_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3947_cast_fp16 = mul(x = var_114_cast_fp16_17, y = var_207_cast_fp16)[name = string("op_3947_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_3948_cast_fp16 = mul(x = current_value_35_cast_fp16, y = var_205_cast_fp16)[name = string("op_3948_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_69_cast_fp16 = add(x = var_3947_cast_fp16, y = var_3948_cast_fp16)[name = string("value_69_cast_fp16")];
            tensor<int32, [4]> var_3952 = const()[name = string("op_3952"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_69_cast_fp16 = reshape(shape = var_3952, x = query_69_cast_fp16)[name = string("mh_q_69_cast_fp16")];
            fp16 var_3954_to_fp16 = const()[name = string("op_3954_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_3955_cast_fp16 = mul(x = mh_q_69_cast_fp16, y = var_3954_to_fp16)[name = string("op_3955_cast_fp16")];
            tensor<int32, [4]> var_3958 = const()[name = string("op_3958"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_3959_cast_fp16 = reshape(shape = var_3958, x = key_69_cast_fp16)[name = string("op_3959_cast_fp16")];
            bool mh_w_103_transpose_x_0 = const()[name = string("mh_w_103_transpose_x_0"), val = bool(true)];
            bool mh_w_103_transpose_y_0 = const()[name = string("mh_w_103_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_103_cast_fp16 = matmul(transpose_x = mh_w_103_transpose_x_0, transpose_y = mh_w_103_transpose_y_0, x = var_3955_cast_fp16, y = var_3959_cast_fp16)[name = string("mh_w_103_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_105_cast_fp16 = add(x = mh_w_103_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_105_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_3967_cast_fp16 = softmax(axis = var_3879, x = mh_w_105_cast_fp16)[name = string("op_3967_cast_fp16")];
            tensor<int32, [4]> var_3968 = const()[name = string("op_3968"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_3969_cast_fp16 = reshape(shape = var_3968, x = value_69_cast_fp16)[name = string("op_3969_cast_fp16")];
            bool attn_69_transpose_x_0 = const()[name = string("attn_69_transpose_x_0"), val = bool(false)];
            bool attn_69_transpose_y_0 = const()[name = string("attn_69_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_69_cast_fp16 = matmul(transpose_x = attn_69_transpose_x_0, transpose_y = attn_69_transpose_y_0, x = var_3969_cast_fp16, y = var_3967_cast_fp16)[name = string("attn_69_cast_fp16")];
            tensor<int32, [4]> var_3972 = const()[name = string("op_3972"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_171_cast_fp16 = reshape(shape = var_3972, x = attn_69_cast_fp16)[name = string("input_171_cast_fp16")];
            string obj_245_pad_type_0 = const()[name = string("obj_245_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_245_strides_0 = const()[name = string("obj_245_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_245_pad_0 = const()[name = string("obj_245_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_245_dilations_0 = const()[name = string("obj_245_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_245_groups_0 = const()[name = string("obj_245_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(325313536))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326100032))))[name = string("layers_17_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_17_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326108288)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_245_cast_fp16 = conv(bias = layers_17_self_attn_o_proj_bias_to_fp16, dilations = obj_245_dilations_0, groups = obj_245_groups_0, pad = obj_245_pad_0, pad_type = obj_245_pad_type_0, strides = obj_245_strides_0, weight = layers_17_self_attn_o_proj_weight_to_fp16_palettized, x = input_171_cast_fp16)[name = string("obj_245_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_105_cast_fp16 = add(x = inputs_103_cast_fp16, y = obj_245_cast_fp16)[name = string("inputs_105_cast_fp16")];
            tensor<int32, [1]> out_105_axes_0 = const()[name = string("out_105_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_3994_to_fp16 = const()[name = string("op_3994_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_105_cast_fp16 = layer_norm(axes = out_105_axes_0, epsilon = var_3994_to_fp16, x = inputs_105_cast_fp16)[name = string("out_105_cast_fp16")];
            tensor<fp16, [1024]> obj_247_gamma_0_to_fp16 = const()[name = string("obj_247_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326110400)))];
            tensor<fp16, [1024]> obj_247_beta_0_to_fp16 = const()[name = string("obj_247_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326112512)))];
            fp16 obj_247_epsilon_0_to_fp16 = const()[name = string("obj_247_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_247_cast_fp16 = batch_norm(beta = obj_247_beta_0_to_fp16, epsilon = obj_247_epsilon_0_to_fp16, gamma = obj_247_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_105_cast_fp16)[name = string("obj_247_cast_fp16")];
            string query_71_pad_type_0 = const()[name = string("query_71_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_71_strides_0 = const()[name = string("query_71_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_71_pad_0 = const()[name = string("query_71_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_71_dilations_0 = const()[name = string("query_71_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_71_groups_0 = const()[name = string("query_71_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326114624))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326901120))))[name = string("layers_17_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_17_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_17_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326909376)))];
            tensor<fp16, [1, 1024, 1, 1]> query_71_cast_fp16 = conv(bias = layers_17_encoder_attn_q_proj_bias_to_fp16, dilations = query_71_dilations_0, groups = query_71_groups_0, pad = query_71_pad_0, pad_type = query_71_pad_type_0, strides = query_71_strides_0, weight = layers_17_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_247_cast_fp16)[name = string("query_71_cast_fp16")];
            string key_71_pad_type_0 = const()[name = string("key_71_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_71_strides_0 = const()[name = string("key_71_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_71_pad_0 = const()[name = string("key_71_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_71_dilations_0 = const()[name = string("key_71_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_71_groups_0 = const()[name = string("key_71_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326911488))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(327697984))))[name = string("layers_17_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_71_cast_fp16 = conv(dilations = key_71_dilations_0, groups = key_71_groups_0, pad = key_71_pad_0, pad_type = key_71_pad_type_0, strides = key_71_strides_0, weight = layers_17_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_71_cast_fp16")];
            string value_71_pad_type_0 = const()[name = string("value_71_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_71_strides_0 = const()[name = string("value_71_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_71_pad_0 = const()[name = string("value_71_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_71_dilations_0 = const()[name = string("value_71_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_71_groups_0 = const()[name = string("value_71_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(327706240))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(328492736))))[name = string("layers_17_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_17_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_17_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(328500992)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_71_cast_fp16 = conv(bias = layers_17_encoder_attn_v_proj_bias_to_fp16, dilations = value_71_dilations_0, groups = value_71_groups_0, pad = value_71_pad_0, pad_type = value_71_pad_type_0, strides = value_71_strides_0, weight = layers_17_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_71_cast_fp16")];
            tensor<int32, [4]> var_4030 = const()[name = string("op_4030"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_71_cast_fp16 = reshape(shape = var_4030, x = query_71_cast_fp16)[name = string("mh_q_71_cast_fp16")];
            fp16 var_4032_to_fp16 = const()[name = string("op_4032_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_4033_cast_fp16 = mul(x = mh_q_71_cast_fp16, y = var_4032_to_fp16)[name = string("op_4033_cast_fp16")];
            tensor<int32, [4]> var_4036 = const()[name = string("op_4036"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_4037_cast_fp16 = reshape(shape = var_4036, x = key_71_cast_fp16)[name = string("op_4037_cast_fp16")];
            bool mh_w_107_transpose_x_0 = const()[name = string("mh_w_107_transpose_x_0"), val = bool(true)];
            bool mh_w_107_transpose_y_0 = const()[name = string("mh_w_107_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_107_cast_fp16 = matmul(transpose_x = mh_w_107_transpose_x_0, transpose_y = mh_w_107_transpose_y_0, x = var_4033_cast_fp16, y = var_4037_cast_fp16)[name = string("mh_w_107_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_251_cast_fp16 = softmax(axis = var_3879, x = mh_w_107_cast_fp16)[name = string("obj_251_cast_fp16")];
            tensor<int32, [4]> var_4041 = const()[name = string("op_4041"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_4042_cast_fp16 = reshape(shape = var_4041, x = value_71_cast_fp16)[name = string("op_4042_cast_fp16")];
            bool attn_71_transpose_x_0 = const()[name = string("attn_71_transpose_x_0"), val = bool(false)];
            bool attn_71_transpose_y_0 = const()[name = string("attn_71_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_71_cast_fp16 = matmul(transpose_x = attn_71_transpose_x_0, transpose_y = attn_71_transpose_y_0, x = var_4042_cast_fp16, y = obj_251_cast_fp16)[name = string("attn_71_cast_fp16")];
            tensor<int32, [4]> var_4045 = const()[name = string("op_4045"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_173_cast_fp16 = reshape(shape = var_4045, x = attn_71_cast_fp16)[name = string("input_173_cast_fp16")];
            string obj_249_pad_type_0 = const()[name = string("obj_249_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_249_strides_0 = const()[name = string("obj_249_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_249_pad_0 = const()[name = string("obj_249_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_249_dilations_0 = const()[name = string("obj_249_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_249_groups_0 = const()[name = string("obj_249_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(328503104))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329289600))))[name = string("layers_17_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_17_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_17_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329297856)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_249_cast_fp16 = conv(bias = layers_17_encoder_attn_o_proj_bias_to_fp16, dilations = obj_249_dilations_0, groups = obj_249_groups_0, pad = obj_249_pad_0, pad_type = obj_249_pad_type_0, strides = obj_249_strides_0, weight = layers_17_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = string("obj_249_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_107_cast_fp16 = add(x = inputs_105_cast_fp16, y = obj_249_cast_fp16)[name = string("inputs_107_cast_fp16")];
            tensor<int32, [1]> out_107_axes_0 = const()[name = string("out_107_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_4063_to_fp16 = const()[name = string("op_4063_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_107_cast_fp16 = layer_norm(axes = out_107_axes_0, epsilon = var_4063_to_fp16, x = inputs_107_cast_fp16)[name = string("out_107_cast_fp16")];
            tensor<fp16, [1024]> input_175_gamma_0_to_fp16 = const()[name = string("input_175_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329299968)))];
            tensor<fp16, [1024]> input_175_beta_0_to_fp16 = const()[name = string("input_175_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329302080)))];
            fp16 input_175_epsilon_0_to_fp16 = const()[name = string("input_175_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_175_cast_fp16 = batch_norm(beta = input_175_beta_0_to_fp16, epsilon = input_175_epsilon_0_to_fp16, gamma = input_175_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_107_cast_fp16)[name = string("input_175_cast_fp16")];
            string input_177_pad_type_0 = const()[name = string("input_177_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_177_strides_0 = const()[name = string("input_177_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_177_pad_0 = const()[name = string("input_177_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_177_dilations_0 = const()[name = string("input_177_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_177_groups_0 = const()[name = string("input_177_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_17_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329304192))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332449984))))[name = string("layers_17_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_17_fc1_bias_to_fp16 = const()[name = string("layers_17_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332482816)))];
            tensor<fp16, [1, 4096, 1, 1]> input_177_cast_fp16 = conv(bias = layers_17_fc1_bias_to_fp16, dilations = input_177_dilations_0, groups = input_177_groups_0, pad = input_177_pad_0, pad_type = input_177_pad_type_0, strides = input_177_strides_0, weight = layers_17_fc1_weight_to_fp16_palettized, x = input_175_cast_fp16)[name = string("input_177_cast_fp16")];
            string input_179_mode_0 = const()[name = string("input_179_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_179_cast_fp16 = gelu(mode = input_179_mode_0, x = input_177_cast_fp16)[name = string("input_179_cast_fp16")];
            string hidden_states_37_pad_type_0 = const()[name = string("hidden_states_37_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_37_strides_0 = const()[name = string("hidden_states_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_37_pad_0 = const()[name = string("hidden_states_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_37_dilations_0 = const()[name = string("hidden_states_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_37_groups_0 = const()[name = string("hidden_states_37_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_17_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332491072))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335636864))))[name = string("layers_17_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_17_fc2_bias_to_fp16 = const()[name = string("layers_17_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335645120)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_37_cast_fp16 = conv(bias = layers_17_fc2_bias_to_fp16, dilations = hidden_states_37_dilations_0, groups = hidden_states_37_groups_0, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = hidden_states_37_strides_0, weight = layers_17_fc2_weight_to_fp16_palettized, x = input_179_cast_fp16)[name = string("hidden_states_37_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_109_cast_fp16 = add(x = inputs_107_cast_fp16, y = hidden_states_37_cast_fp16)[name = string("inputs_109_cast_fp16")];
            int32 var_4098 = const()[name = string("op_4098"), val = int32(3)];
            tensor<int32, [1]> out_109_axes_0 = const()[name = string("out_109_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_4123_to_fp16 = const()[name = string("op_4123_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_109_cast_fp16 = layer_norm(axes = out_109_axes_0, epsilon = var_4123_to_fp16, x = inputs_109_cast_fp16)[name = string("out_109_cast_fp16")];
            tensor<fp16, [1024]> obj_253_gamma_0_to_fp16 = const()[name = string("obj_253_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335647232)))];
            tensor<fp16, [1024]> obj_253_beta_0_to_fp16 = const()[name = string("obj_253_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335649344)))];
            fp16 obj_253_epsilon_0_to_fp16 = const()[name = string("obj_253_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_253_cast_fp16 = batch_norm(beta = obj_253_beta_0_to_fp16, epsilon = obj_253_epsilon_0_to_fp16, gamma = obj_253_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_109_cast_fp16)[name = string("obj_253_cast_fp16")];
            string query_73_pad_type_0 = const()[name = string("query_73_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_73_strides_0 = const()[name = string("query_73_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_73_pad_0 = const()[name = string("query_73_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_73_dilations_0 = const()[name = string("query_73_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_73_groups_0 = const()[name = string("query_73_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335651456))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(336437952))))[name = string("layers_18_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(336446208)))];
            tensor<fp16, [1, 1024, 1, 1]> query_73_cast_fp16 = conv(bias = layers_18_self_attn_q_proj_bias_to_fp16, dilations = query_73_dilations_0, groups = query_73_groups_0, pad = query_73_pad_0, pad_type = query_73_pad_type_0, strides = query_73_strides_0, weight = layers_18_self_attn_q_proj_weight_to_fp16_palettized, x = obj_253_cast_fp16)[name = string("query_73_cast_fp16")];
            string current_key_37_pad_type_0 = const()[name = string("current_key_37_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_37_strides_0 = const()[name = string("current_key_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_37_pad_0 = const()[name = string("current_key_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_37_dilations_0 = const()[name = string("current_key_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_37_groups_0 = const()[name = string("current_key_37_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(336448320))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337234816))))[name = string("layers_18_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_37_cast_fp16 = conv(dilations = current_key_37_dilations_0, groups = current_key_37_groups_0, pad = current_key_37_pad_0, pad_type = current_key_37_pad_type_0, strides = current_key_37_strides_0, weight = layers_18_self_attn_k_proj_weight_to_fp16_palettized, x = obj_253_cast_fp16)[name = string("current_key_37_cast_fp16")];
