program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3600.16.1"}, {"coremlc-version", "3600.25.1"}})]
{
    func f1<ios18>(tensor<fp16, [5, 1, 512, 6]> conv, tensor<fp16, [3, 1, 2, 29, 64]> k, tensor<fp16, [1, 1, 48]> lat48, tensor<fp16, [1, 1, 6, 35]> mask, tensor<fp16, [3, 1, 2, 29, 64]> v) {
            tensor<fp16, [512, 48]> proj_weight_to_fp16 = const()[name = string("proj_weight_to_fp16"), val = tensor<fp16, [512, 48]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            tensor<fp16, [512]> proj_bias_to_fp16 = const()[name = string("proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49280)))];
            tensor<fp16, [1, 1, 512]> linear_0_cast_fp16 = linear(bias = proj_bias_to_fp16, weight = proj_weight_to_fp16, x = lat48)[name = string("linear_0_cast_fp16")];
            bool x_1_transpose_x_0 = const()[name = string("x_1_transpose_x_0"), val = bool(false)];
            bool x_1_transpose_y_0 = const()[name = string("x_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [6, 1]> const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = tensor<fp16, [6, 1]>([[0x1p+0], [0x1p+0], [0x1p+0], [0x1p+0], [0x1p+0], [0x1p+0]])];
            tensor<fp16, [1, 6, 512]> x_1_cast_fp16 = matmul(transpose_x = x_1_transpose_x_0, transpose_y = x_1_transpose_y_0, x = const_0_to_fp16, y = linear_0_cast_fp16)[name = string("x_1_cast_fp16")];
            int32 var_31 = const()[name = string("op_31"), val = int32(-1)];
            bool var_32 = const()[name = string("op_32"), val = bool(true)];
            tensor<fp16, [1, 6, 512]> var_395_cast_fp16 = abs(x = x_1_cast_fp16)[name = string("op_395_cast_fp16")];
            tensor<int32, [1]> var_397_axes_0 = const()[name = string("op_397_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_397_cast_fp16 = reduce_max(axes = var_397_axes_0, keep_dims = var_32, x = var_395_cast_fp16)[name = string("op_397_cast_fp16")];
            fp16 var_33_to_fp16 = const()[name = string("op_33_to_fp16"), val = fp16(0x1.a38p-14)];
            tensor<fp16, [1, 6, 1]> s_1_cast_fp16 = maximum(x = var_397_cast_fp16, y = var_33_to_fp16)[name = string("s_1_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_1_cast_fp16 = real_div(x = x_1_cast_fp16, y = s_1_cast_fp16)[name = string("y_1_cast_fp16")];
            fp16 var_34_promoted_to_fp16 = const()[name = string("op_34_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_400_cast_fp16 = pow(x = y_1_cast_fp16, y = var_34_promoted_to_fp16)[name = string("op_400_cast_fp16")];
            tensor<int32, [1]> var_402_axes_0 = const()[name = string("op_402_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_402_keep_dims_0 = const()[name = string("op_402_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_402_cast_fp16 = reduce_mean(axes = var_402_axes_0, keep_dims = var_402_keep_dims_0, x = var_400_cast_fp16)[name = string("op_402_cast_fp16")];
            fp32 var_403_epsilon_0 = const()[name = string("op_403_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_403_cast_fp16 = inverse(epsilon = var_403_epsilon_0, x = s_1_cast_fp16)[name = string("op_403_cast_fp16")];
            fp16 var_404_to_fp16 = const()[name = string("op_404_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_405_cast_fp16 = mul(x = var_403_cast_fp16, y = var_404_to_fp16)[name = string("op_405_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_406_cast_fp16 = real_div(x = var_405_cast_fp16, y = s_1_cast_fp16)[name = string("op_406_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_407_cast_fp16 = add(x = var_402_cast_fp16, y = var_406_cast_fp16)[name = string("op_407_cast_fp16")];
            fp32 var_408_epsilon_0 = const()[name = string("op_408_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_408_cast_fp16 = rsqrt(epsilon = var_408_epsilon_0, x = var_407_cast_fp16)[name = string("op_408_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_409_cast_fp16 = mul(x = y_1_cast_fp16, y = var_408_cast_fp16)[name = string("op_409_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_0_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50368)))];
            tensor<fp16, [1, 6, 512]> input_1_cast_fp16 = mul(x = core_backbone_layers_0_operator_norm_weight_to_fp16, y = var_409_cast_fp16)[name = string("input_1_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_0_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51456)))];
            tensor<fp16, [1536]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1624384)))];
            tensor<fp16, [1, 6, 1536]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_0_conv_in_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<int32, [3]> bcx_1_perm_0 = const()[name = string("bcx_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_1_begin_0 = const()[name = string("b_1_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_1_end_0 = const()[name = string("b_1_end_0"), val = tensor<int32, [3]>([1, 512, 6])];
            tensor<bool, [3]> b_1_end_mask_0 = const()[name = string("b_1_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 6]> bcx_1_cast_fp16 = transpose(perm = bcx_1_perm_0, x = linear_1_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [1, 512, 6]> b_1_cast_fp16 = slice_by_index(begin = b_1_begin_0, end = b_1_end_0, end_mask = b_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("b_1_cast_fp16")];
            tensor<int32, [3]> c_1_begin_0 = const()[name = string("c_1_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_1_end_0 = const()[name = string("c_1_end_0"), val = tensor<int32, [3]>([1, 1024, 6])];
            tensor<bool, [3]> c_1_end_mask_0 = const()[name = string("c_1_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 6]> c_1_cast_fp16 = slice_by_index(begin = c_1_begin_0, end = c_1_end_0, end_mask = c_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("c_1_cast_fp16")];
            tensor<int32, [3]> xx_1_begin_0 = const()[name = string("xx_1_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_1_end_0 = const()[name = string("xx_1_end_0"), val = tensor<int32, [3]>([1, 1536, 6])];
            tensor<bool, [3]> xx_1_end_mask_0 = const()[name = string("xx_1_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> xx_1_cast_fp16 = slice_by_index(begin = xx_1_begin_0, end = xx_1_end_0, end_mask = xx_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("xx_1_cast_fp16")];
            tensor<fp16, [1, 512, 6]> new_1_cast_fp16 = mul(x = b_1_cast_fp16, y = xx_1_cast_fp16)[name = string("new_1_cast_fp16")];
            tensor<int32, [4]> input_5_begin_0 = const()[name = string("input_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> input_5_end_0 = const()[name = string("input_5_end_0"), val = tensor<int32, [4]>([1, 1, 512, 6])];
            tensor<bool, [4]> input_5_end_mask_0 = const()[name = string("input_5_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_5_squeeze_mask_0 = const()[name = string("input_5_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_5_cast_fp16 = slice_by_index(begin = input_5_begin_0, end = input_5_end_0, end_mask = input_5_end_mask_0, squeeze_mask = input_5_squeeze_mask_0, x = conv)[name = string("input_5_cast_fp16")];
            tensor<int32, [6]> var_424_pad_0 = const()[name = string("op_424_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_424_mode_0 = const()[name = string("op_424_mode_0"), val = string("constant")];
            fp16 const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 12]> var_424_cast_fp16 = pad(constant_val = const_5_to_fp16, mode = var_424_mode_0, pad = var_424_pad_0, x = new_1_cast_fp16)[name = string("op_424_cast_fp16")];
            tensor<int32, [6]> var_426_pad_0 = const()[name = string("op_426_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 6])];
            string var_426_mode_0 = const()[name = string("op_426_mode_0"), val = string("constant")];
            fp16 const_6_to_fp16 = const()[name = string("const_6_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 12]> var_426_cast_fp16 = pad(constant_val = const_6_to_fp16, mode = var_426_mode_0, pad = var_426_pad_0, x = input_5_cast_fp16)[name = string("op_426_cast_fp16")];
            tensor<fp16, [1, 512, 12]> win_1_cast_fp16 = add(x = var_424_cast_fp16, y = var_426_cast_fp16)[name = string("win_1_cast_fp16")];
            tensor<int32, [3]> var_428_begin_0 = const()[name = string("op_428_begin_0"), val = tensor<int32, [3]>([0, 0, 6])];
            tensor<int32, [3]> var_428_end_0 = const()[name = string("op_428_end_0"), val = tensor<int32, [3]>([1, 512, 12])];
            tensor<bool, [3]> var_428_end_mask_0 = const()[name = string("op_428_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_428_cast_fp16 = slice_by_index(begin = var_428_begin_0, end = var_428_end_0, end_mask = var_428_end_mask_0, x = win_1_cast_fp16)[name = string("op_428_cast_fp16")];
            string var_434_pad_type_0 = const()[name = string("op_434_pad_type_0"), val = string("valid")];
            int32 var_434_groups_0 = const()[name = string("op_434_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_434_strides_0 = const()[name = string("op_434_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_434_pad_0 = const()[name = string("op_434_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_434_dilations_0 = const()[name = string("op_434_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_0_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1627520)))];