            string current_value_37_pad_type_0 = const()[name = string("current_value_37_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_37_strides_0 = const()[name = string("current_value_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_37_pad_0 = const()[name = string("current_value_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_37_dilations_0 = const()[name = string("current_value_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_37_groups_0 = const()[name = string("current_value_37_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337243072))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338029568))))[name = string("layers_18_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338037824)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_37_cast_fp16 = conv(bias = layers_18_self_attn_v_proj_bias_to_fp16, dilations = current_value_37_dilations_0, groups = current_value_37_groups_0, pad = current_value_37_pad_0, pad_type = current_value_37_pad_type_0, strides = current_value_37_strides_0, weight = layers_18_self_attn_v_proj_weight_to_fp16_palettized, x = obj_253_cast_fp16)[name = string("current_value_37_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4162_cast_fp16 = mul(x = var_87_cast_fp16_18, y = var_207_cast_fp16)[name = string("op_4162_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4163_cast_fp16 = mul(x = current_key_37_cast_fp16, y = var_205_cast_fp16)[name = string("op_4163_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_73_cast_fp16 = add(x = var_4162_cast_fp16, y = var_4163_cast_fp16)[name = string("key_73_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4166_cast_fp16 = mul(x = var_114_cast_fp16_18, y = var_207_cast_fp16)[name = string("op_4166_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4167_cast_fp16 = mul(x = current_value_37_cast_fp16, y = var_205_cast_fp16)[name = string("op_4167_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_73_cast_fp16 = add(x = var_4166_cast_fp16, y = var_4167_cast_fp16)[name = string("value_73_cast_fp16")];
            tensor<int32, [4]> var_4171 = const()[name = string("op_4171"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_73_cast_fp16 = reshape(shape = var_4171, x = query_73_cast_fp16)[name = string("mh_q_73_cast_fp16")];
            fp16 var_4173_to_fp16 = const()[name = string("op_4173_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_4174_cast_fp16 = mul(x = mh_q_73_cast_fp16, y = var_4173_to_fp16)[name = string("op_4174_cast_fp16")];
            tensor<int32, [4]> var_4177 = const()[name = string("op_4177"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_4178_cast_fp16 = reshape(shape = var_4177, x = key_73_cast_fp16)[name = string("op_4178_cast_fp16")];
            bool mh_w_109_transpose_x_0 = const()[name = string("mh_w_109_transpose_x_0"), val = bool(true)];
            bool mh_w_109_transpose_y_0 = const()[name = string("mh_w_109_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_109_cast_fp16 = matmul(transpose_x = mh_w_109_transpose_x_0, transpose_y = mh_w_109_transpose_y_0, x = var_4174_cast_fp16, y = var_4178_cast_fp16)[name = string("mh_w_109_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_111_cast_fp16 = add(x = mh_w_109_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_111_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_4186_cast_fp16 = softmax(axis = var_4098, x = mh_w_111_cast_fp16)[name = string("op_4186_cast_fp16")];
            tensor<int32, [4]> var_4187 = const()[name = string("op_4187"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_4188_cast_fp16 = reshape(shape = var_4187, x = value_73_cast_fp16)[name = string("op_4188_cast_fp16")];
            bool attn_73_transpose_x_0 = const()[name = string("attn_73_transpose_x_0"), val = bool(false)];
            bool attn_73_transpose_y_0 = const()[name = string("attn_73_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_73_cast_fp16 = matmul(transpose_x = attn_73_transpose_x_0, transpose_y = attn_73_transpose_y_0, x = var_4188_cast_fp16, y = var_4186_cast_fp16)[name = string("attn_73_cast_fp16")];
            tensor<int32, [4]> var_4191 = const()[name = string("op_4191"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_181_cast_fp16 = reshape(shape = var_4191, x = attn_73_cast_fp16)[name = string("input_181_cast_fp16")];
            string obj_259_pad_type_0 = const()[name = string("obj_259_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_259_strides_0 = const()[name = string("obj_259_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_259_pad_0 = const()[name = string("obj_259_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_259_dilations_0 = const()[name = string("obj_259_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_259_groups_0 = const()[name = string("obj_259_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338039936))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338826432))))[name = string("layers_18_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_18_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338834688)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_259_cast_fp16 = conv(bias = layers_18_self_attn_o_proj_bias_to_fp16, dilations = obj_259_dilations_0, groups = obj_259_groups_0, pad = obj_259_pad_0, pad_type = obj_259_pad_type_0, strides = obj_259_strides_0, weight = layers_18_self_attn_o_proj_weight_to_fp16_palettized, x = input_181_cast_fp16)[name = string("obj_259_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_111_cast_fp16 = add(x = inputs_109_cast_fp16, y = obj_259_cast_fp16)[name = string("inputs_111_cast_fp16")];
            tensor<int32, [1]> out_111_axes_0 = const()[name = string("out_111_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_4213_to_fp16 = const()[name = string("op_4213_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_111_cast_fp16 = layer_norm(axes = out_111_axes_0, epsilon = var_4213_to_fp16, x = inputs_111_cast_fp16)[name = string("out_111_cast_fp16")];
            tensor<fp16, [1024]> obj_261_gamma_0_to_fp16 = const()[name = string("obj_261_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338836800)))];
            tensor<fp16, [1024]> obj_261_beta_0_to_fp16 = const()[name = string("obj_261_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338838912)))];
            fp16 obj_261_epsilon_0_to_fp16 = const()[name = string("obj_261_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_261_cast_fp16 = batch_norm(beta = obj_261_beta_0_to_fp16, epsilon = obj_261_epsilon_0_to_fp16, gamma = obj_261_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_111_cast_fp16)[name = string("obj_261_cast_fp16")];
            string query_75_pad_type_0 = const()[name = string("query_75_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_75_strides_0 = const()[name = string("query_75_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_75_pad_0 = const()[name = string("query_75_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_75_dilations_0 = const()[name = string("query_75_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_75_groups_0 = const()[name = string("query_75_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338841024))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339627520))))[name = string("layers_18_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_18_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_18_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339635776)))];
            tensor<fp16, [1, 1024, 1, 1]> query_75_cast_fp16 = conv(bias = layers_18_encoder_attn_q_proj_bias_to_fp16, dilations = query_75_dilations_0, groups = query_75_groups_0, pad = query_75_pad_0, pad_type = query_75_pad_type_0, strides = query_75_strides_0, weight = layers_18_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_261_cast_fp16)[name = string("query_75_cast_fp16")];
            string key_75_pad_type_0 = const()[name = string("key_75_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_75_strides_0 = const()[name = string("key_75_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_75_pad_0 = const()[name = string("key_75_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_75_dilations_0 = const()[name = string("key_75_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_75_groups_0 = const()[name = string("key_75_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339637888))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(340424384))))[name = string("layers_18_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_75_cast_fp16 = conv(dilations = key_75_dilations_0, groups = key_75_groups_0, pad = key_75_pad_0, pad_type = key_75_pad_type_0, strides = key_75_strides_0, weight = layers_18_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_75_cast_fp16")];
            string value_75_pad_type_0 = const()[name = string("value_75_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_75_strides_0 = const()[name = string("value_75_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_75_pad_0 = const()[name = string("value_75_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_75_dilations_0 = const()[name = string("value_75_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_75_groups_0 = const()[name = string("value_75_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(340432640))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341219136))))[name = string("layers_18_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_18_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_18_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341227392)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_75_cast_fp16 = conv(bias = layers_18_encoder_attn_v_proj_bias_to_fp16, dilations = value_75_dilations_0, groups = value_75_groups_0, pad = value_75_pad_0, pad_type = value_75_pad_type_0, strides = value_75_strides_0, weight = layers_18_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_75_cast_fp16")];
            tensor<int32, [4]> var_4249 = const()[name = string("op_4249"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_75_cast_fp16 = reshape(shape = var_4249, x = query_75_cast_fp16)[name = string("mh_q_75_cast_fp16")];
            fp16 var_4251_to_fp16 = const()[name = string("op_4251_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_4252_cast_fp16 = mul(x = mh_q_75_cast_fp16, y = var_4251_to_fp16)[name = string("op_4252_cast_fp16")];
            tensor<int32, [4]> var_4255 = const()[name = string("op_4255"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_4256_cast_fp16 = reshape(shape = var_4255, x = key_75_cast_fp16)[name = string("op_4256_cast_fp16")];
            bool mh_w_113_transpose_x_0 = const()[name = string("mh_w_113_transpose_x_0"), val = bool(true)];
            bool mh_w_113_transpose_y_0 = const()[name = string("mh_w_113_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_113_cast_fp16 = matmul(transpose_x = mh_w_113_transpose_x_0, transpose_y = mh_w_113_transpose_y_0, x = var_4252_cast_fp16, y = var_4256_cast_fp16)[name = string("mh_w_113_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_265_cast_fp16 = softmax(axis = var_4098, x = mh_w_113_cast_fp16)[name = string("obj_265_cast_fp16")];
            tensor<int32, [4]> var_4260 = const()[name = string("op_4260"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_4261_cast_fp16 = reshape(shape = var_4260, x = value_75_cast_fp16)[name = string("op_4261_cast_fp16")];
            bool attn_75_transpose_x_0 = const()[name = string("attn_75_transpose_x_0"), val = bool(false)];
            bool attn_75_transpose_y_0 = const()[name = string("attn_75_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_75_cast_fp16 = matmul(transpose_x = attn_75_transpose_x_0, transpose_y = attn_75_transpose_y_0, x = var_4261_cast_fp16, y = obj_265_cast_fp16)[name = string("attn_75_cast_fp16")];
            tensor<int32, [4]> var_4264 = const()[name = string("op_4264"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_183_cast_fp16 = reshape(shape = var_4264, x = attn_75_cast_fp16)[name = string("input_183_cast_fp16")];
            string obj_263_pad_type_0 = const()[name = string("obj_263_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_263_strides_0 = const()[name = string("obj_263_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_263_pad_0 = const()[name = string("obj_263_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_263_dilations_0 = const()[name = string("obj_263_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_263_groups_0 = const()[name = string("obj_263_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341229504))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342016000))))[name = string("layers_18_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_18_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_18_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342024256)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_263_cast_fp16 = conv(bias = layers_18_encoder_attn_o_proj_bias_to_fp16, dilations = obj_263_dilations_0, groups = obj_263_groups_0, pad = obj_263_pad_0, pad_type = obj_263_pad_type_0, strides = obj_263_strides_0, weight = layers_18_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_183_cast_fp16)[name = string("obj_263_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_113_cast_fp16 = add(x = inputs_111_cast_fp16, y = obj_263_cast_fp16)[name = string("inputs_113_cast_fp16")];
            tensor<int32, [1]> out_113_axes_0 = const()[name = string("out_113_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_4282_to_fp16 = const()[name = string("op_4282_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_113_cast_fp16 = layer_norm(axes = out_113_axes_0, epsilon = var_4282_to_fp16, x = inputs_113_cast_fp16)[name = string("out_113_cast_fp16")];
            tensor<fp16, [1024]> input_185_gamma_0_to_fp16 = const()[name = string("input_185_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342026368)))];
            tensor<fp16, [1024]> input_185_beta_0_to_fp16 = const()[name = string("input_185_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342028480)))];
            fp16 input_185_epsilon_0_to_fp16 = const()[name = string("input_185_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_185_cast_fp16 = batch_norm(beta = input_185_beta_0_to_fp16, epsilon = input_185_epsilon_0_to_fp16, gamma = input_185_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_113_cast_fp16)[name = string("input_185_cast_fp16")];
            string input_187_pad_type_0 = const()[name = string("input_187_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_187_strides_0 = const()[name = string("input_187_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_187_pad_0 = const()[name = string("input_187_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_187_dilations_0 = const()[name = string("input_187_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_187_groups_0 = const()[name = string("input_187_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_18_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342030592))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345176384))))[name = string("layers_18_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_18_fc1_bias_to_fp16 = const()[name = string("layers_18_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345209216)))];
            tensor<fp16, [1, 4096, 1, 1]> input_187_cast_fp16 = conv(bias = layers_18_fc1_bias_to_fp16, dilations = input_187_dilations_0, groups = input_187_groups_0, pad = input_187_pad_0, pad_type = input_187_pad_type_0, strides = input_187_strides_0, weight = layers_18_fc1_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = string("input_187_cast_fp16")];
            string input_189_mode_0 = const()[name = string("input_189_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_189_cast_fp16 = gelu(mode = input_189_mode_0, x = input_187_cast_fp16)[name = string("input_189_cast_fp16")];
            string hidden_states_39_pad_type_0 = const()[name = string("hidden_states_39_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_39_strides_0 = const()[name = string("hidden_states_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_39_pad_0 = const()[name = string("hidden_states_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_39_dilations_0 = const()[name = string("hidden_states_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_39_groups_0 = const()[name = string("hidden_states_39_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_18_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345217472))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(348363264))))[name = string("layers_18_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_18_fc2_bias_to_fp16 = const()[name = string("layers_18_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(348371520)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_39_cast_fp16 = conv(bias = layers_18_fc2_bias_to_fp16, dilations = hidden_states_39_dilations_0, groups = hidden_states_39_groups_0, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = hidden_states_39_strides_0, weight = layers_18_fc2_weight_to_fp16_palettized, x = input_189_cast_fp16)[name = string("hidden_states_39_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_115_cast_fp16 = add(x = inputs_113_cast_fp16, y = hidden_states_39_cast_fp16)[name = string("inputs_115_cast_fp16")];
            int32 var_4317 = const()[name = string("op_4317"), val = int32(3)];
            tensor<int32, [1]> out_115_axes_0 = const()[name = string("out_115_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_4342_to_fp16 = const()[name = string("op_4342_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_115_cast_fp16 = layer_norm(axes = out_115_axes_0, epsilon = var_4342_to_fp16, x = inputs_115_cast_fp16)[name = string("out_115_cast_fp16")];
            tensor<fp16, [1024]> obj_267_gamma_0_to_fp16 = const()[name = string("obj_267_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(348373632)))];
            tensor<fp16, [1024]> obj_267_beta_0_to_fp16 = const()[name = string("obj_267_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(348375744)))];
            fp16 obj_267_epsilon_0_to_fp16 = const()[name = string("obj_267_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_267_cast_fp16 = batch_norm(beta = obj_267_beta_0_to_fp16, epsilon = obj_267_epsilon_0_to_fp16, gamma = obj_267_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_115_cast_fp16)[name = string("obj_267_cast_fp16")];
            string query_77_pad_type_0 = const()[name = string("query_77_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_77_strides_0 = const()[name = string("query_77_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_77_pad_0 = const()[name = string("query_77_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_77_dilations_0 = const()[name = string("query_77_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_77_groups_0 = const()[name = string("query_77_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(348377856))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(349164352))))[name = string("layers_19_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(349172608)))];
            tensor<fp16, [1, 1024, 1, 1]> query_77_cast_fp16 = conv(bias = layers_19_self_attn_q_proj_bias_to_fp16, dilations = query_77_dilations_0, groups = query_77_groups_0, pad = query_77_pad_0, pad_type = query_77_pad_type_0, strides = query_77_strides_0, weight = layers_19_self_attn_q_proj_weight_to_fp16_palettized, x = obj_267_cast_fp16)[name = string("query_77_cast_fp16")];
            string current_key_39_pad_type_0 = const()[name = string("current_key_39_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_39_strides_0 = const()[name = string("current_key_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_39_pad_0 = const()[name = string("current_key_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_39_dilations_0 = const()[name = string("current_key_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_39_groups_0 = const()[name = string("current_key_39_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(349174720))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(349961216))))[name = string("layers_19_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_39_cast_fp16 = conv(dilations = current_key_39_dilations_0, groups = current_key_39_groups_0, pad = current_key_39_pad_0, pad_type = current_key_39_pad_type_0, strides = current_key_39_strides_0, weight = layers_19_self_attn_k_proj_weight_to_fp16_palettized, x = obj_267_cast_fp16)[name = string("current_key_39_cast_fp16")];
            string current_value_39_pad_type_0 = const()[name = string("current_value_39_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_39_strides_0 = const()[name = string("current_value_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_39_pad_0 = const()[name = string("current_value_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_39_dilations_0 = const()[name = string("current_value_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_39_groups_0 = const()[name = string("current_value_39_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(349969472))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350755968))))[name = string("layers_19_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350764224)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_39_cast_fp16 = conv(bias = layers_19_self_attn_v_proj_bias_to_fp16, dilations = current_value_39_dilations_0, groups = current_value_39_groups_0, pad = current_value_39_pad_0, pad_type = current_value_39_pad_type_0, strides = current_value_39_strides_0, weight = layers_19_self_attn_v_proj_weight_to_fp16_palettized, x = obj_267_cast_fp16)[name = string("current_value_39_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4381_cast_fp16 = mul(x = var_87_cast_fp16_19, y = var_207_cast_fp16)[name = string("op_4381_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4382_cast_fp16 = mul(x = current_key_39_cast_fp16, y = var_205_cast_fp16)[name = string("op_4382_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_77_cast_fp16 = add(x = var_4381_cast_fp16, y = var_4382_cast_fp16)[name = string("key_77_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4385_cast_fp16 = mul(x = var_114_cast_fp16_19, y = var_207_cast_fp16)[name = string("op_4385_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4386_cast_fp16 = mul(x = current_value_39_cast_fp16, y = var_205_cast_fp16)[name = string("op_4386_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_77_cast_fp16 = add(x = var_4385_cast_fp16, y = var_4386_cast_fp16)[name = string("value_77_cast_fp16")];
            tensor<int32, [4]> var_4390 = const()[name = string("op_4390"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_77_cast_fp16 = reshape(shape = var_4390, x = query_77_cast_fp16)[name = string("mh_q_77_cast_fp16")];
            fp16 var_4392_to_fp16 = const()[name = string("op_4392_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_4393_cast_fp16 = mul(x = mh_q_77_cast_fp16, y = var_4392_to_fp16)[name = string("op_4393_cast_fp16")];
            tensor<int32, [4]> var_4396 = const()[name = string("op_4396"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_4397_cast_fp16 = reshape(shape = var_4396, x = key_77_cast_fp16)[name = string("op_4397_cast_fp16")];
            bool mh_w_115_transpose_x_0 = const()[name = string("mh_w_115_transpose_x_0"), val = bool(true)];
            bool mh_w_115_transpose_y_0 = const()[name = string("mh_w_115_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_115_cast_fp16 = matmul(transpose_x = mh_w_115_transpose_x_0, transpose_y = mh_w_115_transpose_y_0, x = var_4393_cast_fp16, y = var_4397_cast_fp16)[name = string("mh_w_115_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_117_cast_fp16 = add(x = mh_w_115_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_117_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_4405_cast_fp16 = softmax(axis = var_4317, x = mh_w_117_cast_fp16)[name = string("op_4405_cast_fp16")];
            tensor<int32, [4]> var_4406 = const()[name = string("op_4406"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_4407_cast_fp16 = reshape(shape = var_4406, x = value_77_cast_fp16)[name = string("op_4407_cast_fp16")];
            bool attn_77_transpose_x_0 = const()[name = string("attn_77_transpose_x_0"), val = bool(false)];
            bool attn_77_transpose_y_0 = const()[name = string("attn_77_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_77_cast_fp16 = matmul(transpose_x = attn_77_transpose_x_0, transpose_y = attn_77_transpose_y_0, x = var_4407_cast_fp16, y = var_4405_cast_fp16)[name = string("attn_77_cast_fp16")];
            tensor<int32, [4]> var_4410 = const()[name = string("op_4410"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_191_cast_fp16 = reshape(shape = var_4410, x = attn_77_cast_fp16)[name = string("input_191_cast_fp16")];
            string obj_273_pad_type_0 = const()[name = string("obj_273_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_273_strides_0 = const()[name = string("obj_273_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_273_pad_0 = const()[name = string("obj_273_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_273_dilations_0 = const()[name = string("obj_273_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_273_groups_0 = const()[name = string("obj_273_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350766336))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(351552832))))[name = string("layers_19_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_19_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(351561088)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_273_cast_fp16 = conv(bias = layers_19_self_attn_o_proj_bias_to_fp16, dilations = obj_273_dilations_0, groups = obj_273_groups_0, pad = obj_273_pad_0, pad_type = obj_273_pad_type_0, strides = obj_273_strides_0, weight = layers_19_self_attn_o_proj_weight_to_fp16_palettized, x = input_191_cast_fp16)[name = string("obj_273_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_117_cast_fp16 = add(x = inputs_115_cast_fp16, y = obj_273_cast_fp16)[name = string("inputs_117_cast_fp16")];
            tensor<int32, [1]> out_117_axes_0 = const()[name = string("out_117_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_4432_to_fp16 = const()[name = string("op_4432_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_117_cast_fp16 = layer_norm(axes = out_117_axes_0, epsilon = var_4432_to_fp16, x = inputs_117_cast_fp16)[name = string("out_117_cast_fp16")];
            tensor<fp16, [1024]> obj_275_gamma_0_to_fp16 = const()[name = string("obj_275_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(351563200)))];
            tensor<fp16, [1024]> obj_275_beta_0_to_fp16 = const()[name = string("obj_275_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(351565312)))];
            fp16 obj_275_epsilon_0_to_fp16 = const()[name = string("obj_275_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_275_cast_fp16 = batch_norm(beta = obj_275_beta_0_to_fp16, epsilon = obj_275_epsilon_0_to_fp16, gamma = obj_275_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_117_cast_fp16)[name = string("obj_275_cast_fp16")];
            string query_79_pad_type_0 = const()[name = string("query_79_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_79_strides_0 = const()[name = string("query_79_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_79_pad_0 = const()[name = string("query_79_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_79_dilations_0 = const()[name = string("query_79_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_79_groups_0 = const()[name = string("query_79_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(351567424))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(352353920))))[name = string("layers_19_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_19_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_19_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(352362176)))];
            tensor<fp16, [1, 1024, 1, 1]> query_79_cast_fp16 = conv(bias = layers_19_encoder_attn_q_proj_bias_to_fp16, dilations = query_79_dilations_0, groups = query_79_groups_0, pad = query_79_pad_0, pad_type = query_79_pad_type_0, strides = query_79_strides_0, weight = layers_19_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_275_cast_fp16)[name = string("query_79_cast_fp16")];
            string key_79_pad_type_0 = const()[name = string("key_79_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_79_strides_0 = const()[name = string("key_79_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_79_pad_0 = const()[name = string("key_79_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_79_dilations_0 = const()[name = string("key_79_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_79_groups_0 = const()[name = string("key_79_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(352364288))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353150784))))[name = string("layers_19_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_79_cast_fp16 = conv(dilations = key_79_dilations_0, groups = key_79_groups_0, pad = key_79_pad_0, pad_type = key_79_pad_type_0, strides = key_79_strides_0, weight = layers_19_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_79_cast_fp16")];
            string value_79_pad_type_0 = const()[name = string("value_79_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_79_strides_0 = const()[name = string("value_79_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_79_pad_0 = const()[name = string("value_79_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_79_dilations_0 = const()[name = string("value_79_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_79_groups_0 = const()[name = string("value_79_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353159040))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353945536))))[name = string("layers_19_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_19_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_19_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353953792)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_79_cast_fp16 = conv(bias = layers_19_encoder_attn_v_proj_bias_to_fp16, dilations = value_79_dilations_0, groups = value_79_groups_0, pad = value_79_pad_0, pad_type = value_79_pad_type_0, strides = value_79_strides_0, weight = layers_19_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_79_cast_fp16")];
            tensor<int32, [4]> var_4468 = const()[name = string("op_4468"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_79_cast_fp16 = reshape(shape = var_4468, x = query_79_cast_fp16)[name = string("mh_q_79_cast_fp16")];
            fp16 var_4470_to_fp16 = const()[name = string("op_4470_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_4471_cast_fp16 = mul(x = mh_q_79_cast_fp16, y = var_4470_to_fp16)[name = string("op_4471_cast_fp16")];
            tensor<int32, [4]> var_4474 = const()[name = string("op_4474"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_4475_cast_fp16 = reshape(shape = var_4474, x = key_79_cast_fp16)[name = string("op_4475_cast_fp16")];
            bool mh_w_119_transpose_x_0 = const()[name = string("mh_w_119_transpose_x_0"), val = bool(true)];
            bool mh_w_119_transpose_y_0 = const()[name = string("mh_w_119_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_119_cast_fp16 = matmul(transpose_x = mh_w_119_transpose_x_0, transpose_y = mh_w_119_transpose_y_0, x = var_4471_cast_fp16, y = var_4475_cast_fp16)[name = string("mh_w_119_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_279_cast_fp16 = softmax(axis = var_4317, x = mh_w_119_cast_fp16)[name = string("obj_279_cast_fp16")];
            tensor<int32, [4]> var_4479 = const()[name = string("op_4479"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_4480_cast_fp16 = reshape(shape = var_4479, x = value_79_cast_fp16)[name = string("op_4480_cast_fp16")];
            bool attn_79_transpose_x_0 = const()[name = string("attn_79_transpose_x_0"), val = bool(false)];
            bool attn_79_transpose_y_0 = const()[name = string("attn_79_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_79_cast_fp16 = matmul(transpose_x = attn_79_transpose_x_0, transpose_y = attn_79_transpose_y_0, x = var_4480_cast_fp16, y = obj_279_cast_fp16)[name = string("attn_79_cast_fp16")];
            tensor<int32, [4]> var_4483 = const()[name = string("op_4483"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_193_cast_fp16 = reshape(shape = var_4483, x = attn_79_cast_fp16)[name = string("input_193_cast_fp16")];
            string obj_277_pad_type_0 = const()[name = string("obj_277_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_277_strides_0 = const()[name = string("obj_277_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_277_pad_0 = const()[name = string("obj_277_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_277_dilations_0 = const()[name = string("obj_277_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_277_groups_0 = const()[name = string("obj_277_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353955904))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354742400))))[name = string("layers_19_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_19_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_19_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354750656)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_277_cast_fp16 = conv(bias = layers_19_encoder_attn_o_proj_bias_to_fp16, dilations = obj_277_dilations_0, groups = obj_277_groups_0, pad = obj_277_pad_0, pad_type = obj_277_pad_type_0, strides = obj_277_strides_0, weight = layers_19_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = string("obj_277_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_119_cast_fp16 = add(x = inputs_117_cast_fp16, y = obj_277_cast_fp16)[name = string("inputs_119_cast_fp16")];
            tensor<int32, [1]> out_119_axes_0 = const()[name = string("out_119_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_4501_to_fp16 = const()[name = string("op_4501_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_119_cast_fp16 = layer_norm(axes = out_119_axes_0, epsilon = var_4501_to_fp16, x = inputs_119_cast_fp16)[name = string("out_119_cast_fp16")];
            tensor<fp16, [1024]> input_195_gamma_0_to_fp16 = const()[name = string("input_195_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354752768)))];
            tensor<fp16, [1024]> input_195_beta_0_to_fp16 = const()[name = string("input_195_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354754880)))];
            fp16 input_195_epsilon_0_to_fp16 = const()[name = string("input_195_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_195_cast_fp16 = batch_norm(beta = input_195_beta_0_to_fp16, epsilon = input_195_epsilon_0_to_fp16, gamma = input_195_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_119_cast_fp16)[name = string("input_195_cast_fp16")];
            string input_197_pad_type_0 = const()[name = string("input_197_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_197_strides_0 = const()[name = string("input_197_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_197_pad_0 = const()[name = string("input_197_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_197_dilations_0 = const()[name = string("input_197_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_197_groups_0 = const()[name = string("input_197_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_19_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354756992))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357902784))))[name = string("layers_19_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_19_fc1_bias_to_fp16 = const()[name = string("layers_19_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357935616)))];
            tensor<fp16, [1, 4096, 1, 1]> input_197_cast_fp16 = conv(bias = layers_19_fc1_bias_to_fp16, dilations = input_197_dilations_0, groups = input_197_groups_0, pad = input_197_pad_0, pad_type = input_197_pad_type_0, strides = input_197_strides_0, weight = layers_19_fc1_weight_to_fp16_palettized, x = input_195_cast_fp16)[name = string("input_197_cast_fp16")];
            string input_199_mode_0 = const()[name = string("input_199_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_199_cast_fp16 = gelu(mode = input_199_mode_0, x = input_197_cast_fp16)[name = string("input_199_cast_fp16")];
            string hidden_states_41_pad_type_0 = const()[name = string("hidden_states_41_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_41_strides_0 = const()[name = string("hidden_states_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_41_pad_0 = const()[name = string("hidden_states_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_41_dilations_0 = const()[name = string("hidden_states_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_41_groups_0 = const()[name = string("hidden_states_41_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_19_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357943872))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361089664))))[name = string("layers_19_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_19_fc2_bias_to_fp16 = const()[name = string("layers_19_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361097920)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_41_cast_fp16 = conv(bias = layers_19_fc2_bias_to_fp16, dilations = hidden_states_41_dilations_0, groups = hidden_states_41_groups_0, pad = hidden_states_41_pad_0, pad_type = hidden_states_41_pad_type_0, strides = hidden_states_41_strides_0, weight = layers_19_fc2_weight_to_fp16_palettized, x = input_199_cast_fp16)[name = string("hidden_states_41_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_121_cast_fp16 = add(x = inputs_119_cast_fp16, y = hidden_states_41_cast_fp16)[name = string("inputs_121_cast_fp16")];