            tensor<fp16, [1, 512, 6]> var_434_cast_fp16 = conv(dilations = var_434_dilations_0, groups = var_434_groups_0, pad = var_434_pad_0, pad_type = var_434_pad_type_0, strides = var_434_strides_0, weight = core_backbone_layers_0_conv_conv_weight_to_fp16, x = win_1_cast_fp16)[name = string("op_434_cast_fp16")];
            tensor<fp16, [1, 512, 6]> var_436_cast_fp16 = mul(x = c_1_cast_fp16, y = var_434_cast_fp16)[name = string("op_436_cast_fp16")];
            tensor<int32, [3]> input_7_perm_0 = const()[name = string("input_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_0_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1634752)))];
            tensor<fp16, [512]> linear_2_bias_0_to_fp16 = const()[name = string("linear_2_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2159104)))];
            tensor<fp16, [1, 6, 512]> input_7_cast_fp16 = transpose(perm = input_7_perm_0, x = var_436_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [1, 6, 512]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_0_conv_out_proj_weight_to_fp16, x = input_7_cast_fp16)[name = string("linear_2_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_441_cast_fp16 = abs(x = linear_2_cast_fp16)[name = string("op_441_cast_fp16")];
            tensor<int32, [1]> var_443_axes_0 = const()[name = string("op_443_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_443_cast_fp16 = reduce_max(axes = var_443_axes_0, keep_dims = var_32, x = var_441_cast_fp16)[name = string("op_443_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_3_cast_fp16 = maximum(x = var_443_cast_fp16, y = var_33_to_fp16)[name = string("s_3_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_5_cast_fp16 = real_div(x = linear_2_cast_fp16, y = s_3_cast_fp16)[name = string("y_5_cast_fp16")];
            fp16 var_34_promoted_1_to_fp16 = const()[name = string("op_34_promoted_1_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_446_cast_fp16 = pow(x = y_5_cast_fp16, y = var_34_promoted_1_to_fp16)[name = string("op_446_cast_fp16")];
            tensor<int32, [1]> var_448_axes_0 = const()[name = string("op_448_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_448_keep_dims_0 = const()[name = string("op_448_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_448_cast_fp16 = reduce_mean(axes = var_448_axes_0, keep_dims = var_448_keep_dims_0, x = var_446_cast_fp16)[name = string("op_448_cast_fp16")];
            fp32 var_449_epsilon_0 = const()[name = string("op_449_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_449_cast_fp16 = inverse(epsilon = var_449_epsilon_0, x = s_3_cast_fp16)[name = string("op_449_cast_fp16")];
            fp16 var_450_to_fp16 = const()[name = string("op_450_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_451_cast_fp16 = mul(x = var_449_cast_fp16, y = var_450_to_fp16)[name = string("op_451_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_452_cast_fp16 = real_div(x = var_451_cast_fp16, y = s_3_cast_fp16)[name = string("op_452_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_453_cast_fp16 = add(x = var_448_cast_fp16, y = var_452_cast_fp16)[name = string("op_453_cast_fp16")];
            fp32 var_454_epsilon_0 = const()[name = string("op_454_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_454_cast_fp16 = rsqrt(epsilon = var_454_epsilon_0, x = var_453_cast_fp16)[name = string("op_454_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_455_cast_fp16 = mul(x = y_5_cast_fp16, y = var_454_cast_fp16)[name = string("op_455_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_0_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2160192)))];
            tensor<fp16, [1, 6, 512]> a_1_cast_fp16 = mul(x = core_backbone_layers_0_post_operator_norm_weight_to_fp16, y = var_455_cast_fp16)[name = string("a_1_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_5_cast_fp16 = add(x = x_1_cast_fp16, y = a_1_cast_fp16)[name = string("x_5_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_459_cast_fp16 = abs(x = x_5_cast_fp16)[name = string("op_459_cast_fp16")];
            tensor<int32, [1]> var_461_axes_0 = const()[name = string("op_461_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_461_cast_fp16 = reduce_max(axes = var_461_axes_0, keep_dims = var_32, x = var_459_cast_fp16)[name = string("op_461_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_5_cast_fp16 = maximum(x = var_461_cast_fp16, y = var_33_to_fp16)[name = string("s_5_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_7_cast_fp16 = real_div(x = x_5_cast_fp16, y = s_5_cast_fp16)[name = string("y_7_cast_fp16")];
            fp16 var_34_promoted_2_to_fp16 = const()[name = string("op_34_promoted_2_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_464_cast_fp16 = pow(x = y_7_cast_fp16, y = var_34_promoted_2_to_fp16)[name = string("op_464_cast_fp16")];
            tensor<int32, [1]> var_466_axes_0 = const()[name = string("op_466_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_466_keep_dims_0 = const()[name = string("op_466_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_466_cast_fp16 = reduce_mean(axes = var_466_axes_0, keep_dims = var_466_keep_dims_0, x = var_464_cast_fp16)[name = string("op_466_cast_fp16")];
            fp32 var_467_epsilon_0 = const()[name = string("op_467_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_467_cast_fp16 = inverse(epsilon = var_467_epsilon_0, x = s_5_cast_fp16)[name = string("op_467_cast_fp16")];
            fp16 var_468_to_fp16 = const()[name = string("op_468_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_469_cast_fp16 = mul(x = var_467_cast_fp16, y = var_468_to_fp16)[name = string("op_469_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_470_cast_fp16 = real_div(x = var_469_cast_fp16, y = s_5_cast_fp16)[name = string("op_470_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_471_cast_fp16 = add(x = var_466_cast_fp16, y = var_470_cast_fp16)[name = string("op_471_cast_fp16")];
            fp32 var_472_epsilon_0 = const()[name = string("op_472_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_472_cast_fp16 = rsqrt(epsilon = var_472_epsilon_0, x = var_471_cast_fp16)[name = string("op_472_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_473_cast_fp16 = mul(x = y_7_cast_fp16, y = var_472_cast_fp16)[name = string("op_473_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_0_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2161280)))];
            tensor<fp16, [1, 6, 512]> input_9_cast_fp16 = mul(x = core_backbone_layers_0_ffn_norm_weight_to_fp16, y = var_473_cast_fp16)[name = string("input_9_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_0_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2162368)))];
            tensor<fp16, [2304]> linear_3_bias_0_to_fp16 = const()[name = string("linear_3_bias_0_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4521728)))];
            tensor<fp16, [1, 6, 2304]> linear_3_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w1_weight_to_fp16, x = input_9_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_0_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4526400)))];
            tensor<fp16, [1, 6, 2304]> linear_4_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w3_weight_to_fp16, x = input_9_cast_fp16)[name = string("linear_4_cast_fp16")];
            fp16 var_482_to_fp16 = const()[name = string("op_482_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 6, 2304]> var_483_cast_fp16 = mul(x = linear_3_cast_fp16, y = var_482_to_fp16)[name = string("op_483_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_486_cast_fp16 = tanh(x = var_483_cast_fp16)[name = string("op_486_cast_fp16")];
            fp16 var_487_to_fp16 = const()[name = string("op_487_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 6, 2304]> var_488_cast_fp16 = add(x = var_486_cast_fp16, y = var_487_to_fp16)[name = string("op_488_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_489_cast_fp16 = mul(x = var_483_cast_fp16, y = var_488_cast_fp16)[name = string("op_489_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> input_11_cast_fp16 = mul(x = var_489_cast_fp16, y = linear_4_cast_fp16)[name = string("input_11_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_0_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6885760)))];
            tensor<fp16, [1, 6, 512]> linear_5_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w2_weight_to_fp16, x = input_11_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_494_cast_fp16 = abs(x = linear_5_cast_fp16)[name = string("op_494_cast_fp16")];
            tensor<int32, [1]> var_496_axes_0 = const()[name = string("op_496_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_496_cast_fp16 = reduce_max(axes = var_496_axes_0, keep_dims = var_32, x = var_494_cast_fp16)[name = string("op_496_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_7_cast_fp16 = maximum(x = var_496_cast_fp16, y = var_33_to_fp16)[name = string("s_7_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_9_cast_fp16 = real_div(x = linear_5_cast_fp16, y = s_7_cast_fp16)[name = string("y_9_cast_fp16")];