            int32 var_4536 = const()[name = string("op_4536"), val = int32(3)];
            tensor<int32, [1]> out_121_axes_0 = const()[name = string("out_121_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_4561_to_fp16 = const()[name = string("op_4561_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_121_cast_fp16 = layer_norm(axes = out_121_axes_0, epsilon = var_4561_to_fp16, x = inputs_121_cast_fp16)[name = string("out_121_cast_fp16")];
            tensor<fp16, [1024]> obj_281_gamma_0_to_fp16 = const()[name = string("obj_281_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361100032)))];
            tensor<fp16, [1024]> obj_281_beta_0_to_fp16 = const()[name = string("obj_281_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361102144)))];
            fp16 obj_281_epsilon_0_to_fp16 = const()[name = string("obj_281_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_281_cast_fp16 = batch_norm(beta = obj_281_beta_0_to_fp16, epsilon = obj_281_epsilon_0_to_fp16, gamma = obj_281_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_121_cast_fp16)[name = string("obj_281_cast_fp16")];
            string query_81_pad_type_0 = const()[name = string("query_81_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_81_strides_0 = const()[name = string("query_81_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_81_pad_0 = const()[name = string("query_81_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_81_dilations_0 = const()[name = string("query_81_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_81_groups_0 = const()[name = string("query_81_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361104256))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361890752))))[name = string("layers_20_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361899008)))];
            tensor<fp16, [1, 1024, 1, 1]> query_81_cast_fp16 = conv(bias = layers_20_self_attn_q_proj_bias_to_fp16, dilations = query_81_dilations_0, groups = query_81_groups_0, pad = query_81_pad_0, pad_type = query_81_pad_type_0, strides = query_81_strides_0, weight = layers_20_self_attn_q_proj_weight_to_fp16_palettized, x = obj_281_cast_fp16)[name = string("query_81_cast_fp16")];
            string current_key_41_pad_type_0 = const()[name = string("current_key_41_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_41_strides_0 = const()[name = string("current_key_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_41_pad_0 = const()[name = string("current_key_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_41_dilations_0 = const()[name = string("current_key_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_41_groups_0 = const()[name = string("current_key_41_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361901120))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362687616))))[name = string("layers_20_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_41_cast_fp16 = conv(dilations = current_key_41_dilations_0, groups = current_key_41_groups_0, pad = current_key_41_pad_0, pad_type = current_key_41_pad_type_0, strides = current_key_41_strides_0, weight = layers_20_self_attn_k_proj_weight_to_fp16_palettized, x = obj_281_cast_fp16)[name = string("current_key_41_cast_fp16")];
            string current_value_41_pad_type_0 = const()[name = string("current_value_41_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_41_strides_0 = const()[name = string("current_value_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_41_pad_0 = const()[name = string("current_value_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_41_dilations_0 = const()[name = string("current_value_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_41_groups_0 = const()[name = string("current_value_41_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362695872))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(363482368))))[name = string("layers_20_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(363490624)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_41_cast_fp16 = conv(bias = layers_20_self_attn_v_proj_bias_to_fp16, dilations = current_value_41_dilations_0, groups = current_value_41_groups_0, pad = current_value_41_pad_0, pad_type = current_value_41_pad_type_0, strides = current_value_41_strides_0, weight = layers_20_self_attn_v_proj_weight_to_fp16_palettized, x = obj_281_cast_fp16)[name = string("current_value_41_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4600_cast_fp16 = mul(x = var_87_cast_fp16_20, y = var_207_cast_fp16)[name = string("op_4600_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4601_cast_fp16 = mul(x = current_key_41_cast_fp16, y = var_205_cast_fp16)[name = string("op_4601_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_81_cast_fp16 = add(x = var_4600_cast_fp16, y = var_4601_cast_fp16)[name = string("key_81_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4604_cast_fp16 = mul(x = var_114_cast_fp16_20, y = var_207_cast_fp16)[name = string("op_4604_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4605_cast_fp16 = mul(x = current_value_41_cast_fp16, y = var_205_cast_fp16)[name = string("op_4605_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_81_cast_fp16 = add(x = var_4604_cast_fp16, y = var_4605_cast_fp16)[name = string("value_81_cast_fp16")];
            tensor<int32, [4]> var_4609 = const()[name = string("op_4609"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_81_cast_fp16 = reshape(shape = var_4609, x = query_81_cast_fp16)[name = string("mh_q_81_cast_fp16")];
            fp16 var_4611_to_fp16 = const()[name = string("op_4611_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_4612_cast_fp16 = mul(x = mh_q_81_cast_fp16, y = var_4611_to_fp16)[name = string("op_4612_cast_fp16")];
            tensor<int32, [4]> var_4615 = const()[name = string("op_4615"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_4616_cast_fp16 = reshape(shape = var_4615, x = key_81_cast_fp16)[name = string("op_4616_cast_fp16")];
            bool mh_w_121_transpose_x_0 = const()[name = string("mh_w_121_transpose_x_0"), val = bool(true)];
            bool mh_w_121_transpose_y_0 = const()[name = string("mh_w_121_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_121_cast_fp16 = matmul(transpose_x = mh_w_121_transpose_x_0, transpose_y = mh_w_121_transpose_y_0, x = var_4612_cast_fp16, y = var_4616_cast_fp16)[name = string("mh_w_121_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_123_cast_fp16 = add(x = mh_w_121_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_123_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_4624_cast_fp16 = softmax(axis = var_4536, x = mh_w_123_cast_fp16)[name = string("op_4624_cast_fp16")];
            tensor<int32, [4]> var_4625 = const()[name = string("op_4625"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_4626_cast_fp16 = reshape(shape = var_4625, x = value_81_cast_fp16)[name = string("op_4626_cast_fp16")];
            bool attn_81_transpose_x_0 = const()[name = string("attn_81_transpose_x_0"), val = bool(false)];
            bool attn_81_transpose_y_0 = const()[name = string("attn_81_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_81_cast_fp16 = matmul(transpose_x = attn_81_transpose_x_0, transpose_y = attn_81_transpose_y_0, x = var_4626_cast_fp16, y = var_4624_cast_fp16)[name = string("attn_81_cast_fp16")];
            tensor<int32, [4]> var_4629 = const()[name = string("op_4629"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_201_cast_fp16 = reshape(shape = var_4629, x = attn_81_cast_fp16)[name = string("input_201_cast_fp16")];
            string obj_287_pad_type_0 = const()[name = string("obj_287_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_287_strides_0 = const()[name = string("obj_287_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_287_pad_0 = const()[name = string("obj_287_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_287_dilations_0 = const()[name = string("obj_287_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_287_groups_0 = const()[name = string("obj_287_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(363492736))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364279232))))[name = string("layers_20_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_20_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364287488)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_287_cast_fp16 = conv(bias = layers_20_self_attn_o_proj_bias_to_fp16, dilations = obj_287_dilations_0, groups = obj_287_groups_0, pad = obj_287_pad_0, pad_type = obj_287_pad_type_0, strides = obj_287_strides_0, weight = layers_20_self_attn_o_proj_weight_to_fp16_palettized, x = input_201_cast_fp16)[name = string("obj_287_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_123_cast_fp16 = add(x = inputs_121_cast_fp16, y = obj_287_cast_fp16)[name = string("inputs_123_cast_fp16")];
            tensor<int32, [1]> out_123_axes_0 = const()[name = string("out_123_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_4651_to_fp16 = const()[name = string("op_4651_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_123_cast_fp16 = layer_norm(axes = out_123_axes_0, epsilon = var_4651_to_fp16, x = inputs_123_cast_fp16)[name = string("out_123_cast_fp16")];
            tensor<fp16, [1024]> obj_289_gamma_0_to_fp16 = const()[name = string("obj_289_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364289600)))];
            tensor<fp16, [1024]> obj_289_beta_0_to_fp16 = const()[name = string("obj_289_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364291712)))];
            fp16 obj_289_epsilon_0_to_fp16 = const()[name = string("obj_289_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_289_cast_fp16 = batch_norm(beta = obj_289_beta_0_to_fp16, epsilon = obj_289_epsilon_0_to_fp16, gamma = obj_289_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_123_cast_fp16)[name = string("obj_289_cast_fp16")];
            string query_83_pad_type_0 = const()[name = string("query_83_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_83_strides_0 = const()[name = string("query_83_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_83_pad_0 = const()[name = string("query_83_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_83_dilations_0 = const()[name = string("query_83_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_83_groups_0 = const()[name = string("query_83_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364293824))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365080320))))[name = string("layers_20_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_20_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_20_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365088576)))];
            tensor<fp16, [1, 1024, 1, 1]> query_83_cast_fp16 = conv(bias = layers_20_encoder_attn_q_proj_bias_to_fp16, dilations = query_83_dilations_0, groups = query_83_groups_0, pad = query_83_pad_0, pad_type = query_83_pad_type_0, strides = query_83_strides_0, weight = layers_20_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_289_cast_fp16)[name = string("query_83_cast_fp16")];
            string key_83_pad_type_0 = const()[name = string("key_83_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_83_strides_0 = const()[name = string("key_83_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_83_pad_0 = const()[name = string("key_83_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_83_dilations_0 = const()[name = string("key_83_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_83_groups_0 = const()[name = string("key_83_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365090688))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365877184))))[name = string("layers_20_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_83_cast_fp16 = conv(dilations = key_83_dilations_0, groups = key_83_groups_0, pad = key_83_pad_0, pad_type = key_83_pad_type_0, strides = key_83_strides_0, weight = layers_20_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_83_cast_fp16")];
            string value_83_pad_type_0 = const()[name = string("value_83_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_83_strides_0 = const()[name = string("value_83_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_83_pad_0 = const()[name = string("value_83_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_83_dilations_0 = const()[name = string("value_83_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_83_groups_0 = const()[name = string("value_83_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365885440))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(366671936))))[name = string("layers_20_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_20_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_20_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(366680192)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_83_cast_fp16 = conv(bias = layers_20_encoder_attn_v_proj_bias_to_fp16, dilations = value_83_dilations_0, groups = value_83_groups_0, pad = value_83_pad_0, pad_type = value_83_pad_type_0, strides = value_83_strides_0, weight = layers_20_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_83_cast_fp16")];
            tensor<int32, [4]> var_4687 = const()[name = string("op_4687"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_83_cast_fp16 = reshape(shape = var_4687, x = query_83_cast_fp16)[name = string("mh_q_83_cast_fp16")];
            fp16 var_4689_to_fp16 = const()[name = string("op_4689_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_4690_cast_fp16 = mul(x = mh_q_83_cast_fp16, y = var_4689_to_fp16)[name = string("op_4690_cast_fp16")];
            tensor<int32, [4]> var_4693 = const()[name = string("op_4693"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_4694_cast_fp16 = reshape(shape = var_4693, x = key_83_cast_fp16)[name = string("op_4694_cast_fp16")];
            bool mh_w_125_transpose_x_0 = const()[name = string("mh_w_125_transpose_x_0"), val = bool(true)];
            bool mh_w_125_transpose_y_0 = const()[name = string("mh_w_125_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_125_cast_fp16 = matmul(transpose_x = mh_w_125_transpose_x_0, transpose_y = mh_w_125_transpose_y_0, x = var_4690_cast_fp16, y = var_4694_cast_fp16)[name = string("mh_w_125_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_293_cast_fp16 = softmax(axis = var_4536, x = mh_w_125_cast_fp16)[name = string("obj_293_cast_fp16")];
            tensor<int32, [4]> var_4698 = const()[name = string("op_4698"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_4699_cast_fp16 = reshape(shape = var_4698, x = value_83_cast_fp16)[name = string("op_4699_cast_fp16")];
            bool attn_83_transpose_x_0 = const()[name = string("attn_83_transpose_x_0"), val = bool(false)];
            bool attn_83_transpose_y_0 = const()[name = string("attn_83_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_83_cast_fp16 = matmul(transpose_x = attn_83_transpose_x_0, transpose_y = attn_83_transpose_y_0, x = var_4699_cast_fp16, y = obj_293_cast_fp16)[name = string("attn_83_cast_fp16")];
            tensor<int32, [4]> var_4702 = const()[name = string("op_4702"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_203_cast_fp16 = reshape(shape = var_4702, x = attn_83_cast_fp16)[name = string("input_203_cast_fp16")];
            string obj_291_pad_type_0 = const()[name = string("obj_291_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_291_strides_0 = const()[name = string("obj_291_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_291_pad_0 = const()[name = string("obj_291_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_291_dilations_0 = const()[name = string("obj_291_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_291_groups_0 = const()[name = string("obj_291_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(366682304))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(367468800))))[name = string("layers_20_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_20_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_20_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(367477056)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_291_cast_fp16 = conv(bias = layers_20_encoder_attn_o_proj_bias_to_fp16, dilations = obj_291_dilations_0, groups = obj_291_groups_0, pad = obj_291_pad_0, pad_type = obj_291_pad_type_0, strides = obj_291_strides_0, weight = layers_20_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_203_cast_fp16)[name = string("obj_291_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_125_cast_fp16 = add(x = inputs_123_cast_fp16, y = obj_291_cast_fp16)[name = string("inputs_125_cast_fp16")];
            tensor<int32, [1]> out_125_axes_0 = const()[name = string("out_125_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_4723_to_fp16 = const()[name = string("op_4723_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_125_cast_fp16 = layer_norm(axes = out_125_axes_0, epsilon = var_4723_to_fp16, x = inputs_125_cast_fp16)[name = string("out_125_cast_fp16")];
            tensor<fp16, [1024]> input_205_gamma_0_to_fp16 = const()[name = string("input_205_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(367479168)))];
            tensor<fp16, [1024]> input_205_beta_0_to_fp16 = const()[name = string("input_205_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(367481280)))];
            fp16 input_205_epsilon_0_to_fp16 = const()[name = string("input_205_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_205_cast_fp16 = batch_norm(beta = input_205_beta_0_to_fp16, epsilon = input_205_epsilon_0_to_fp16, gamma = input_205_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_125_cast_fp16)[name = string("input_205_cast_fp16")];
            string input_207_pad_type_0 = const()[name = string("input_207_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_207_strides_0 = const()[name = string("input_207_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_207_pad_0 = const()[name = string("input_207_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_207_dilations_0 = const()[name = string("input_207_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_207_groups_0 = const()[name = string("input_207_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_20_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(367483392))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370629184))))[name = string("layers_20_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_20_fc1_bias_to_fp16 = const()[name = string("layers_20_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370662016)))];
            tensor<fp16, [1, 4096, 1, 1]> input_207_cast_fp16 = conv(bias = layers_20_fc1_bias_to_fp16, dilations = input_207_dilations_0, groups = input_207_groups_0, pad = input_207_pad_0, pad_type = input_207_pad_type_0, strides = input_207_strides_0, weight = layers_20_fc1_weight_to_fp16_palettized, x = input_205_cast_fp16)[name = string("input_207_cast_fp16")];
            string input_209_mode_0 = const()[name = string("input_209_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_209_cast_fp16 = gelu(mode = input_209_mode_0, x = input_207_cast_fp16)[name = string("input_209_cast_fp16")];
            string hidden_states_43_pad_type_0 = const()[name = string("hidden_states_43_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_43_strides_0 = const()[name = string("hidden_states_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_43_pad_0 = const()[name = string("hidden_states_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_43_dilations_0 = const()[name = string("hidden_states_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_43_groups_0 = const()[name = string("hidden_states_43_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_20_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370670272))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373816064))))[name = string("layers_20_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_20_fc2_bias_to_fp16 = const()[name = string("layers_20_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373824320)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_43_cast_fp16 = conv(bias = layers_20_fc2_bias_to_fp16, dilations = hidden_states_43_dilations_0, groups = hidden_states_43_groups_0, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = hidden_states_43_strides_0, weight = layers_20_fc2_weight_to_fp16_palettized, x = input_209_cast_fp16)[name = string("hidden_states_43_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_127_cast_fp16 = add(x = inputs_125_cast_fp16, y = hidden_states_43_cast_fp16)[name = string("inputs_127_cast_fp16")];
            int32 var_4759 = const()[name = string("op_4759"), val = int32(3)];
            tensor<int32, [1]> out_127_axes_0 = const()[name = string("out_127_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_4784_to_fp16 = const()[name = string("op_4784_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_127_cast_fp16 = layer_norm(axes = out_127_axes_0, epsilon = var_4784_to_fp16, x = inputs_127_cast_fp16)[name = string("out_127_cast_fp16")];
            tensor<fp16, [1024]> obj_295_gamma_0_to_fp16 = const()[name = string("obj_295_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373826432)))];
            tensor<fp16, [1024]> obj_295_beta_0_to_fp16 = const()[name = string("obj_295_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373828544)))];
            fp16 obj_295_epsilon_0_to_fp16 = const()[name = string("obj_295_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_295_cast_fp16 = batch_norm(beta = obj_295_beta_0_to_fp16, epsilon = obj_295_epsilon_0_to_fp16, gamma = obj_295_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_127_cast_fp16)[name = string("obj_295_cast_fp16")];
            string query_85_pad_type_0 = const()[name = string("query_85_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_85_strides_0 = const()[name = string("query_85_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_85_pad_0 = const()[name = string("query_85_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_85_dilations_0 = const()[name = string("query_85_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_85_groups_0 = const()[name = string("query_85_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373830656))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(374617152))))[name = string("layers_21_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(374625408)))];
            tensor<fp16, [1, 1024, 1, 1]> query_85_cast_fp16 = conv(bias = layers_21_self_attn_q_proj_bias_to_fp16, dilations = query_85_dilations_0, groups = query_85_groups_0, pad = query_85_pad_0, pad_type = query_85_pad_type_0, strides = query_85_strides_0, weight = layers_21_self_attn_q_proj_weight_to_fp16_palettized, x = obj_295_cast_fp16)[name = string("query_85_cast_fp16")];
            string current_key_43_pad_type_0 = const()[name = string("current_key_43_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_43_strides_0 = const()[name = string("current_key_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_43_pad_0 = const()[name = string("current_key_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_43_dilations_0 = const()[name = string("current_key_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_43_groups_0 = const()[name = string("current_key_43_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(374627520))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375414016))))[name = string("layers_21_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_43_cast_fp16 = conv(dilations = current_key_43_dilations_0, groups = current_key_43_groups_0, pad = current_key_43_pad_0, pad_type = current_key_43_pad_type_0, strides = current_key_43_strides_0, weight = layers_21_self_attn_k_proj_weight_to_fp16_palettized, x = obj_295_cast_fp16)[name = string("current_key_43_cast_fp16")];
            string current_value_43_pad_type_0 = const()[name = string("current_value_43_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_43_strides_0 = const()[name = string("current_value_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_43_pad_0 = const()[name = string("current_value_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_43_dilations_0 = const()[name = string("current_value_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_43_groups_0 = const()[name = string("current_value_43_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375422272))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376208768))))[name = string("layers_21_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376217024)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_43_cast_fp16 = conv(bias = layers_21_self_attn_v_proj_bias_to_fp16, dilations = current_value_43_dilations_0, groups = current_value_43_groups_0, pad = current_value_43_pad_0, pad_type = current_value_43_pad_type_0, strides = current_value_43_strides_0, weight = layers_21_self_attn_v_proj_weight_to_fp16_palettized, x = obj_295_cast_fp16)[name = string("current_value_43_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4823_cast_fp16 = mul(x = var_87_cast_fp16_21, y = var_207_cast_fp16)[name = string("op_4823_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4824_cast_fp16 = mul(x = current_key_43_cast_fp16, y = var_205_cast_fp16)[name = string("op_4824_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_85_cast_fp16 = add(x = var_4823_cast_fp16, y = var_4824_cast_fp16)[name = string("key_85_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4827_cast_fp16 = mul(x = var_114_cast_fp16_21, y = var_207_cast_fp16)[name = string("op_4827_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_4828_cast_fp16 = mul(x = current_value_43_cast_fp16, y = var_205_cast_fp16)[name = string("op_4828_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_85_cast_fp16 = add(x = var_4827_cast_fp16, y = var_4828_cast_fp16)[name = string("value_85_cast_fp16")];
            tensor<int32, [4]> var_4832 = const()[name = string("op_4832"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_85_cast_fp16 = reshape(shape = var_4832, x = query_85_cast_fp16)[name = string("mh_q_85_cast_fp16")];
            fp16 var_4834_to_fp16 = const()[name = string("op_4834_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_4835_cast_fp16 = mul(x = mh_q_85_cast_fp16, y = var_4834_to_fp16)[name = string("op_4835_cast_fp16")];
            tensor<int32, [4]> var_4838 = const()[name = string("op_4838"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_4839_cast_fp16 = reshape(shape = var_4838, x = key_85_cast_fp16)[name = string("op_4839_cast_fp16")];
            bool mh_w_127_transpose_x_0 = const()[name = string("mh_w_127_transpose_x_0"), val = bool(true)];
            bool mh_w_127_transpose_y_0 = const()[name = string("mh_w_127_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_127_cast_fp16 = matmul(transpose_x = mh_w_127_transpose_x_0, transpose_y = mh_w_127_transpose_y_0, x = var_4835_cast_fp16, y = var_4839_cast_fp16)[name = string("mh_w_127_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_129_cast_fp16 = add(x = mh_w_127_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_129_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_4847_cast_fp16 = softmax(axis = var_4759, x = mh_w_129_cast_fp16)[name = string("op_4847_cast_fp16")];
            tensor<int32, [4]> var_4848 = const()[name = string("op_4848"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_4849_cast_fp16 = reshape(shape = var_4848, x = value_85_cast_fp16)[name = string("op_4849_cast_fp16")];
            bool attn_85_transpose_x_0 = const()[name = string("attn_85_transpose_x_0"), val = bool(false)];
            bool attn_85_transpose_y_0 = const()[name = string("attn_85_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_85_cast_fp16 = matmul(transpose_x = attn_85_transpose_x_0, transpose_y = attn_85_transpose_y_0, x = var_4849_cast_fp16, y = var_4847_cast_fp16)[name = string("attn_85_cast_fp16")];
            tensor<int32, [4]> var_4852 = const()[name = string("op_4852"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_211_cast_fp16 = reshape(shape = var_4852, x = attn_85_cast_fp16)[name = string("input_211_cast_fp16")];
            string obj_301_pad_type_0 = const()[name = string("obj_301_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_301_strides_0 = const()[name = string("obj_301_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_301_pad_0 = const()[name = string("obj_301_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_301_dilations_0 = const()[name = string("obj_301_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_301_groups_0 = const()[name = string("obj_301_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376219136))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377005632))))[name = string("layers_21_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_21_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377013888)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_301_cast_fp16 = conv(bias = layers_21_self_attn_o_proj_bias_to_fp16, dilations = obj_301_dilations_0, groups = obj_301_groups_0, pad = obj_301_pad_0, pad_type = obj_301_pad_type_0, strides = obj_301_strides_0, weight = layers_21_self_attn_o_proj_weight_to_fp16_palettized, x = input_211_cast_fp16)[name = string("obj_301_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_129_cast_fp16 = add(x = inputs_127_cast_fp16, y = obj_301_cast_fp16)[name = string("inputs_129_cast_fp16")];
            tensor<int32, [1]> out_129_axes_0 = const()[name = string("out_129_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_4874_to_fp16 = const()[name = string("op_4874_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_129_cast_fp16 = layer_norm(axes = out_129_axes_0, epsilon = var_4874_to_fp16, x = inputs_129_cast_fp16)[name = string("out_129_cast_fp16")];
            tensor<fp16, [1024]> obj_303_gamma_0_to_fp16 = const()[name = string("obj_303_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377016000)))];
            tensor<fp16, [1024]> obj_303_beta_0_to_fp16 = const()[name = string("obj_303_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377018112)))];
            fp16 obj_303_epsilon_0_to_fp16 = const()[name = string("obj_303_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_303_cast_fp16 = batch_norm(beta = obj_303_beta_0_to_fp16, epsilon = obj_303_epsilon_0_to_fp16, gamma = obj_303_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_129_cast_fp16)[name = string("obj_303_cast_fp16")];
            string query_87_pad_type_0 = const()[name = string("query_87_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_87_strides_0 = const()[name = string("query_87_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_87_pad_0 = const()[name = string("query_87_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_87_dilations_0 = const()[name = string("query_87_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_87_groups_0 = const()[name = string("query_87_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377020224))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377806720))))[name = string("layers_21_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_21_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_21_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377814976)))];
            tensor<fp16, [1, 1024, 1, 1]> query_87_cast_fp16 = conv(bias = layers_21_encoder_attn_q_proj_bias_to_fp16, dilations = query_87_dilations_0, groups = query_87_groups_0, pad = query_87_pad_0, pad_type = query_87_pad_type_0, strides = query_87_strides_0, weight = layers_21_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_303_cast_fp16)[name = string("query_87_cast_fp16")];
            string key_87_pad_type_0 = const()[name = string("key_87_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_87_strides_0 = const()[name = string("key_87_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_87_pad_0 = const()[name = string("key_87_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_87_dilations_0 = const()[name = string("key_87_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_87_groups_0 = const()[name = string("key_87_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377817088))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378603584))))[name = string("layers_21_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_87_cast_fp16 = conv(dilations = key_87_dilations_0, groups = key_87_groups_0, pad = key_87_pad_0, pad_type = key_87_pad_type_0, strides = key_87_strides_0, weight = layers_21_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_87_cast_fp16")];
            string value_87_pad_type_0 = const()[name = string("value_87_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_87_strides_0 = const()[name = string("value_87_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_87_pad_0 = const()[name = string("value_87_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_87_dilations_0 = const()[name = string("value_87_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_87_groups_0 = const()[name = string("value_87_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378611840))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(379398336))))[name = string("layers_21_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_21_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_21_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(379406592)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_87_cast_fp16 = conv(bias = layers_21_encoder_attn_v_proj_bias_to_fp16, dilations = value_87_dilations_0, groups = value_87_groups_0, pad = value_87_pad_0, pad_type = value_87_pad_type_0, strides = value_87_strides_0, weight = layers_21_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_87_cast_fp16")];
            tensor<int32, [4]> var_4910 = const()[name = string("op_4910"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_87_cast_fp16 = reshape(shape = var_4910, x = query_87_cast_fp16)[name = string("mh_q_87_cast_fp16")];
            fp16 var_4912_to_fp16 = const()[name = string("op_4912_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_4913_cast_fp16 = mul(x = mh_q_87_cast_fp16, y = var_4912_to_fp16)[name = string("op_4913_cast_fp16")];
            tensor<int32, [4]> var_4916 = const()[name = string("op_4916"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_4917_cast_fp16 = reshape(shape = var_4916, x = key_87_cast_fp16)[name = string("op_4917_cast_fp16")];
            bool mh_w_131_transpose_x_0 = const()[name = string("mh_w_131_transpose_x_0"), val = bool(true)];
            bool mh_w_131_transpose_y_0 = const()[name = string("mh_w_131_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_131_cast_fp16 = matmul(transpose_x = mh_w_131_transpose_x_0, transpose_y = mh_w_131_transpose_y_0, x = var_4913_cast_fp16, y = var_4917_cast_fp16)[name = string("mh_w_131_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_307_cast_fp16 = softmax(axis = var_4759, x = mh_w_131_cast_fp16)[name = string("obj_307_cast_fp16")];
            tensor<int32, [4]> var_4921 = const()[name = string("op_4921"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_4922_cast_fp16 = reshape(shape = var_4921, x = value_87_cast_fp16)[name = string("op_4922_cast_fp16")];
            bool attn_87_transpose_x_0 = const()[name = string("attn_87_transpose_x_0"), val = bool(false)];
            bool attn_87_transpose_y_0 = const()[name = string("attn_87_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_87_cast_fp16 = matmul(transpose_x = attn_87_transpose_x_0, transpose_y = attn_87_transpose_y_0, x = var_4922_cast_fp16, y = obj_307_cast_fp16)[name = string("attn_87_cast_fp16")];
            tensor<int32, [4]> var_4925 = const()[name = string("op_4925"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_213_cast_fp16 = reshape(shape = var_4925, x = attn_87_cast_fp16)[name = string("input_213_cast_fp16")];
            string obj_305_pad_type_0 = const()[name = string("obj_305_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_305_strides_0 = const()[name = string("obj_305_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_305_pad_0 = const()[name = string("obj_305_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_305_dilations_0 = const()[name = string("obj_305_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_305_groups_0 = const()[name = string("obj_305_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(379408704))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380195200))))[name = string("layers_21_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_21_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_21_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380203456)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_305_cast_fp16 = conv(bias = layers_21_encoder_attn_o_proj_bias_to_fp16, dilations = obj_305_dilations_0, groups = obj_305_groups_0, pad = obj_305_pad_0, pad_type = obj_305_pad_type_0, strides = obj_305_strides_0, weight = layers_21_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_213_cast_fp16)[name = string("obj_305_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_131_cast_fp16 = add(x = inputs_129_cast_fp16, y = obj_305_cast_fp16)[name = string("inputs_131_cast_fp16")];
            tensor<int32, [1]> out_131_axes_0 = const()[name = string("out_131_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_4943_to_fp16 = const()[name = string("op_4943_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_131_cast_fp16 = layer_norm(axes = out_131_axes_0, epsilon = var_4943_to_fp16, x = inputs_131_cast_fp16)[name = string("out_131_cast_fp16")];
            tensor<fp16, [1024]> input_215_gamma_0_to_fp16 = const()[name = string("input_215_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380205568)))];
            tensor<fp16, [1024]> input_215_beta_0_to_fp16 = const()[name = string("input_215_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380207680)))];