            fp16 var_34_promoted_3_to_fp16 = const()[name = string("op_34_promoted_3_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_499_cast_fp16 = pow(x = y_9_cast_fp16, y = var_34_promoted_3_to_fp16)[name = string("op_499_cast_fp16")];
            tensor<int32, [1]> var_501_axes_0 = const()[name = string("op_501_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_501_keep_dims_0 = const()[name = string("op_501_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_501_cast_fp16 = reduce_mean(axes = var_501_axes_0, keep_dims = var_501_keep_dims_0, x = var_499_cast_fp16)[name = string("op_501_cast_fp16")];
            fp32 var_502_epsilon_0 = const()[name = string("op_502_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_502_cast_fp16 = inverse(epsilon = var_502_epsilon_0, x = s_7_cast_fp16)[name = string("op_502_cast_fp16")];
            fp16 var_503_to_fp16 = const()[name = string("op_503_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_504_cast_fp16 = mul(x = var_502_cast_fp16, y = var_503_to_fp16)[name = string("op_504_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_505_cast_fp16 = real_div(x = var_504_cast_fp16, y = s_7_cast_fp16)[name = string("op_505_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_506_cast_fp16 = add(x = var_501_cast_fp16, y = var_505_cast_fp16)[name = string("op_506_cast_fp16")];
            fp32 var_507_epsilon_0 = const()[name = string("op_507_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_507_cast_fp16 = rsqrt(epsilon = var_507_epsilon_0, x = var_506_cast_fp16)[name = string("op_507_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_508_cast_fp16 = mul(x = y_9_cast_fp16, y = var_507_cast_fp16)[name = string("op_508_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_0_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9245120)))];
            tensor<fp16, [1, 6, 512]> f_1_cast_fp16 = mul(x = core_backbone_layers_0_post_ffn_norm_weight_to_fp16, y = var_508_cast_fp16)[name = string("f_1_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_11_cast_fp16 = add(x = x_5_cast_fp16, y = f_1_cast_fp16)[name = string("x_11_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_512_cast_fp16 = abs(x = x_11_cast_fp16)[name = string("op_512_cast_fp16")];
            tensor<int32, [1]> var_514_axes_0 = const()[name = string("op_514_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_514_cast_fp16 = reduce_max(axes = var_514_axes_0, keep_dims = var_32, x = var_512_cast_fp16)[name = string("op_514_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_9_cast_fp16 = maximum(x = var_514_cast_fp16, y = var_33_to_fp16)[name = string("s_9_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_11_cast_fp16 = real_div(x = x_11_cast_fp16, y = s_9_cast_fp16)[name = string("y_11_cast_fp16")];
            fp16 var_34_promoted_4_to_fp16 = const()[name = string("op_34_promoted_4_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_517_cast_fp16 = pow(x = y_11_cast_fp16, y = var_34_promoted_4_to_fp16)[name = string("op_517_cast_fp16")];
            tensor<int32, [1]> var_519_axes_0 = const()[name = string("op_519_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_519_keep_dims_0 = const()[name = string("op_519_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_519_cast_fp16 = reduce_mean(axes = var_519_axes_0, keep_dims = var_519_keep_dims_0, x = var_517_cast_fp16)[name = string("op_519_cast_fp16")];
            fp32 var_520_epsilon_0 = const()[name = string("op_520_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_520_cast_fp16 = inverse(epsilon = var_520_epsilon_0, x = s_9_cast_fp16)[name = string("op_520_cast_fp16")];
            fp16 var_521_to_fp16 = const()[name = string("op_521_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_522_cast_fp16 = mul(x = var_520_cast_fp16, y = var_521_to_fp16)[name = string("op_522_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_523_cast_fp16 = real_div(x = var_522_cast_fp16, y = s_9_cast_fp16)[name = string("op_523_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_524_cast_fp16 = add(x = var_519_cast_fp16, y = var_523_cast_fp16)[name = string("op_524_cast_fp16")];
            fp32 var_525_epsilon_0 = const()[name = string("op_525_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_525_cast_fp16 = rsqrt(epsilon = var_525_epsilon_0, x = var_524_cast_fp16)[name = string("op_525_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_526_cast_fp16 = mul(x = y_11_cast_fp16, y = var_525_cast_fp16)[name = string("op_526_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_1_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9246208)))];
            tensor<fp16, [1, 6, 512]> input_13_cast_fp16 = mul(x = core_backbone_layers_1_operator_norm_weight_to_fp16, y = var_526_cast_fp16)[name = string("input_13_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_1_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9247296)))];
            tensor<fp16, [1, 6, 1536]> linear_6_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_1_conv_in_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("linear_6_cast_fp16")];
            tensor<int32, [3]> bcx_3_perm_0 = const()[name = string("bcx_3_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_5_begin_0 = const()[name = string("b_5_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_5_end_0 = const()[name = string("b_5_end_0"), val = tensor<int32, [3]>([1, 512, 6])];
            tensor<bool, [3]> b_5_end_mask_0 = const()[name = string("b_5_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 6]> bcx_3_cast_fp16 = transpose(perm = bcx_3_perm_0, x = linear_6_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [1, 512, 6]> b_5_cast_fp16 = slice_by_index(begin = b_5_begin_0, end = b_5_end_0, end_mask = b_5_end_mask_0, x = bcx_3_cast_fp16)[name = string("b_5_cast_fp16")];
            tensor<int32, [3]> c_3_begin_0 = const()[name = string("c_3_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_3_end_0 = const()[name = string("c_3_end_0"), val = tensor<int32, [3]>([1, 1024, 6])];
            tensor<bool, [3]> c_3_end_mask_0 = const()[name = string("c_3_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 6]> c_3_cast_fp16 = slice_by_index(begin = c_3_begin_0, end = c_3_end_0, end_mask = c_3_end_mask_0, x = bcx_3_cast_fp16)[name = string("c_3_cast_fp16")];
            tensor<int32, [3]> xx_3_begin_0 = const()[name = string("xx_3_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_3_end_0 = const()[name = string("xx_3_end_0"), val = tensor<int32, [3]>([1, 1536, 6])];
            tensor<bool, [3]> xx_3_end_mask_0 = const()[name = string("xx_3_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> xx_3_cast_fp16 = slice_by_index(begin = xx_3_begin_0, end = xx_3_end_0, end_mask = xx_3_end_mask_0, x = bcx_3_cast_fp16)[name = string("xx_3_cast_fp16")];
            tensor<fp16, [1, 512, 6]> new_3_cast_fp16 = mul(x = b_5_cast_fp16, y = xx_3_cast_fp16)[name = string("new_3_cast_fp16")];
            tensor<int32, [4]> input_17_begin_0 = const()[name = string("input_17_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])];
            tensor<int32, [4]> input_17_end_0 = const()[name = string("input_17_end_0"), val = tensor<int32, [4]>([2, 1, 512, 6])];
            tensor<bool, [4]> input_17_end_mask_0 = const()[name = string("input_17_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_17_squeeze_mask_0 = const()[name = string("input_17_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_17_cast_fp16 = slice_by_index(begin = input_17_begin_0, end = input_17_end_0, end_mask = input_17_end_mask_0, squeeze_mask = input_17_squeeze_mask_0, x = conv)[name = string("input_17_cast_fp16")];
            tensor<int32, [6]> var_541_pad_0 = const()[name = string("op_541_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_541_mode_0 = const()[name = string("op_541_mode_0"), val = string("constant")];
            fp16 const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 12]> var_541_cast_fp16 = pad(constant_val = const_7_to_fp16, mode = var_541_mode_0, pad = var_541_pad_0, x = new_3_cast_fp16)[name = string("op_541_cast_fp16")];
            tensor<int32, [6]> var_543_pad_0 = const()[name = string("op_543_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 6])];
            string var_543_mode_0 = const()[name = string("op_543_mode_0"), val = string("constant")];
            fp16 const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 12]> var_543_cast_fp16 = pad(constant_val = const_8_to_fp16, mode = var_543_mode_0, pad = var_543_pad_0, x = input_17_cast_fp16)[name = string("op_543_cast_fp16")];
            tensor<fp16, [1, 512, 12]> win_5_cast_fp16 = add(x = var_541_cast_fp16, y = var_543_cast_fp16)[name = string("win_5_cast_fp16")];
            tensor<int32, [3]> var_545_begin_0 = const()[name = string("op_545_begin_0"), val = tensor<int32, [3]>([0, 0, 6])];
            tensor<int32, [3]> var_545_end_0 = const()[name = string("op_545_end_0"), val = tensor<int32, [3]>([1, 512, 12])];
            tensor<bool, [3]> var_545_end_mask_0 = const()[name = string("op_545_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_545_cast_fp16 = slice_by_index(begin = var_545_begin_0, end = var_545_end_0, end_mask = var_545_end_mask_0, x = win_5_cast_fp16)[name = string("op_545_cast_fp16")];
            string var_551_pad_type_0 = const()[name = string("op_551_pad_type_0"), val = string("valid")];
            int32 var_551_groups_0 = const()[name = string("op_551_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_551_strides_0 = const()[name = string("op_551_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_551_pad_0 = const()[name = string("op_551_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_551_dilations_0 = const()[name = string("op_551_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_1_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10820224)))];
            tensor<fp16, [1, 512, 6]> var_551_cast_fp16 = conv(dilations = var_551_dilations_0, groups = var_551_groups_0, pad = var_551_pad_0, pad_type = var_551_pad_type_0, strides = var_551_strides_0, weight = core_backbone_layers_1_conv_conv_weight_to_fp16, x = win_5_cast_fp16)[name = string("op_551_cast_fp16")];
            tensor<fp16, [1, 512, 6]> var_553_cast_fp16 = mul(x = c_3_cast_fp16, y = var_551_cast_fp16)[name = string("op_553_cast_fp16")];
            tensor<int32, [3]> input_19_perm_0 = const()[name = string("input_19_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_1_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10827456)))];
            tensor<fp16, [1, 6, 512]> input_19_cast_fp16 = transpose(perm = input_19_perm_0, x = var_553_cast_fp16)[name = string("transpose_24")];
            tensor<fp16, [1, 6, 512]> linear_7_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_1_conv_out_proj_weight_to_fp16, x = input_19_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_558_cast_fp16 = abs(x = linear_7_cast_fp16)[name = string("op_558_cast_fp16")];
            tensor<int32, [1]> var_560_axes_0 = const()[name = string("op_560_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_560_cast_fp16 = reduce_max(axes = var_560_axes_0, keep_dims = var_32, x = var_558_cast_fp16)[name = string("op_560_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_11_cast_fp16 = maximum(x = var_560_cast_fp16, y = var_33_to_fp16)[name = string("s_11_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_15_cast_fp16 = real_div(x = linear_7_cast_fp16, y = s_11_cast_fp16)[name = string("y_15_cast_fp16")];
            fp16 var_34_promoted_5_to_fp16 = const()[name = string("op_34_promoted_5_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_563_cast_fp16 = pow(x = y_15_cast_fp16, y = var_34_promoted_5_to_fp16)[name = string("op_563_cast_fp16")];
            tensor<int32, [1]> var_565_axes_0 = const()[name = string("op_565_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_565_keep_dims_0 = const()[name = string("op_565_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_565_cast_fp16 = reduce_mean(axes = var_565_axes_0, keep_dims = var_565_keep_dims_0, x = var_563_cast_fp16)[name = string("op_565_cast_fp16")];
            fp32 var_566_epsilon_0 = const()[name = string("op_566_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_566_cast_fp16 = inverse(epsilon = var_566_epsilon_0, x = s_11_cast_fp16)[name = string("op_566_cast_fp16")];
            fp16 var_567_to_fp16 = const()[name = string("op_567_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_568_cast_fp16 = mul(x = var_566_cast_fp16, y = var_567_to_fp16)[name = string("op_568_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_569_cast_fp16 = real_div(x = var_568_cast_fp16, y = s_11_cast_fp16)[name = string("op_569_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_570_cast_fp16 = add(x = var_565_cast_fp16, y = var_569_cast_fp16)[name = string("op_570_cast_fp16")];
            fp32 var_571_epsilon_0 = const()[name = string("op_571_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_571_cast_fp16 = rsqrt(epsilon = var_571_epsilon_0, x = var_570_cast_fp16)[name = string("op_571_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_572_cast_fp16 = mul(x = y_15_cast_fp16, y = var_571_cast_fp16)[name = string("op_572_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_1_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11351808)))];
            tensor<fp16, [1, 6, 512]> a_3_cast_fp16 = mul(x = core_backbone_layers_1_post_operator_norm_weight_to_fp16, y = var_572_cast_fp16)[name = string("a_3_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_15_cast_fp16 = add(x = x_11_cast_fp16, y = a_3_cast_fp16)[name = string("x_15_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_576_cast_fp16 = abs(x = x_15_cast_fp16)[name = string("op_576_cast_fp16")];
            tensor<int32, [1]> var_578_axes_0 = const()[name = string("op_578_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_578_cast_fp16 = reduce_max(axes = var_578_axes_0, keep_dims = var_32, x = var_576_cast_fp16)[name = string("op_578_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_13_cast_fp16 = maximum(x = var_578_cast_fp16, y = var_33_to_fp16)[name = string("s_13_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_17_cast_fp16 = real_div(x = x_15_cast_fp16, y = s_13_cast_fp16)[name = string("y_17_cast_fp16")];
            fp16 var_34_promoted_6_to_fp16 = const()[name = string("op_34_promoted_6_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_581_cast_fp16 = pow(x = y_17_cast_fp16, y = var_34_promoted_6_to_fp16)[name = string("op_581_cast_fp16")];
            tensor<int32, [1]> var_583_axes_0 = const()[name = string("op_583_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_583_keep_dims_0 = const()[name = string("op_583_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_583_cast_fp16 = reduce_mean(axes = var_583_axes_0, keep_dims = var_583_keep_dims_0, x = var_581_cast_fp16)[name = string("op_583_cast_fp16")];
            fp32 var_584_epsilon_0 = const()[name = string("op_584_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_584_cast_fp16 = inverse(epsilon = var_584_epsilon_0, x = s_13_cast_fp16)[name = string("op_584_cast_fp16")];
            fp16 var_585_to_fp16 = const()[name = string("op_585_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_586_cast_fp16 = mul(x = var_584_cast_fp16, y = var_585_to_fp16)[name = string("op_586_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_587_cast_fp16 = real_div(x = var_586_cast_fp16, y = s_13_cast_fp16)[name = string("op_587_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_588_cast_fp16 = add(x = var_583_cast_fp16, y = var_587_cast_fp16)[name = string("op_588_cast_fp16")];
            fp32 var_589_epsilon_0 = const()[name = string("op_589_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_589_cast_fp16 = rsqrt(epsilon = var_589_epsilon_0, x = var_588_cast_fp16)[name = string("op_589_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_590_cast_fp16 = mul(x = y_17_cast_fp16, y = var_589_cast_fp16)[name = string("op_590_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_1_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11352896)))];
            tensor<fp16, [1, 6, 512]> input_21_cast_fp16 = mul(x = core_backbone_layers_1_ffn_norm_weight_to_fp16, y = var_590_cast_fp16)[name = string("input_21_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_1_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11353984)))];
            tensor<fp16, [1, 6, 2304]> linear_8_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w1_weight_to_fp16, x = input_21_cast_fp16)[name = string("linear_8_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_1_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13713344)))];
            tensor<fp16, [1, 6, 2304]> linear_9_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w3_weight_to_fp16, x = input_21_cast_fp16)[name = string("linear_9_cast_fp16")];
            fp16 var_599_to_fp16 = const()[name = string("op_599_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 6, 2304]> var_600_cast_fp16 = mul(x = linear_8_cast_fp16, y = var_599_to_fp16)[name = string("op_600_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_603_cast_fp16 = tanh(x = var_600_cast_fp16)[name = string("op_603_cast_fp16")];
            fp16 var_604_to_fp16 = const()[name = string("op_604_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 6, 2304]> var_605_cast_fp16 = add(x = var_603_cast_fp16, y = var_604_to_fp16)[name = string("op_605_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_606_cast_fp16 = mul(x = var_600_cast_fp16, y = var_605_cast_fp16)[name = string("op_606_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> input_23_cast_fp16 = mul(x = var_606_cast_fp16, y = linear_9_cast_fp16)[name = string("input_23_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_1_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16072704)))];
            tensor<fp16, [1, 6, 512]> linear_10_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w2_weight_to_fp16, x = input_23_cast_fp16)[name = string("linear_10_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_611_cast_fp16 = abs(x = linear_10_cast_fp16)[name = string("op_611_cast_fp16")];
            tensor<int32, [1]> var_613_axes_0 = const()[name = string("op_613_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_613_cast_fp16 = reduce_max(axes = var_613_axes_0, keep_dims = var_32, x = var_611_cast_fp16)[name = string("op_613_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_15_cast_fp16 = maximum(x = var_613_cast_fp16, y = var_33_to_fp16)[name = string("s_15_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_19_cast_fp16 = real_div(x = linear_10_cast_fp16, y = s_15_cast_fp16)[name = string("y_19_cast_fp16")];