            fp16 input_215_epsilon_0_to_fp16 = const()[name = string("input_215_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_215_cast_fp16 = batch_norm(beta = input_215_beta_0_to_fp16, epsilon = input_215_epsilon_0_to_fp16, gamma = input_215_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_131_cast_fp16)[name = string("input_215_cast_fp16")];
            string input_217_pad_type_0 = const()[name = string("input_217_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_217_strides_0 = const()[name = string("input_217_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_217_pad_0 = const()[name = string("input_217_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_217_dilations_0 = const()[name = string("input_217_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_217_groups_0 = const()[name = string("input_217_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_21_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380209792))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(383355584))))[name = string("layers_21_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_21_fc1_bias_to_fp16 = const()[name = string("layers_21_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(383388416)))];
            tensor<fp16, [1, 4096, 1, 1]> input_217_cast_fp16 = conv(bias = layers_21_fc1_bias_to_fp16, dilations = input_217_dilations_0, groups = input_217_groups_0, pad = input_217_pad_0, pad_type = input_217_pad_type_0, strides = input_217_strides_0, weight = layers_21_fc1_weight_to_fp16_palettized, x = input_215_cast_fp16)[name = string("input_217_cast_fp16")];
            string input_219_mode_0 = const()[name = string("input_219_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_219_cast_fp16 = gelu(mode = input_219_mode_0, x = input_217_cast_fp16)[name = string("input_219_cast_fp16")];
            string hidden_states_45_pad_type_0 = const()[name = string("hidden_states_45_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_45_strides_0 = const()[name = string("hidden_states_45_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_45_pad_0 = const()[name = string("hidden_states_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_45_dilations_0 = const()[name = string("hidden_states_45_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_45_groups_0 = const()[name = string("hidden_states_45_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_21_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(383396672))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(386542464))))[name = string("layers_21_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_21_fc2_bias_to_fp16 = const()[name = string("layers_21_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(386550720)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_45_cast_fp16 = conv(bias = layers_21_fc2_bias_to_fp16, dilations = hidden_states_45_dilations_0, groups = hidden_states_45_groups_0, pad = hidden_states_45_pad_0, pad_type = hidden_states_45_pad_type_0, strides = hidden_states_45_strides_0, weight = layers_21_fc2_weight_to_fp16_palettized, x = input_219_cast_fp16)[name = string("hidden_states_45_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_133_cast_fp16 = add(x = inputs_131_cast_fp16, y = hidden_states_45_cast_fp16)[name = string("inputs_133_cast_fp16")];
            int32 var_4978 = const()[name = string("op_4978"), val = int32(3)];
            tensor<int32, [1]> out_133_axes_0 = const()[name = string("out_133_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_5003_to_fp16 = const()[name = string("op_5003_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_133_cast_fp16 = layer_norm(axes = out_133_axes_0, epsilon = var_5003_to_fp16, x = inputs_133_cast_fp16)[name = string("out_133_cast_fp16")];
            tensor<fp16, [1024]> obj_309_gamma_0_to_fp16 = const()[name = string("obj_309_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(386552832)))];
            tensor<fp16, [1024]> obj_309_beta_0_to_fp16 = const()[name = string("obj_309_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(386554944)))];
            fp16 obj_309_epsilon_0_to_fp16 = const()[name = string("obj_309_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_309_cast_fp16 = batch_norm(beta = obj_309_beta_0_to_fp16, epsilon = obj_309_epsilon_0_to_fp16, gamma = obj_309_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_133_cast_fp16)[name = string("obj_309_cast_fp16")];
            string query_89_pad_type_0 = const()[name = string("query_89_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_89_strides_0 = const()[name = string("query_89_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_89_pad_0 = const()[name = string("query_89_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_89_dilations_0 = const()[name = string("query_89_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_89_groups_0 = const()[name = string("query_89_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(386557056))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387343552))))[name = string("layers_22_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387351808)))];
            tensor<fp16, [1, 1024, 1, 1]> query_89_cast_fp16 = conv(bias = layers_22_self_attn_q_proj_bias_to_fp16, dilations = query_89_dilations_0, groups = query_89_groups_0, pad = query_89_pad_0, pad_type = query_89_pad_type_0, strides = query_89_strides_0, weight = layers_22_self_attn_q_proj_weight_to_fp16_palettized, x = obj_309_cast_fp16)[name = string("query_89_cast_fp16")];
            string current_key_45_pad_type_0 = const()[name = string("current_key_45_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_45_strides_0 = const()[name = string("current_key_45_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_45_pad_0 = const()[name = string("current_key_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_45_dilations_0 = const()[name = string("current_key_45_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_45_groups_0 = const()[name = string("current_key_45_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387353920))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(388140416))))[name = string("layers_22_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_45_cast_fp16 = conv(dilations = current_key_45_dilations_0, groups = current_key_45_groups_0, pad = current_key_45_pad_0, pad_type = current_key_45_pad_type_0, strides = current_key_45_strides_0, weight = layers_22_self_attn_k_proj_weight_to_fp16_palettized, x = obj_309_cast_fp16)[name = string("current_key_45_cast_fp16")];
            string current_value_45_pad_type_0 = const()[name = string("current_value_45_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_45_strides_0 = const()[name = string("current_value_45_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_45_pad_0 = const()[name = string("current_value_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_45_dilations_0 = const()[name = string("current_value_45_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_45_groups_0 = const()[name = string("current_value_45_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(388148672))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(388935168))))[name = string("layers_22_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(388943424)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_45_cast_fp16 = conv(bias = layers_22_self_attn_v_proj_bias_to_fp16, dilations = current_value_45_dilations_0, groups = current_value_45_groups_0, pad = current_value_45_pad_0, pad_type = current_value_45_pad_type_0, strides = current_value_45_strides_0, weight = layers_22_self_attn_v_proj_weight_to_fp16_palettized, x = obj_309_cast_fp16)[name = string("current_value_45_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_5042_cast_fp16 = mul(x = var_87_cast_fp16_22, y = var_207_cast_fp16)[name = string("op_5042_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_5043_cast_fp16 = mul(x = current_key_45_cast_fp16, y = var_205_cast_fp16)[name = string("op_5043_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_89_cast_fp16 = add(x = var_5042_cast_fp16, y = var_5043_cast_fp16)[name = string("key_89_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_5046_cast_fp16 = mul(x = var_114_cast_fp16_22, y = var_207_cast_fp16)[name = string("op_5046_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_5047_cast_fp16 = mul(x = current_value_45_cast_fp16, y = var_205_cast_fp16)[name = string("op_5047_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_89_cast_fp16 = add(x = var_5046_cast_fp16, y = var_5047_cast_fp16)[name = string("value_89_cast_fp16")];
            tensor<int32, [4]> var_5051 = const()[name = string("op_5051"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_89_cast_fp16 = reshape(shape = var_5051, x = query_89_cast_fp16)[name = string("mh_q_89_cast_fp16")];
            fp16 var_5053_to_fp16 = const()[name = string("op_5053_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_5054_cast_fp16 = mul(x = mh_q_89_cast_fp16, y = var_5053_to_fp16)[name = string("op_5054_cast_fp16")];
            tensor<int32, [4]> var_5057 = const()[name = string("op_5057"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_5058_cast_fp16 = reshape(shape = var_5057, x = key_89_cast_fp16)[name = string("op_5058_cast_fp16")];
            bool mh_w_133_transpose_x_0 = const()[name = string("mh_w_133_transpose_x_0"), val = bool(true)];
            bool mh_w_133_transpose_y_0 = const()[name = string("mh_w_133_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_133_cast_fp16 = matmul(transpose_x = mh_w_133_transpose_x_0, transpose_y = mh_w_133_transpose_y_0, x = var_5054_cast_fp16, y = var_5058_cast_fp16)[name = string("mh_w_133_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_135_cast_fp16 = add(x = mh_w_133_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_135_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_5066_cast_fp16 = softmax(axis = var_4978, x = mh_w_135_cast_fp16)[name = string("op_5066_cast_fp16")];
            tensor<int32, [4]> var_5067 = const()[name = string("op_5067"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_5068_cast_fp16 = reshape(shape = var_5067, x = value_89_cast_fp16)[name = string("op_5068_cast_fp16")];
            bool attn_89_transpose_x_0 = const()[name = string("attn_89_transpose_x_0"), val = bool(false)];
            bool attn_89_transpose_y_0 = const()[name = string("attn_89_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_89_cast_fp16 = matmul(transpose_x = attn_89_transpose_x_0, transpose_y = attn_89_transpose_y_0, x = var_5068_cast_fp16, y = var_5066_cast_fp16)[name = string("attn_89_cast_fp16")];
            tensor<int32, [4]> var_5071 = const()[name = string("op_5071"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_221_cast_fp16 = reshape(shape = var_5071, x = attn_89_cast_fp16)[name = string("input_221_cast_fp16")];
            string obj_315_pad_type_0 = const()[name = string("obj_315_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_315_strides_0 = const()[name = string("obj_315_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_315_pad_0 = const()[name = string("obj_315_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_315_dilations_0 = const()[name = string("obj_315_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_315_groups_0 = const()[name = string("obj_315_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(388945536))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(389732032))))[name = string("layers_22_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_22_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(389740288)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_315_cast_fp16 = conv(bias = layers_22_self_attn_o_proj_bias_to_fp16, dilations = obj_315_dilations_0, groups = obj_315_groups_0, pad = obj_315_pad_0, pad_type = obj_315_pad_type_0, strides = obj_315_strides_0, weight = layers_22_self_attn_o_proj_weight_to_fp16_palettized, x = input_221_cast_fp16)[name = string("obj_315_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_135_cast_fp16 = add(x = inputs_133_cast_fp16, y = obj_315_cast_fp16)[name = string("inputs_135_cast_fp16")];
            tensor<int32, [1]> out_135_axes_0 = const()[name = string("out_135_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_5093_to_fp16 = const()[name = string("op_5093_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_135_cast_fp16 = layer_norm(axes = out_135_axes_0, epsilon = var_5093_to_fp16, x = inputs_135_cast_fp16)[name = string("out_135_cast_fp16")];
            tensor<fp16, [1024]> obj_317_gamma_0_to_fp16 = const()[name = string("obj_317_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(389742400)))];
            tensor<fp16, [1024]> obj_317_beta_0_to_fp16 = const()[name = string("obj_317_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(389744512)))];
            fp16 obj_317_epsilon_0_to_fp16 = const()[name = string("obj_317_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_317_cast_fp16 = batch_norm(beta = obj_317_beta_0_to_fp16, epsilon = obj_317_epsilon_0_to_fp16, gamma = obj_317_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_135_cast_fp16)[name = string("obj_317_cast_fp16")];
            string query_91_pad_type_0 = const()[name = string("query_91_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_91_strides_0 = const()[name = string("query_91_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_91_pad_0 = const()[name = string("query_91_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_91_dilations_0 = const()[name = string("query_91_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_91_groups_0 = const()[name = string("query_91_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(389746624))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390533120))))[name = string("layers_22_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_22_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_22_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390541376)))];
            tensor<fp16, [1, 1024, 1, 1]> query_91_cast_fp16 = conv(bias = layers_22_encoder_attn_q_proj_bias_to_fp16, dilations = query_91_dilations_0, groups = query_91_groups_0, pad = query_91_pad_0, pad_type = query_91_pad_type_0, strides = query_91_strides_0, weight = layers_22_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_317_cast_fp16)[name = string("query_91_cast_fp16")];
            string key_91_pad_type_0 = const()[name = string("key_91_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_91_strides_0 = const()[name = string("key_91_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_91_pad_0 = const()[name = string("key_91_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_91_dilations_0 = const()[name = string("key_91_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_91_groups_0 = const()[name = string("key_91_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390543488))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(391329984))))[name = string("layers_22_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_91_cast_fp16 = conv(dilations = key_91_dilations_0, groups = key_91_groups_0, pad = key_91_pad_0, pad_type = key_91_pad_type_0, strides = key_91_strides_0, weight = layers_22_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_91_cast_fp16")];
            string value_91_pad_type_0 = const()[name = string("value_91_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_91_strides_0 = const()[name = string("value_91_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_91_pad_0 = const()[name = string("value_91_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_91_dilations_0 = const()[name = string("value_91_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_91_groups_0 = const()[name = string("value_91_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(391338240))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392124736))))[name = string("layers_22_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_22_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_22_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392132992)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_91_cast_fp16 = conv(bias = layers_22_encoder_attn_v_proj_bias_to_fp16, dilations = value_91_dilations_0, groups = value_91_groups_0, pad = value_91_pad_0, pad_type = value_91_pad_type_0, strides = value_91_strides_0, weight = layers_22_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_91_cast_fp16")];
            tensor<int32, [4]> var_5129 = const()[name = string("op_5129"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_91_cast_fp16 = reshape(shape = var_5129, x = query_91_cast_fp16)[name = string("mh_q_91_cast_fp16")];
            fp16 var_5131_to_fp16 = const()[name = string("op_5131_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_5132_cast_fp16 = mul(x = mh_q_91_cast_fp16, y = var_5131_to_fp16)[name = string("op_5132_cast_fp16")];
            tensor<int32, [4]> var_5135 = const()[name = string("op_5135"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_5136_cast_fp16 = reshape(shape = var_5135, x = key_91_cast_fp16)[name = string("op_5136_cast_fp16")];
            bool mh_w_137_transpose_x_0 = const()[name = string("mh_w_137_transpose_x_0"), val = bool(true)];
            bool mh_w_137_transpose_y_0 = const()[name = string("mh_w_137_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_137_cast_fp16 = matmul(transpose_x = mh_w_137_transpose_x_0, transpose_y = mh_w_137_transpose_y_0, x = var_5132_cast_fp16, y = var_5136_cast_fp16)[name = string("mh_w_137_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_321_cast_fp16 = softmax(axis = var_4978, x = mh_w_137_cast_fp16)[name = string("obj_321_cast_fp16")];
            tensor<int32, [4]> var_5140 = const()[name = string("op_5140"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_5141_cast_fp16 = reshape(shape = var_5140, x = value_91_cast_fp16)[name = string("op_5141_cast_fp16")];
            bool attn_91_transpose_x_0 = const()[name = string("attn_91_transpose_x_0"), val = bool(false)];
            bool attn_91_transpose_y_0 = const()[name = string("attn_91_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_91_cast_fp16 = matmul(transpose_x = attn_91_transpose_x_0, transpose_y = attn_91_transpose_y_0, x = var_5141_cast_fp16, y = obj_321_cast_fp16)[name = string("attn_91_cast_fp16")];
            tensor<int32, [4]> var_5144 = const()[name = string("op_5144"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_223_cast_fp16 = reshape(shape = var_5144, x = attn_91_cast_fp16)[name = string("input_223_cast_fp16")];