            fp16 var_34_promoted_7_to_fp16 = const()[name = string("op_34_promoted_7_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_616_cast_fp16 = pow(x = y_19_cast_fp16, y = var_34_promoted_7_to_fp16)[name = string("op_616_cast_fp16")];
            tensor<int32, [1]> var_618_axes_0 = const()[name = string("op_618_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_618_keep_dims_0 = const()[name = string("op_618_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_618_cast_fp16 = reduce_mean(axes = var_618_axes_0, keep_dims = var_618_keep_dims_0, x = var_616_cast_fp16)[name = string("op_618_cast_fp16")];
            fp32 var_619_epsilon_0 = const()[name = string("op_619_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_619_cast_fp16 = inverse(epsilon = var_619_epsilon_0, x = s_15_cast_fp16)[name = string("op_619_cast_fp16")];
            fp16 var_620_to_fp16 = const()[name = string("op_620_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_621_cast_fp16 = mul(x = var_619_cast_fp16, y = var_620_to_fp16)[name = string("op_621_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_622_cast_fp16 = real_div(x = var_621_cast_fp16, y = s_15_cast_fp16)[name = string("op_622_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_623_cast_fp16 = add(x = var_618_cast_fp16, y = var_622_cast_fp16)[name = string("op_623_cast_fp16")];
            fp32 var_624_epsilon_0 = const()[name = string("op_624_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_624_cast_fp16 = rsqrt(epsilon = var_624_epsilon_0, x = var_623_cast_fp16)[name = string("op_624_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_625_cast_fp16 = mul(x = y_19_cast_fp16, y = var_624_cast_fp16)[name = string("op_625_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_1_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18432064)))];
            tensor<fp16, [1, 6, 512]> f_3_cast_fp16 = mul(x = core_backbone_layers_1_post_ffn_norm_weight_to_fp16, y = var_625_cast_fp16)[name = string("f_3_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_21_cast_fp16 = add(x = x_15_cast_fp16, y = f_3_cast_fp16)[name = string("x_21_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_629_cast_fp16 = abs(x = x_21_cast_fp16)[name = string("op_629_cast_fp16")];
            tensor<int32, [1]> var_631_axes_0 = const()[name = string("op_631_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_631_cast_fp16 = reduce_max(axes = var_631_axes_0, keep_dims = var_32, x = var_629_cast_fp16)[name = string("op_631_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_17_cast_fp16 = maximum(x = var_631_cast_fp16, y = var_33_to_fp16)[name = string("s_17_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_21_cast_fp16 = real_div(x = x_21_cast_fp16, y = s_17_cast_fp16)[name = string("y_21_cast_fp16")];
            fp16 var_34_promoted_8_to_fp16 = const()[name = string("op_34_promoted_8_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_634_cast_fp16 = pow(x = y_21_cast_fp16, y = var_34_promoted_8_to_fp16)[name = string("op_634_cast_fp16")];
            tensor<int32, [1]> var_636_axes_0 = const()[name = string("op_636_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_636_keep_dims_0 = const()[name = string("op_636_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_636_cast_fp16 = reduce_mean(axes = var_636_axes_0, keep_dims = var_636_keep_dims_0, x = var_634_cast_fp16)[name = string("op_636_cast_fp16")];
            fp32 var_637_epsilon_0 = const()[name = string("op_637_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_637_cast_fp16 = inverse(epsilon = var_637_epsilon_0, x = s_17_cast_fp16)[name = string("op_637_cast_fp16")];
            fp16 var_638_to_fp16 = const()[name = string("op_638_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_639_cast_fp16 = mul(x = var_637_cast_fp16, y = var_638_to_fp16)[name = string("op_639_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_640_cast_fp16 = real_div(x = var_639_cast_fp16, y = s_17_cast_fp16)[name = string("op_640_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_641_cast_fp16 = add(x = var_636_cast_fp16, y = var_640_cast_fp16)[name = string("op_641_cast_fp16")];
            fp32 var_642_epsilon_0 = const()[name = string("op_642_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_642_cast_fp16 = rsqrt(epsilon = var_642_epsilon_0, x = var_641_cast_fp16)[name = string("op_642_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_643_cast_fp16 = mul(x = y_21_cast_fp16, y = var_642_cast_fp16)[name = string("op_643_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_2_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18433152)))];
            tensor<fp16, [1, 6, 512]> input_25_cast_fp16 = mul(x = core_backbone_layers_2_operator_norm_weight_to_fp16, y = var_643_cast_fp16)[name = string("input_25_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18434240)))];
            tensor<fp16, [1, 6, 512]> linear_11_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_q_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_11_cast_fp16")];
            tensor<int32, [4]> concat_0 = const()[name = string("concat_0"), val = tensor<int32, [4]>([1, 6, 8, 64])];
            tensor<fp16, [1, 6, 8, 64]> x_23_cast_fp16 = reshape(shape = concat_0, x = linear_11_cast_fp16)[name = string("x_23_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> var_650_cast_fp16 = abs(x = x_23_cast_fp16)[name = string("op_650_cast_fp16")];
            tensor<int32, [1]> var_652_axes_0 = const()[name = string("op_652_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 8, 1]> var_652_cast_fp16 = reduce_max(axes = var_652_axes_0, keep_dims = var_32, x = var_650_cast_fp16)[name = string("op_652_cast_fp16")];
            tensor<fp16, [1, 6, 8, 1]> s_19_cast_fp16 = maximum(x = var_652_cast_fp16, y = var_33_to_fp16)[name = string("s_19_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> y_23_cast_fp16 = real_div(x = x_23_cast_fp16, y = s_19_cast_fp16)[name = string("y_23_cast_fp16")];
            fp16 var_34_promoted_9_to_fp16 = const()[name = string("op_34_promoted_9_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 8, 64]> var_655_cast_fp16 = pow(x = y_23_cast_fp16, y = var_34_promoted_9_to_fp16)[name = string("op_655_cast_fp16")];
            tensor<int32, [1]> var_657_axes_0 = const()[name = string("op_657_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_657_keep_dims_0 = const()[name = string("op_657_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 8, 1]> var_657_cast_fp16 = reduce_mean(axes = var_657_axes_0, keep_dims = var_657_keep_dims_0, x = var_655_cast_fp16)[name = string("op_657_cast_fp16")];
            fp32 var_658_epsilon_0 = const()[name = string("op_658_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 8, 1]> var_658_cast_fp16 = inverse(epsilon = var_658_epsilon_0, x = s_19_cast_fp16)[name = string("op_658_cast_fp16")];
            fp16 var_659_to_fp16 = const()[name = string("op_659_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 8, 1]> var_660_cast_fp16 = mul(x = var_658_cast_fp16, y = var_659_to_fp16)[name = string("op_660_cast_fp16")];
            tensor<fp16, [1, 6, 8, 1]> var_661_cast_fp16 = real_div(x = var_660_cast_fp16, y = s_19_cast_fp16)[name = string("op_661_cast_fp16")];
            tensor<fp16, [1, 6, 8, 1]> var_662_cast_fp16 = add(x = var_657_cast_fp16, y = var_661_cast_fp16)[name = string("op_662_cast_fp16")];
            fp32 var_663_epsilon_0 = const()[name = string("op_663_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 8, 1]> var_663_cast_fp16 = rsqrt(epsilon = var_663_epsilon_0, x = var_662_cast_fp16)[name = string("op_663_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> var_664_cast_fp16 = mul(x = y_23_cast_fp16, y = var_663_cast_fp16)[name = string("op_664_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_2_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18958592)))];
            tensor<fp16, [1, 6, 8, 64]> var_665_cast_fp16 = mul(x = core_backbone_layers_2_self_attn_q_norm_weight_to_fp16, y = var_664_cast_fp16)[name = string("op_665_cast_fp16")];
            tensor<fp16, [128, 512]> core_backbone_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18958784)))];
            tensor<fp16, [128]> linear_12_bias_0_to_fp16 = const()[name = string("linear_12_bias_0_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19089920)))];
            tensor<fp16, [1, 6, 128]> linear_12_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_k_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_12_cast_fp16")];
            tensor<int32, [4]> concat_1 = const()[name = string("concat_1"), val = tensor<int32, [4]>([1, 6, 2, 64])];
            tensor<fp16, [1, 6, 2, 64]> x_25_cast_fp16 = reshape(shape = concat_1, x = linear_12_cast_fp16)[name = string("x_25_cast_fp16")];
            tensor<fp16, [1, 6, 2, 64]> var_672_cast_fp16 = abs(x = x_25_cast_fp16)[name = string("op_672_cast_fp16")];
            tensor<int32, [1]> var_674_axes_0 = const()[name = string("op_674_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 2, 1]> var_674_cast_fp16 = reduce_max(axes = var_674_axes_0, keep_dims = var_32, x = var_672_cast_fp16)[name = string("op_674_cast_fp16")];
            tensor<fp16, [1, 6, 2, 1]> s_21_cast_fp16 = maximum(x = var_674_cast_fp16, y = var_33_to_fp16)[name = string("s_21_cast_fp16")];
            tensor<fp16, [1, 6, 2, 64]> y_25_cast_fp16 = real_div(x = x_25_cast_fp16, y = s_21_cast_fp16)[name = string("y_25_cast_fp16")];
            fp16 var_34_promoted_10_to_fp16 = const()[name = string("op_34_promoted_10_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 2, 64]> var_677_cast_fp16 = pow(x = y_25_cast_fp16, y = var_34_promoted_10_to_fp16)[name = string("op_677_cast_fp16")];
            tensor<int32, [1]> var_679_axes_0 = const()[name = string("op_679_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_679_keep_dims_0 = const()[name = string("op_679_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 2, 1]> var_679_cast_fp16 = reduce_mean(axes = var_679_axes_0, keep_dims = var_679_keep_dims_0, x = var_677_cast_fp16)[name = string("op_679_cast_fp16")];
            fp32 var_680_epsilon_0 = const()[name = string("op_680_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 2, 1]> var_680_cast_fp16 = inverse(epsilon = var_680_epsilon_0, x = s_21_cast_fp16)[name = string("op_680_cast_fp16")];
            fp16 var_681_to_fp16 = const()[name = string("op_681_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 2, 1]> var_682_cast_fp16 = mul(x = var_680_cast_fp16, y = var_681_to_fp16)[name = string("op_682_cast_fp16")];
            tensor<fp16, [1, 6, 2, 1]> var_683_cast_fp16 = real_div(x = var_682_cast_fp16, y = s_21_cast_fp16)[name = string("op_683_cast_fp16")];
            tensor<fp16, [1, 6, 2, 1]> var_684_cast_fp16 = add(x = var_679_cast_fp16, y = var_683_cast_fp16)[name = string("op_684_cast_fp16")];
            fp32 var_685_epsilon_0 = const()[name = string("op_685_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 2, 1]> var_685_cast_fp16 = rsqrt(epsilon = var_685_epsilon_0, x = var_684_cast_fp16)[name = string("op_685_cast_fp16")];
            tensor<fp16, [1, 6, 2, 64]> var_686_cast_fp16 = mul(x = y_25_cast_fp16, y = var_685_cast_fp16)[name = string("op_686_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_2_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19090240)))];
            tensor<fp16, [1, 6, 2, 64]> var_687_cast_fp16 = mul(x = core_backbone_layers_2_self_attn_k_norm_weight_to_fp16, y = var_686_cast_fp16)[name = string("op_687_cast_fp16")];
            tensor<int32, [4]> new_5_perm_0 = const()[name = string("new_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [128, 512]> core_backbone_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19090432)))];
            tensor<fp16, [1, 6, 128]> linear_13_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_v_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_13_cast_fp16")];
            tensor<int32, [4]> concat_2 = const()[name = string("concat_2"), val = tensor<int32, [4]>([1, 6, 2, 64])];
            tensor<fp16, [1, 6, 2, 64]> var_692_cast_fp16 = reshape(shape = concat_2, x = linear_13_cast_fp16)[name = string("op_692_cast_fp16")];
            tensor<int32, [4]> new_7_perm_0 = const()[name = string("new_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_29_begin_0 = const()[name = string("input_29_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
            tensor<int32, [5]> input_29_end_0 = const()[name = string("input_29_end_0"), val = tensor<int32, [5]>([1, 1, 2, 29, 64])];
            tensor<bool, [5]> input_29_end_mask_0 = const()[name = string("input_29_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_29_squeeze_mask_0 = const()[name = string("input_29_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_29_cast_fp16 = slice_by_index(begin = input_29_begin_0, end = input_29_end_0, end_mask = input_29_end_mask_0, squeeze_mask = input_29_squeeze_mask_0, x = k)[name = string("input_29_cast_fp16")];
            tensor<int32, [8]> var_697_pad_0 = const()[name = string("op_697_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_697_mode_0 = const()[name = string("op_697_mode_0"), val = string("constant")];
            fp16 const_9_to_fp16 = const()[name = string("const_9_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 6, 64]> new_5_cast_fp16 = transpose(perm = new_5_perm_0, x = var_687_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [1, 2, 35, 64]> var_697_cast_fp16 = pad(constant_val = const_9_to_fp16, mode = var_697_mode_0, pad = var_697_pad_0, x = new_5_cast_fp16)[name = string("op_697_cast_fp16")];
            tensor<int32, [8]> var_699_pad_0 = const()[name = string("op_699_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 6, 0, 0])];
            string var_699_mode_0 = const()[name = string("op_699_mode_0"), val = string("constant")];
            fp16 const_10_to_fp16 = const()[name = string("const_10_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 35, 64]> var_699_cast_fp16 = pad(constant_val = const_10_to_fp16, mode = var_699_mode_0, pad = var_699_pad_0, x = input_29_cast_fp16)[name = string("op_699_cast_fp16")];
            tensor<fp16, [1, 2, 35, 64]> x_29_cast_fp16 = add(x = var_697_cast_fp16, y = var_699_cast_fp16)[name = string("x_29_cast_fp16")];
            tensor<int32, [5]> input_33_begin_0 = const()[name = string("input_33_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
            tensor<int32, [5]> input_33_end_0 = const()[name = string("input_33_end_0"), val = tensor<int32, [5]>([1, 1, 2, 29, 64])];
            tensor<bool, [5]> input_33_end_mask_0 = const()[name = string("input_33_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_33_squeeze_mask_0 = const()[name = string("input_33_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_33_cast_fp16 = slice_by_index(begin = input_33_begin_0, end = input_33_end_0, end_mask = input_33_end_mask_0, squeeze_mask = input_33_squeeze_mask_0, x = v)[name = string("input_33_cast_fp16")];
            tensor<int32, [8]> var_704_pad_0 = const()[name = string("op_704_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_704_mode_0 = const()[name = string("op_704_mode_0"), val = string("constant")];
            fp16 const_11_to_fp16 = const()[name = string("const_11_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 6, 64]> new_7_cast_fp16 = transpose(perm = new_7_perm_0, x = var_692_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [1, 2, 35, 64]> var_704_cast_fp16 = pad(constant_val = const_11_to_fp16, mode = var_704_mode_0, pad = var_704_pad_0, x = new_7_cast_fp16)[name = string("op_704_cast_fp16")];
            tensor<int32, [8]> var_706_pad_0 = const()[name = string("op_706_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 6, 0, 0])];
            string var_706_mode_0 = const()[name = string("op_706_mode_0"), val = string("constant")];
            fp16 const_12_to_fp16 = const()[name = string("const_12_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 35, 64]> var_706_cast_fp16 = pad(constant_val = const_12_to_fp16, mode = var_706_mode_0, pad = var_706_pad_0, x = input_33_cast_fp16)[name = string("op_706_cast_fp16")];
            tensor<fp16, [1, 2, 35, 64]> v_3_cast_fp16 = add(x = var_704_cast_fp16, y = var_706_cast_fp16)[name = string("v_3_cast_fp16")];
            tensor<fp16, [1, 6, 1, 64]> const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = tensor<fp16, [1, 6, 1, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19221568)))];
            tensor<fp16, [1, 6, 8, 64]> var_708_cast_fp16 = mul(x = var_665_cast_fp16, y = const_33_to_fp16)[name = string("op_708_cast_fp16")];
            tensor<int32, [2]> var_709_split_sizes_0 = const()[name = string("op_709_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 const_34 = const()[name = string("const_34"), val = int32(3)];
            tensor<fp16, [1, 6, 8, 32]> var_709_cast_fp16_0, tensor<fp16, [1, 6, 8, 32]> var_709_cast_fp16_1 = split(axis = const_34, split_sizes = var_709_split_sizes_0, x = var_665_cast_fp16)[name = string("op_709_cast_fp16")];
            fp16 const_13_promoted_to_fp16 = const()[name = string("const_13_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 6, 8, 32]> var_711_cast_fp16 = mul(x = var_709_cast_fp16_1, y = const_13_promoted_to_fp16)[name = string("op_711_cast_fp16")];
            bool var_713_interleave_0 = const()[name = string("op_713_interleave_0"), val = bool(false)];
            int32 const_35 = const()[name = string("const_35"), val = int32(3)];
            tensor<fp16, [1, 6, 8, 64]> var_713_cast_fp16 = concat(axis = const_35, interleave = var_713_interleave_0, values = (var_711_cast_fp16, var_709_cast_fp16_0))[name = string("op_713_cast_fp16")];
            tensor<fp16, [1, 6, 1, 64]> const_36_to_fp16 = const()[name = string("const_36_to_fp16"), val = tensor<fp16, [1, 6, 1, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19222400)))];
            tensor<fp16, [1, 6, 8, 64]> var_714_cast_fp16 = mul(x = var_713_cast_fp16, y = const_36_to_fp16)[name = string("op_714_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> q_1_cast_fp16 = add(x = var_708_cast_fp16, y = var_714_cast_fp16)[name = string("q_1_cast_fp16")];
            tensor<fp16, [1, 1, 35, 64]> const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = tensor<fp16, [1, 1, 35, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19223232)))];
            tensor<fp16, [1, 2, 35, 64]> var_716_cast_fp16 = mul(x = x_29_cast_fp16, y = const_1_to_fp16)[name = string("op_716_cast_fp16")];