            string obj_319_pad_type_0 = const()[name = string("obj_319_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_319_strides_0 = const()[name = string("obj_319_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_319_pad_0 = const()[name = string("obj_319_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_319_dilations_0 = const()[name = string("obj_319_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_319_groups_0 = const()[name = string("obj_319_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392135104))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392921600))))[name = string("layers_22_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_22_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_22_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392929856)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_319_cast_fp16 = conv(bias = layers_22_encoder_attn_o_proj_bias_to_fp16, dilations = obj_319_dilations_0, groups = obj_319_groups_0, pad = obj_319_pad_0, pad_type = obj_319_pad_type_0, strides = obj_319_strides_0, weight = layers_22_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_223_cast_fp16)[name = string("obj_319_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_137_cast_fp16 = add(x = inputs_135_cast_fp16, y = obj_319_cast_fp16)[name = string("inputs_137_cast_fp16")];
            tensor<int32, [1]> out_137_axes_0 = const()[name = string("out_137_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_5162_to_fp16 = const()[name = string("op_5162_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_137_cast_fp16 = layer_norm(axes = out_137_axes_0, epsilon = var_5162_to_fp16, x = inputs_137_cast_fp16)[name = string("out_137_cast_fp16")];
            tensor<fp16, [1024]> input_225_gamma_0_to_fp16 = const()[name = string("input_225_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392931968)))];
            tensor<fp16, [1024]> input_225_beta_0_to_fp16 = const()[name = string("input_225_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392934080)))];
            fp16 input_225_epsilon_0_to_fp16 = const()[name = string("input_225_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_225_cast_fp16 = batch_norm(beta = input_225_beta_0_to_fp16, epsilon = input_225_epsilon_0_to_fp16, gamma = input_225_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_137_cast_fp16)[name = string("input_225_cast_fp16")];
            string input_227_pad_type_0 = const()[name = string("input_227_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_227_strides_0 = const()[name = string("input_227_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_227_pad_0 = const()[name = string("input_227_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_227_dilations_0 = const()[name = string("input_227_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_227_groups_0 = const()[name = string("input_227_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_22_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392936192))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396081984))))[name = string("layers_22_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_22_fc1_bias_to_fp16 = const()[name = string("layers_22_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396114816)))];
            tensor<fp16, [1, 4096, 1, 1]> input_227_cast_fp16 = conv(bias = layers_22_fc1_bias_to_fp16, dilations = input_227_dilations_0, groups = input_227_groups_0, pad = input_227_pad_0, pad_type = input_227_pad_type_0, strides = input_227_strides_0, weight = layers_22_fc1_weight_to_fp16_palettized, x = input_225_cast_fp16)[name = string("input_227_cast_fp16")];
            string input_229_mode_0 = const()[name = string("input_229_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_229_cast_fp16 = gelu(mode = input_229_mode_0, x = input_227_cast_fp16)[name = string("input_229_cast_fp16")];
            string hidden_states_47_pad_type_0 = const()[name = string("hidden_states_47_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_47_strides_0 = const()[name = string("hidden_states_47_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_47_pad_0 = const()[name = string("hidden_states_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_47_dilations_0 = const()[name = string("hidden_states_47_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_47_groups_0 = const()[name = string("hidden_states_47_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_22_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396123072))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399268864))))[name = string("layers_22_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_22_fc2_bias_to_fp16 = const()[name = string("layers_22_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399277120)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_47_cast_fp16 = conv(bias = layers_22_fc2_bias_to_fp16, dilations = hidden_states_47_dilations_0, groups = hidden_states_47_groups_0, pad = hidden_states_47_pad_0, pad_type = hidden_states_47_pad_type_0, strides = hidden_states_47_strides_0, weight = layers_22_fc2_weight_to_fp16_palettized, x = input_229_cast_fp16)[name = string("hidden_states_47_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_139_cast_fp16 = add(x = inputs_137_cast_fp16, y = hidden_states_47_cast_fp16)[name = string("inputs_139_cast_fp16")];
            int32 var_5197 = const()[name = string("op_5197"), val = int32(3)];
            tensor<int32, [1]> out_139_axes_0 = const()[name = string("out_139_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_5222_to_fp16 = const()[name = string("op_5222_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_139_cast_fp16 = layer_norm(axes = out_139_axes_0, epsilon = var_5222_to_fp16, x = inputs_139_cast_fp16)[name = string("out_139_cast_fp16")];
            tensor<fp16, [1024]> obj_323_gamma_0_to_fp16 = const()[name = string("obj_323_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399279232)))];
            tensor<fp16, [1024]> obj_323_beta_0_to_fp16 = const()[name = string("obj_323_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399281344)))];
            fp16 obj_323_epsilon_0_to_fp16 = const()[name = string("obj_323_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_323_cast_fp16 = batch_norm(beta = obj_323_beta_0_to_fp16, epsilon = obj_323_epsilon_0_to_fp16, gamma = obj_323_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_139_cast_fp16)[name = string("obj_323_cast_fp16")];
            string query_93_pad_type_0 = const()[name = string("query_93_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_93_strides_0 = const()[name = string("query_93_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_93_pad_0 = const()[name = string("query_93_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_93_dilations_0 = const()[name = string("query_93_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_93_groups_0 = const()[name = string("query_93_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399283456))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(400069952))))[name = string("layers_23_self_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_23_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(400078208)))];
            tensor<fp16, [1, 1024, 1, 1]> query_93_cast_fp16 = conv(bias = layers_23_self_attn_q_proj_bias_to_fp16, dilations = query_93_dilations_0, groups = query_93_groups_0, pad = query_93_pad_0, pad_type = query_93_pad_type_0, strides = query_93_strides_0, weight = layers_23_self_attn_q_proj_weight_to_fp16_palettized, x = obj_323_cast_fp16)[name = string("query_93_cast_fp16")];
            string current_key_pad_type_0 = const()[name = string("current_key_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_key_strides_0 = const()[name = string("current_key_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_key_pad_0 = const()[name = string("current_key_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_key_dilations_0 = const()[name = string("current_key_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_key_groups_0 = const()[name = string("current_key_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(400080320))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(400866816))))[name = string("layers_23_self_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1]> current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_23_self_attn_k_proj_weight_to_fp16_palettized, x = obj_323_cast_fp16)[name = string("current_key_cast_fp16")];
            string current_value_pad_type_0 = const()[name = string("current_value_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> current_value_strides_0 = const()[name = string("current_value_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> current_value_pad_0 = const()[name = string("current_value_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> current_value_dilations_0 = const()[name = string("current_value_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 current_value_groups_0 = const()[name = string("current_value_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(400875072))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(401661568))))[name = string("layers_23_self_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_23_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(401669824)))];
            tensor<fp16, [1, 1024, 1, 1]> current_value_cast_fp16 = conv(bias = layers_23_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_23_self_attn_v_proj_weight_to_fp16_palettized, x = obj_323_cast_fp16)[name = string("current_value_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_5261_cast_fp16 = mul(x = var_87_cast_fp16_23, y = var_207_cast_fp16)[name = string("op_5261_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_5262_cast_fp16 = mul(x = current_key_cast_fp16, y = var_205_cast_fp16)[name = string("op_5262_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> key_93_cast_fp16 = add(x = var_5261_cast_fp16, y = var_5262_cast_fp16)[name = string("key_93_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_5265_cast_fp16 = mul(x = var_114_cast_fp16_23, y = var_207_cast_fp16)[name = string("op_5265_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> var_5266_cast_fp16 = mul(x = current_value_cast_fp16, y = var_205_cast_fp16)[name = string("op_5266_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 448]> value_93_cast_fp16 = add(x = var_5265_cast_fp16, y = var_5266_cast_fp16)[name = string("value_93_cast_fp16")];
            tensor<int32, [4]> var_5270 = const()[name = string("op_5270"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_93_cast_fp16 = reshape(shape = var_5270, x = query_93_cast_fp16)[name = string("mh_q_93_cast_fp16")];
            fp16 var_5272_to_fp16 = const()[name = string("op_5272_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_5273_cast_fp16 = mul(x = mh_q_93_cast_fp16, y = var_5272_to_fp16)[name = string("op_5273_cast_fp16")];
            tensor<int32, [4]> var_5276 = const()[name = string("op_5276"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_5277_cast_fp16 = reshape(shape = var_5276, x = key_93_cast_fp16)[name = string("op_5277_cast_fp16")];
            bool mh_w_139_transpose_x_0 = const()[name = string("mh_w_139_transpose_x_0"), val = bool(true)];
            bool mh_w_139_transpose_y_0 = const()[name = string("mh_w_139_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 448]> mh_w_139_cast_fp16 = matmul(transpose_x = mh_w_139_transpose_x_0, transpose_y = mh_w_139_transpose_y_0, x = var_5273_cast_fp16, y = var_5277_cast_fp16)[name = string("mh_w_139_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> mh_w_141_cast_fp16 = add(x = mh_w_139_cast_fp16, y = var_229_cast_fp16)[name = string("mh_w_141_cast_fp16")];
            tensor<fp16, [1, 16, 1, 448]> var_5285_cast_fp16 = softmax(axis = var_5197, x = mh_w_141_cast_fp16)[name = string("op_5285_cast_fp16")];
            tensor<int32, [4]> var_5286 = const()[name = string("op_5286"), val = tensor<int32, [4]>([1, 16, 64, 448])];
            tensor<fp16, [1, 16, 64, 448]> var_5287_cast_fp16 = reshape(shape = var_5286, x = value_93_cast_fp16)[name = string("op_5287_cast_fp16")];
            bool attn_93_transpose_x_0 = const()[name = string("attn_93_transpose_x_0"), val = bool(false)];
            bool attn_93_transpose_y_0 = const()[name = string("attn_93_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_93_cast_fp16 = matmul(transpose_x = attn_93_transpose_x_0, transpose_y = attn_93_transpose_y_0, x = var_5287_cast_fp16, y = var_5285_cast_fp16)[name = string("attn_93_cast_fp16")];
            tensor<int32, [4]> var_5290 = const()[name = string("op_5290"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_231_cast_fp16 = reshape(shape = var_5290, x = attn_93_cast_fp16)[name = string("input_231_cast_fp16")];
            string obj_329_pad_type_0 = const()[name = string("obj_329_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_329_strides_0 = const()[name = string("obj_329_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_329_pad_0 = const()[name = string("obj_329_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_329_dilations_0 = const()[name = string("obj_329_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_329_groups_0 = const()[name = string("obj_329_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(401671936))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(402458432))))[name = string("layers_23_self_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_23_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(402466688)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_329_cast_fp16 = conv(bias = layers_23_self_attn_o_proj_bias_to_fp16, dilations = obj_329_dilations_0, groups = obj_329_groups_0, pad = obj_329_pad_0, pad_type = obj_329_pad_type_0, strides = obj_329_strides_0, weight = layers_23_self_attn_o_proj_weight_to_fp16_palettized, x = input_231_cast_fp16)[name = string("obj_329_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_141_cast_fp16 = add(x = inputs_139_cast_fp16, y = obj_329_cast_fp16)[name = string("inputs_141_cast_fp16")];
            tensor<int32, [1]> out_141_axes_0 = const()[name = string("out_141_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_5312_to_fp16 = const()[name = string("op_5312_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_141_cast_fp16 = layer_norm(axes = out_141_axes_0, epsilon = var_5312_to_fp16, x = inputs_141_cast_fp16)[name = string("out_141_cast_fp16")];
            tensor<fp16, [1024]> obj_331_gamma_0_to_fp16 = const()[name = string("obj_331_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(402468800)))];
            tensor<fp16, [1024]> obj_331_beta_0_to_fp16 = const()[name = string("obj_331_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(402470912)))];
            fp16 obj_331_epsilon_0_to_fp16 = const()[name = string("obj_331_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> obj_331_cast_fp16 = batch_norm(beta = obj_331_beta_0_to_fp16, epsilon = obj_331_epsilon_0_to_fp16, gamma = obj_331_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_141_cast_fp16)[name = string("obj_331_cast_fp16")];
            string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_strides_0 = const()[name = string("query_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_pad_0 = const()[name = string("query_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_encoder_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(402473024))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403259520))))[name = string("layers_23_encoder_attn_q_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_23_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_23_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403267776)))];
            tensor<fp16, [1, 1024, 1, 1]> query_cast_fp16 = conv(bias = layers_23_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_23_encoder_attn_q_proj_weight_to_fp16_palettized, x = obj_331_cast_fp16)[name = string("query_cast_fp16")];
            string key_pad_type_0 = const()[name = string("key_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_strides_0 = const()[name = string("key_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_pad_0 = const()[name = string("key_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_dilations_0 = const()[name = string("key_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_groups_0 = const()[name = string("key_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_encoder_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403269888))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404056384))))[name = string("layers_23_encoder_attn_k_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 1500]> key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_23_encoder_attn_k_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("key_cast_fp16")];
            string value_pad_type_0 = const()[name = string("value_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_strides_0 = const()[name = string("value_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_pad_0 = const()[name = string("value_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_dilations_0 = const()[name = string("value_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_groups_0 = const()[name = string("value_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_encoder_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404064640))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404851136))))[name = string("layers_23_encoder_attn_v_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_23_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("layers_23_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404859392)))];
            tensor<fp16, [1, 1024, 1, 1500]> value_cast_fp16 = conv(bias = layers_23_encoder_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_23_encoder_attn_v_proj_weight_to_fp16_palettized, x = encoder_output_embeds)[name = string("value_cast_fp16")];
            tensor<int32, [4]> var_5348 = const()[name = string("op_5348"), val = tensor<int32, [4]>([1, 16, 64, 1])];
            tensor<fp16, [1, 16, 64, 1]> mh_q_cast_fp16 = reshape(shape = var_5348, x = query_cast_fp16)[name = string("mh_q_cast_fp16")];
            fp16 var_5350_to_fp16 = const()[name = string("op_5350_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 64, 1]> var_5351_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_5350_to_fp16)[name = string("op_5351_cast_fp16")];
            tensor<int32, [4]> var_5354 = const()[name = string("op_5354"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_5355_cast_fp16 = reshape(shape = var_5354, x = key_cast_fp16)[name = string("op_5355_cast_fp16")];
            bool mh_w_transpose_x_0 = const()[name = string("mh_w_transpose_x_0"), val = bool(true)];
            bool mh_w_transpose_y_0 = const()[name = string("mh_w_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 1, 1500]> mh_w_cast_fp16 = matmul(transpose_x = mh_w_transpose_x_0, transpose_y = mh_w_transpose_y_0, x = var_5351_cast_fp16, y = var_5355_cast_fp16)[name = string("mh_w_cast_fp16")];
            tensor<fp16, [1, 16, 1, 1500]> obj_335_cast_fp16 = softmax(axis = var_5197, x = mh_w_cast_fp16)[name = string("obj_335_cast_fp16")];