            tensor<int32, [2]> var_717_split_sizes_0 = const()[name = string("op_717_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 var_717_axis_0 = const()[name = string("op_717_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 2, 35, 32]> var_717_cast_fp16_0, tensor<fp16, [1, 2, 35, 32]> var_717_cast_fp16_1 = split(axis = var_717_axis_0, split_sizes = var_717_split_sizes_0, x = x_29_cast_fp16)[name = string("op_717_cast_fp16")];
            fp16 const_14_promoted_to_fp16 = const()[name = string("const_14_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 2, 35, 32]> var_719_cast_fp16 = mul(x = var_717_cast_fp16_1, y = const_14_promoted_to_fp16)[name = string("op_719_cast_fp16")];
            bool var_721_interleave_0 = const()[name = string("op_721_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 2, 35, 64]> var_721_cast_fp16 = concat(axis = var_31, interleave = var_721_interleave_0, values = (var_719_cast_fp16, var_717_cast_fp16_0))[name = string("op_721_cast_fp16")];
            tensor<fp16, [1, 1, 35, 64]> const_2_to_fp16 = const()[name = string("const_2_to_fp16"), val = tensor<fp16, [1, 1, 35, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19227776)))];
            tensor<fp16, [1, 2, 35, 64]> var_722_cast_fp16 = mul(x = var_721_cast_fp16, y = const_2_to_fp16)[name = string("op_722_cast_fp16")];
            tensor<fp16, [1, 2, 35, 64]> k_3_cast_fp16 = add(x = var_716_cast_fp16, y = var_722_cast_fp16)[name = string("k_3_cast_fp16")];
            tensor<int32, [1]> var_724_axes_0 = const()[name = string("op_724_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 35, 64]> var_724_cast_fp16 = expand_dims(axes = var_724_axes_0, x = k_3_cast_fp16)[name = string("op_724_cast_fp16")];
            tensor<int32, [5]> var_726_reps_0 = const()[name = string("op_726_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 35, 64]> var_726_cast_fp16 = tile(reps = var_726_reps_0, x = var_724_cast_fp16)[name = string("op_726_cast_fp16")];
            tensor<int32, [4]> concat_4 = const()[name = string("concat_4"), val = tensor<int32, [4]>([1, 8, 35, 64])];
            tensor<fp16, [1, 8, 35, 64]> k_5_cast_fp16 = reshape(shape = concat_4, x = var_726_cast_fp16)[name = string("k_5_cast_fp16")];
            tensor<int32, [1]> var_728_axes_0 = const()[name = string("op_728_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 35, 64]> var_728_cast_fp16 = expand_dims(axes = var_728_axes_0, x = v_3_cast_fp16)[name = string("op_728_cast_fp16")];
            tensor<int32, [5]> var_730_reps_0 = const()[name = string("op_730_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 35, 64]> var_730_cast_fp16 = tile(reps = var_730_reps_0, x = var_728_cast_fp16)[name = string("op_730_cast_fp16")];
            tensor<int32, [4]> concat_6 = const()[name = string("concat_6"), val = tensor<int32, [4]>([1, 8, 35, 64])];
            tensor<fp16, [1, 8, 35, 64]> v_5_cast_fp16 = reshape(shape = concat_6, x = var_730_cast_fp16)[name = string("v_5_cast_fp16")];
            tensor<int32, [4]> transpose_0_perm_0 = const()[name = string("transpose_0_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_733_transpose_x_1 = const()[name = string("op_733_transpose_x_1"), val = bool(false)];
            bool var_733_transpose_y_1 = const()[name = string("op_733_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 6, 64]> transpose_0_cast_fp16 = transpose(perm = transpose_0_perm_0, x = q_1_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [1, 8, 6, 35]> var_733_cast_fp16 = matmul(transpose_x = var_733_transpose_x_1, transpose_y = var_733_transpose_y_1, x = transpose_0_cast_fp16, y = k_5_cast_fp16)[name = string("op_733_cast_fp16")];
            fp16 var_734_to_fp16 = const()[name = string("op_734_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 8, 6, 35]> s_23_cast_fp16 = mul(x = var_733_cast_fp16, y = var_734_to_fp16)[name = string("s_23_cast_fp16")];
            tensor<fp16, [1, 8, 6, 35]> x_31_cast_fp16 = add(x = s_23_cast_fp16, y = mask)[name = string("x_31_cast_fp16")];
            tensor<int32, [1]> var_738_axes_0 = const()[name = string("op_738_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 6, 1]> var_738_cast_fp16 = reduce_max(axes = var_738_axes_0, keep_dims = var_32, x = x_31_cast_fp16)[name = string("op_738_cast_fp16")];
            tensor<fp16, [1, 8, 6, 35]> x_33_cast_fp16 = sub(x = x_31_cast_fp16, y = var_738_cast_fp16)[name = string("x_33_cast_fp16")];
            tensor<fp16, [1, 8, 6, 35]> x_35_cast_fp16 = exp(x = x_33_cast_fp16)[name = string("x_35_cast_fp16")];
            tensor<int32, [1]> var_742_axes_0 = const()[name = string("op_742_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_742_keep_dims_0 = const()[name = string("op_742_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 6, 1]> var_742_cast_fp16 = reduce_sum(axes = var_742_axes_0, keep_dims = var_742_keep_dims_0, x = x_35_cast_fp16)[name = string("op_742_cast_fp16")];
            tensor<fp16, [1, 8, 6, 35]> var_743_cast_fp16 = real_div(x = x_35_cast_fp16, y = var_742_cast_fp16)[name = string("op_743_cast_fp16")];
            bool o_1_transpose_x_0 = const()[name = string("o_1_transpose_x_0"), val = bool(false)];
            bool o_1_transpose_y_0 = const()[name = string("o_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 6, 64]> o_1_cast_fp16 = matmul(transpose_x = o_1_transpose_x_0, transpose_y = o_1_transpose_y_0, x = var_743_cast_fp16, y = v_5_cast_fp16)[name = string("o_1_cast_fp16")];
            tensor<int32, [4]> o_3_perm_0 = const()[name = string("o_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [8, 512]> core_backbone_layers_2_self_attn_gate_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_gate_proj_weight_to_fp16"), val = tensor<fp16, [8, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19232320)))];
            tensor<fp16, [8]> linear_14_bias_0_to_fp16 = const()[name = string("linear_14_bias_0_to_fp16"), val = tensor<fp16, [8]>([0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0])];
            tensor<fp16, [1, 6, 8]> linear_14_cast_fp16 = linear(bias = linear_14_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_gate_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_14_cast_fp16")];
            fp16 var_749_to_fp16 = const()[name = string("op_749_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 6, 8]> var_750_cast_fp16 = mul(x = linear_14_cast_fp16, y = var_749_to_fp16)[name = string("op_750_cast_fp16")];
            tensor<fp16, [1, 6, 8]> var_751_cast_fp16 = tanh(x = var_750_cast_fp16)[name = string("op_751_cast_fp16")];
            fp16 var_752_to_fp16 = const()[name = string("op_752_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 6, 8]> var_753_cast_fp16 = add(x = var_751_cast_fp16, y = var_752_to_fp16)[name = string("op_753_cast_fp16")];
            fp16 var_754_to_fp16 = const()[name = string("op_754_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 6, 8]> var_755_cast_fp16 = mul(x = var_753_cast_fp16, y = var_754_to_fp16)[name = string("op_755_cast_fp16")];
            tensor<int32, [1]> var_756_axes_0 = const()[name = string("op_756_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 8, 1]> var_756_cast_fp16 = expand_dims(axes = var_756_axes_0, x = var_755_cast_fp16)[name = string("op_756_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> o_3_cast_fp16 = transpose(perm = o_3_perm_0, x = o_1_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [1, 6, 8, 64]> o_5_cast_fp16 = mul(x = o_3_cast_fp16, y = var_756_cast_fp16)[name = string("o_5_cast_fp16")];
            tensor<int32, [3]> concat_7 = const()[name = string("concat_7"), val = tensor<int32, [3]>([1, 6, 512])];
            tensor<fp16, [1, 6, 512]> input_35_cast_fp16 = reshape(shape = concat_7, x = o_5_cast_fp16)[name = string("input_35_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19240576)))];
            tensor<fp16, [1, 6, 512]> linear_15_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_o_proj_weight_to_fp16, x = input_35_cast_fp16)[name = string("linear_15_cast_fp16")];
            tensor<int32, [4]> var_763_begin_0 = const()[name = string("op_763_begin_0"), val = tensor<int32, [4]>([0, 0, 6, 0])];
            tensor<int32, [4]> var_763_end_0 = const()[name = string("op_763_end_0"), val = tensor<int32, [4]>([1, 2, 35, 64])];
            tensor<bool, [4]> var_763_end_mask_0 = const()[name = string("op_763_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_763_cast_fp16 = slice_by_index(begin = var_763_begin_0, end = var_763_end_0, end_mask = var_763_end_mask_0, x = x_29_cast_fp16)[name = string("op_763_cast_fp16")];
            tensor<int32, [4]> var_766_begin_0 = const()[name = string("op_766_begin_0"), val = tensor<int32, [4]>([0, 0, 6, 0])];
            tensor<int32, [4]> var_766_end_0 = const()[name = string("op_766_end_0"), val = tensor<int32, [4]>([1, 2, 35, 64])];
            tensor<bool, [4]> var_766_end_mask_0 = const()[name = string("op_766_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_766_cast_fp16 = slice_by_index(begin = var_766_begin_0, end = var_766_end_0, end_mask = var_766_end_mask_0, x = v_3_cast_fp16)[name = string("op_766_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_768_cast_fp16 = abs(x = linear_15_cast_fp16)[name = string("op_768_cast_fp16")];