            tensor<int32, [4]> var_5359 = const()[name = string("op_5359"), val = tensor<int32, [4]>([1, 16, 64, 1500])];
            tensor<fp16, [1, 16, 64, 1500]> var_5360_cast_fp16 = reshape(shape = var_5359, x = value_cast_fp16)[name = string("op_5360_cast_fp16")];
            bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)];
            bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 16, 64, 1]> attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_5360_cast_fp16, y = obj_335_cast_fp16)[name = string("attn_cast_fp16")];
            tensor<int32, [4]> var_5363 = const()[name = string("op_5363"), val = tensor<int32, [4]>([1, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 1]> input_233_cast_fp16 = reshape(shape = var_5363, x = attn_cast_fp16)[name = string("input_233_cast_fp16")];
            string obj_333_pad_type_0 = const()[name = string("obj_333_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_333_strides_0 = const()[name = string("obj_333_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_333_pad_0 = const()[name = string("obj_333_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_333_dilations_0 = const()[name = string("obj_333_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_333_groups_0 = const()[name = string("obj_333_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_encoder_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404861504))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(405648000))))[name = string("layers_23_encoder_attn_o_proj_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_23_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_23_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(405656256)))];
            tensor<fp16, [1, 1024, 1, 1]> obj_333_cast_fp16 = conv(bias = layers_23_encoder_attn_o_proj_bias_to_fp16, dilations = obj_333_dilations_0, groups = obj_333_groups_0, pad = obj_333_pad_0, pad_type = obj_333_pad_type_0, strides = obj_333_strides_0, weight = layers_23_encoder_attn_o_proj_weight_to_fp16_palettized, x = input_233_cast_fp16)[name = string("obj_333_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_143_cast_fp16 = add(x = inputs_141_cast_fp16, y = obj_333_cast_fp16)[name = string("inputs_143_cast_fp16")];
            tensor<int32, [1]> out_143_axes_0 = const()[name = string("out_143_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_5384_to_fp16 = const()[name = string("op_5384_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_143_cast_fp16 = layer_norm(axes = out_143_axes_0, epsilon = var_5384_to_fp16, x = inputs_143_cast_fp16)[name = string("out_143_cast_fp16")];
            tensor<fp16, [1024]> input_235_gamma_0_to_fp16 = const()[name = string("input_235_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(405658368)))];
            tensor<fp16, [1024]> input_235_beta_0_to_fp16 = const()[name = string("input_235_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(405660480)))];
            fp16 input_235_epsilon_0_to_fp16 = const()[name = string("input_235_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> input_235_cast_fp16 = batch_norm(beta = input_235_beta_0_to_fp16, epsilon = input_235_epsilon_0_to_fp16, gamma = input_235_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_143_cast_fp16)[name = string("input_235_cast_fp16")];
            string input_237_pad_type_0 = const()[name = string("input_237_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_237_strides_0 = const()[name = string("input_237_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_237_pad_0 = const()[name = string("input_237_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_237_dilations_0 = const()[name = string("input_237_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_237_groups_0 = const()[name = string("input_237_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_23_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(405662592))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408808384))))[name = string("layers_23_fc1_weight_to_fp16_palettized")];
            tensor<fp16, [4096]> layers_23_fc1_bias_to_fp16 = const()[name = string("layers_23_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408841216)))];
            tensor<fp16, [1, 4096, 1, 1]> input_237_cast_fp16 = conv(bias = layers_23_fc1_bias_to_fp16, dilations = input_237_dilations_0, groups = input_237_groups_0, pad = input_237_pad_0, pad_type = input_237_pad_type_0, strides = input_237_strides_0, weight = layers_23_fc1_weight_to_fp16_palettized, x = input_235_cast_fp16)[name = string("input_237_cast_fp16")];
            string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 4096, 1, 1]> input_cast_fp16 = gelu(mode = input_mode_0, x = input_237_cast_fp16)[name = string("input_cast_fp16")];
            string hidden_states_49_pad_type_0 = const()[name = string("hidden_states_49_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_49_strides_0 = const()[name = string("hidden_states_49_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_49_pad_0 = const()[name = string("hidden_states_49_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_49_dilations_0 = const()[name = string("hidden_states_49_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_49_groups_0 = const()[name = string("hidden_states_49_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> layers_23_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408849472))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411995264))))[name = string("layers_23_fc2_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> layers_23_fc2_bias_to_fp16 = const()[name = string("layers_23_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412003520)))];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_49_cast_fp16 = conv(bias = layers_23_fc2_bias_to_fp16, dilations = hidden_states_49_dilations_0, groups = hidden_states_49_groups_0, pad = hidden_states_49_pad_0, pad_type = hidden_states_49_pad_type_0, strides = hidden_states_49_strides_0, weight = layers_23_fc2_weight_to_fp16_palettized, x = input_cast_fp16)[name = string("hidden_states_49_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> inputs_cast_fp16 = add(x = inputs_143_cast_fp16, y = hidden_states_49_cast_fp16)[name = string("inputs_cast_fp16")];
            tensor<int32, [1]> out_axes_0 = const()[name = string("out_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_5427_to_fp16 = const()[name = string("op_5427_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_5427_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")];
            tensor<fp16, [1024]> hidden_states_gamma_0_to_fp16 = const()[name = string("hidden_states_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412005632)))];
            tensor<fp16, [1024]> hidden_states_beta_0_to_fp16 = const()[name = string("hidden_states_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412007744)))];
            fp16 hidden_states_epsilon_0_to_fp16 = const()[name = string("hidden_states_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 1]> hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = string("hidden_states_cast_fp16")];
            tensor<int32, [1]> var_5438_axes_0 = const()[name = string("op_5438_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 1024, 1]> var_5438_cast_fp16 = squeeze(axes = var_5438_axes_0, x = hidden_states_cast_fp16)[name = string("op_5438_cast_fp16")];
            tensor<int32, [3]> var_5441_perm_0 = const()[name = string("op_5441_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [51865]> linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor<fp16, [51865]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412009856)))];
            tensor<fp16, [1, 1, 1024]> var_5441_cast_fp16 = transpose(perm = var_5441_perm_0, x = var_5438_cast_fp16)[name = string("transpose_0")];
            tensor<fp16, [1, 1, 51865]> logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_5441_cast_fp16)[name = string("linear_0_cast_fp16")];
            int32 var_5445 = const()[name = string("op_5445"), val = int32(1)];
            bool obj_339_interleave_0 = const()[name = string("obj_339_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 24576, 1, 1]> key_cache_updates = concat(axis = var_5445, interleave = obj_339_interleave_0, values = (current_key_1_cast_fp16, current_key_3_cast_fp16, current_key_5_cast_fp16, current_key_7_cast_fp16, current_key_9_cast_fp16, current_key_11_cast_fp16, current_key_13_cast_fp16, current_key_15_cast_fp16, current_key_17_cast_fp16, current_key_19_cast_fp16, current_key_21_cast_fp16, current_key_23_cast_fp16, current_key_25_cast_fp16, current_key_27_cast_fp16, current_key_29_cast_fp16, current_key_31_cast_fp16, current_key_33_cast_fp16, current_key_35_cast_fp16, current_key_37_cast_fp16, current_key_39_cast_fp16, current_key_41_cast_fp16, current_key_43_cast_fp16, current_key_45_cast_fp16, current_key_cast_fp16))[name = string("obj_339_cast_fp16")];
            int32 var_5448 = const()[name = string("op_5448"), val = int32(1)];
            bool obj_341_interleave_0 = const()[name = string("obj_341_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 24576, 1, 1]> value_cache_updates = concat(axis = var_5448, interleave = obj_341_interleave_0, values = (current_value_1_cast_fp16, current_value_3_cast_fp16, current_value_5_cast_fp16, current_value_7_cast_fp16, current_value_9_cast_fp16, current_value_11_cast_fp16, current_value_13_cast_fp16, current_value_15_cast_fp16, current_value_17_cast_fp16, current_value_19_cast_fp16, current_value_21_cast_fp16, current_value_23_cast_fp16, current_value_25_cast_fp16, current_value_27_cast_fp16, current_value_29_cast_fp16, current_value_31_cast_fp16, current_value_33_cast_fp16, current_value_35_cast_fp16, current_value_37_cast_fp16, current_value_39_cast_fp16, current_value_41_cast_fp16, current_value_43_cast_fp16, current_value_45_cast_fp16, current_value_cast_fp16))[name = string("obj_341_cast_fp16")];
            tensor<int32, [4]> var_5459_begin_0 = const()[name = string("op_5459_begin_0"), val = tensor<int32, [4]>([0, 15, 0, 0])];
            tensor<int32, [4]> var_5459_end_0 = const()[name = string("op_5459_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])];
            tensor<bool, [4]> var_5459_end_mask_0 = const()[name = string("op_5459_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 1, 1, 1500]> var_5459_cast_fp16 = slice_by_index(begin = var_5459_begin_0, end = var_5459_end_0, end_mask = var_5459_end_mask_0, x = obj_195_cast_fp16)[name = string("op_5459_cast_fp16")];
            tensor<int32, [4]> var_5462_begin_0 = const()[name = string("op_5462_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5462_end_0 = const()[name = string("op_5462_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])];
            tensor<bool, [4]> var_5462_end_mask_0 = const()[name = string("op_5462_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])];
            tensor<bool, [4]> var_5462_squeeze_mask_0 = const()[name = string("op_5462_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])];
            tensor<fp16, [1, 1, 1500]> var_5462_cast_fp16 = slice_by_index(begin = var_5462_begin_0, end = var_5462_end_0, end_mask = var_5462_end_mask_0, squeeze_mask = var_5462_squeeze_mask_0, x = var_5459_cast_fp16)[name = string("op_5462_cast_fp16")];
            tensor<int32, [4]> var_5477_begin_0 = const()[name = string("op_5477_begin_0"), val = tensor<int32, [4]>([0, 4, 0, 0])];
            tensor<int32, [4]> var_5477_end_0 = const()[name = string("op_5477_end_0"), val = tensor<int32, [4]>([1, 5, 1, 1500])];
            tensor<bool, [4]> var_5477_end_mask_0 = const()[name = string("op_5477_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 1, 1, 1500]> var_5477_cast_fp16 = slice_by_index(begin = var_5477_begin_0, end = var_5477_end_0, end_mask = var_5477_end_mask_0, x = obj_223_cast_fp16)[name = string("op_5477_cast_fp16")];
            tensor<int32, [4]> var_5480_begin_0 = const()[name = string("op_5480_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5480_end_0 = const()[name = string("op_5480_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])];
            tensor<bool, [4]> var_5480_end_mask_0 = const()[name = string("op_5480_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])];
            tensor<bool, [4]> var_5480_squeeze_mask_0 = const()[name = string("op_5480_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])];
            tensor<fp16, [1, 1, 1500]> var_5480_cast_fp16 = slice_by_index(begin = var_5480_begin_0, end = var_5480_end_0, end_mask = var_5480_end_mask_0, squeeze_mask = var_5480_squeeze_mask_0, x = var_5477_cast_fp16)[name = string("op_5480_cast_fp16")];
            tensor<int32, [4]> var_5495_begin_0 = const()[name = string("op_5495_begin_0"), val = tensor<int32, [4]>([0, 15, 0, 0])];
            tensor<int32, [4]> var_5495_end_0 = const()[name = string("op_5495_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])];
            tensor<bool, [4]> var_5495_end_mask_0 = const()[name = string("op_5495_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 1, 1, 1500]> var_5495_cast_fp16 = slice_by_index(begin = var_5495_begin_0, end = var_5495_end_0, end_mask = var_5495_end_mask_0, x = obj_223_cast_fp16)[name = string("op_5495_cast_fp16")];
            tensor<int32, [4]> var_5498_begin_0 = const()[name = string("op_5498_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5498_end_0 = const()[name = string("op_5498_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])];
            tensor<bool, [4]> var_5498_end_mask_0 = const()[name = string("op_5498_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])];
            tensor<bool, [4]> var_5498_squeeze_mask_0 = const()[name = string("op_5498_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])];
            tensor<fp16, [1, 1, 1500]> var_5498_cast_fp16 = slice_by_index(begin = var_5498_begin_0, end = var_5498_end_0, end_mask = var_5498_end_mask_0, squeeze_mask = var_5498_squeeze_mask_0, x = var_5495_cast_fp16)[name = string("op_5498_cast_fp16")];
            tensor<int32, [4]> var_5513_begin_0 = const()[name = string("op_5513_begin_0"), val = tensor<int32, [4]>([0, 1, 0, 0])];
            tensor<int32, [4]> var_5513_end_0 = const()[name = string("op_5513_end_0"), val = tensor<int32, [4]>([1, 2, 1, 1500])];
            tensor<bool, [4]> var_5513_end_mask_0 = const()[name = string("op_5513_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 1, 1, 1500]> var_5513_cast_fp16 = slice_by_index(begin = var_5513_begin_0, end = var_5513_end_0, end_mask = var_5513_end_mask_0, x = obj_237_cast_fp16)[name = string("op_5513_cast_fp16")];
            tensor<int32, [4]> var_5516_begin_0 = const()[name = string("op_5516_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5516_end_0 = const()[name = string("op_5516_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])];
            tensor<bool, [4]> var_5516_end_mask_0 = const()[name = string("op_5516_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])];
            tensor<bool, [4]> var_5516_squeeze_mask_0 = const()[name = string("op_5516_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])];
            tensor<fp16, [1, 1, 1500]> var_5516_cast_fp16 = slice_by_index(begin = var_5516_begin_0, end = var_5516_end_0, end_mask = var_5516_end_mask_0, squeeze_mask = var_5516_squeeze_mask_0, x = var_5513_cast_fp16)[name = string("op_5516_cast_fp16")];
            tensor<int32, [4]> var_5531_begin_0 = const()[name = string("op_5531_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5531_end_0 = const()[name = string("op_5531_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])];
            tensor<bool, [4]> var_5531_end_mask_0 = const()[name = string("op_5531_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 1, 1, 1500]> var_5531_cast_fp16 = slice_by_index(begin = var_5531_begin_0, end = var_5531_end_0, end_mask = var_5531_end_mask_0, x = obj_293_cast_fp16)[name = string("op_5531_cast_fp16")];
            tensor<int32, [4]> var_5534_begin_0 = const()[name = string("op_5534_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5534_end_0 = const()[name = string("op_5534_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])];
            tensor<bool, [4]> var_5534_end_mask_0 = const()[name = string("op_5534_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])];
            tensor<bool, [4]> var_5534_squeeze_mask_0 = const()[name = string("op_5534_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])];
            tensor<fp16, [1, 1, 1500]> var_5534_cast_fp16 = slice_by_index(begin = var_5534_begin_0, end = var_5534_end_0, end_mask = var_5534_end_mask_0, squeeze_mask = var_5534_squeeze_mask_0, x = var_5531_cast_fp16)[name = string("op_5534_cast_fp16")];
            tensor<int32, [4]> var_5549_begin_0 = const()[name = string("op_5549_begin_0"), val = tensor<int32, [4]>([0, 4, 0, 0])];
            tensor<int32, [4]> var_5549_end_0 = const()[name = string("op_5549_end_0"), val = tensor<int32, [4]>([1, 5, 1, 1500])];
            tensor<bool, [4]> var_5549_end_mask_0 = const()[name = string("op_5549_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 1, 1, 1500]> var_5549_cast_fp16 = slice_by_index(begin = var_5549_begin_0, end = var_5549_end_0, end_mask = var_5549_end_mask_0, x = obj_335_cast_fp16)[name = string("op_5549_cast_fp16")];
            tensor<int32, [4]> var_5552_begin_0 = const()[name = string("op_5552_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5552_end_0 = const()[name = string("op_5552_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])];
            tensor<bool, [4]> var_5552_end_mask_0 = const()[name = string("op_5552_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])];
            tensor<bool, [4]> var_5552_squeeze_mask_0 = const()[name = string("op_5552_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])];
            tensor<fp16, [1, 1, 1500]> var_5552_cast_fp16 = slice_by_index(begin = var_5552_begin_0, end = var_5552_end_0, end_mask = var_5552_end_mask_0, squeeze_mask = var_5552_squeeze_mask_0, x = var_5549_cast_fp16)[name = string("op_5552_cast_fp16")];
            int32 var_5559 = const()[name = string("op_5559"), val = int32(1)];
            bool var_5560_interleave_0 = const()[name = string("op_5560_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 6, 1500]> var_5560_cast_fp16 = concat(axis = var_5559, interleave = var_5560_interleave_0, values = (var_5462_cast_fp16, var_5480_cast_fp16, var_5498_cast_fp16, var_5516_cast_fp16, var_5534_cast_fp16, var_5552_cast_fp16))[name = string("op_5560_cast_fp16")];
            tensor<int32, [1]> obj_axes_0 = const()[name = string("obj_axes_0"), val = tensor<int32, [1]>([1])];
            bool obj_keep_dims_0 = const()[name = string("obj_keep_dims_0"), val = bool(false)];
            tensor<fp16, [1, 1500]> alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = obj_keep_dims_0, x = var_5560_cast_fp16)[name = string("obj_cast_fp16")];
        } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights);
}