            tensor<int32, [1]> var_770_axes_0 = const()[name = string("op_770_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_770_cast_fp16 = reduce_max(axes = var_770_axes_0, keep_dims = var_32, x = var_768_cast_fp16)[name = string("op_770_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_25_cast_fp16 = maximum(x = var_770_cast_fp16, y = var_33_to_fp16)[name = string("s_25_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_27_cast_fp16 = real_div(x = linear_15_cast_fp16, y = s_25_cast_fp16)[name = string("y_27_cast_fp16")];
            fp16 var_34_promoted_11_to_fp16 = const()[name = string("op_34_promoted_11_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_773_cast_fp16 = pow(x = y_27_cast_fp16, y = var_34_promoted_11_to_fp16)[name = string("op_773_cast_fp16")];
            tensor<int32, [1]> var_775_axes_0 = const()[name = string("op_775_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_775_keep_dims_0 = const()[name = string("op_775_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_775_cast_fp16 = reduce_mean(axes = var_775_axes_0, keep_dims = var_775_keep_dims_0, x = var_773_cast_fp16)[name = string("op_775_cast_fp16")];
            fp32 var_776_epsilon_0 = const()[name = string("op_776_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_776_cast_fp16 = inverse(epsilon = var_776_epsilon_0, x = s_25_cast_fp16)[name = string("op_776_cast_fp16")];
            fp16 var_777_to_fp16 = const()[name = string("op_777_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_778_cast_fp16 = mul(x = var_776_cast_fp16, y = var_777_to_fp16)[name = string("op_778_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_779_cast_fp16 = real_div(x = var_778_cast_fp16, y = s_25_cast_fp16)[name = string("op_779_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_780_cast_fp16 = add(x = var_775_cast_fp16, y = var_779_cast_fp16)[name = string("op_780_cast_fp16")];
            fp32 var_781_epsilon_0 = const()[name = string("op_781_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_781_cast_fp16 = rsqrt(epsilon = var_781_epsilon_0, x = var_780_cast_fp16)[name = string("op_781_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_782_cast_fp16 = mul(x = y_27_cast_fp16, y = var_781_cast_fp16)[name = string("op_782_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_2_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19764928)))];
            tensor<fp16, [1, 6, 512]> a_5_cast_fp16 = mul(x = core_backbone_layers_2_post_operator_norm_weight_to_fp16, y = var_782_cast_fp16)[name = string("a_5_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_41_cast_fp16 = add(x = x_21_cast_fp16, y = a_5_cast_fp16)[name = string("x_41_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_786_cast_fp16 = abs(x = x_41_cast_fp16)[name = string("op_786_cast_fp16")];
            tensor<int32, [1]> var_788_axes_0 = const()[name = string("op_788_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_788_cast_fp16 = reduce_max(axes = var_788_axes_0, keep_dims = var_32, x = var_786_cast_fp16)[name = string("op_788_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_27_cast_fp16 = maximum(x = var_788_cast_fp16, y = var_33_to_fp16)[name = string("s_27_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_29_cast_fp16 = real_div(x = x_41_cast_fp16, y = s_27_cast_fp16)[name = string("y_29_cast_fp16")];
            fp16 var_34_promoted_12_to_fp16 = const()[name = string("op_34_promoted_12_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_791_cast_fp16 = pow(x = y_29_cast_fp16, y = var_34_promoted_12_to_fp16)[name = string("op_791_cast_fp16")];
            tensor<int32, [1]> var_793_axes_0 = const()[name = string("op_793_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_793_keep_dims_0 = const()[name = string("op_793_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_793_cast_fp16 = reduce_mean(axes = var_793_axes_0, keep_dims = var_793_keep_dims_0, x = var_791_cast_fp16)[name = string("op_793_cast_fp16")];
            fp32 var_794_epsilon_0 = const()[name = string("op_794_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_794_cast_fp16 = inverse(epsilon = var_794_epsilon_0, x = s_27_cast_fp16)[name = string("op_794_cast_fp16")];
            fp16 var_795_to_fp16 = const()[name = string("op_795_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_796_cast_fp16 = mul(x = var_794_cast_fp16, y = var_795_to_fp16)[name = string("op_796_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_797_cast_fp16 = real_div(x = var_796_cast_fp16, y = s_27_cast_fp16)[name = string("op_797_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_798_cast_fp16 = add(x = var_793_cast_fp16, y = var_797_cast_fp16)[name = string("op_798_cast_fp16")];
            fp32 var_799_epsilon_0 = const()[name = string("op_799_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_799_cast_fp16 = rsqrt(epsilon = var_799_epsilon_0, x = var_798_cast_fp16)[name = string("op_799_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_800_cast_fp16 = mul(x = y_29_cast_fp16, y = var_799_cast_fp16)[name = string("op_800_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_2_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19766016)))];
            tensor<fp16, [1, 6, 512]> input_37_cast_fp16 = mul(x = core_backbone_layers_2_ffn_norm_weight_to_fp16, y = var_800_cast_fp16)[name = string("input_37_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_2_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19767104)))];
            tensor<fp16, [1, 6, 2304]> linear_16_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w1_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_16_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_2_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22126464)))];
            tensor<fp16, [1, 6, 2304]> linear_17_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w3_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_17_cast_fp16")];
            fp16 var_809_to_fp16 = const()[name = string("op_809_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 6, 2304]> var_810_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_809_to_fp16)[name = string("op_810_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_813_cast_fp16 = tanh(x = var_810_cast_fp16)[name = string("op_813_cast_fp16")];
            fp16 var_814_to_fp16 = const()[name = string("op_814_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 6, 2304]> var_815_cast_fp16 = add(x = var_813_cast_fp16, y = var_814_to_fp16)[name = string("op_815_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_816_cast_fp16 = mul(x = var_810_cast_fp16, y = var_815_cast_fp16)[name = string("op_816_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> input_39_cast_fp16 = mul(x = var_816_cast_fp16, y = linear_17_cast_fp16)[name = string("input_39_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_2_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24485824)))];
            tensor<fp16, [1, 6, 512]> linear_18_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w2_weight_to_fp16, x = input_39_cast_fp16)[name = string("linear_18_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_821_cast_fp16 = abs(x = linear_18_cast_fp16)[name = string("op_821_cast_fp16")];
            tensor<int32, [1]> var_823_axes_0 = const()[name = string("op_823_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_823_cast_fp16 = reduce_max(axes = var_823_axes_0, keep_dims = var_32, x = var_821_cast_fp16)[name = string("op_823_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_29_cast_fp16 = maximum(x = var_823_cast_fp16, y = var_33_to_fp16)[name = string("s_29_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_31_cast_fp16 = real_div(x = linear_18_cast_fp16, y = s_29_cast_fp16)[name = string("y_31_cast_fp16")];
            fp16 var_34_promoted_13_to_fp16 = const()[name = string("op_34_promoted_13_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_826_cast_fp16 = pow(x = y_31_cast_fp16, y = var_34_promoted_13_to_fp16)[name = string("op_826_cast_fp16")];
            tensor<int32, [1]> var_828_axes_0 = const()[name = string("op_828_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_828_keep_dims_0 = const()[name = string("op_828_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_828_cast_fp16 = reduce_mean(axes = var_828_axes_0, keep_dims = var_828_keep_dims_0, x = var_826_cast_fp16)[name = string("op_828_cast_fp16")];
            fp32 var_829_epsilon_0 = const()[name = string("op_829_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_829_cast_fp16 = inverse(epsilon = var_829_epsilon_0, x = s_29_cast_fp16)[name = string("op_829_cast_fp16")];
            fp16 var_830_to_fp16 = const()[name = string("op_830_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_831_cast_fp16 = mul(x = var_829_cast_fp16, y = var_830_to_fp16)[name = string("op_831_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_832_cast_fp16 = real_div(x = var_831_cast_fp16, y = s_29_cast_fp16)[name = string("op_832_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_833_cast_fp16 = add(x = var_828_cast_fp16, y = var_832_cast_fp16)[name = string("op_833_cast_fp16")];
            fp32 var_834_epsilon_0 = const()[name = string("op_834_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_834_cast_fp16 = rsqrt(epsilon = var_834_epsilon_0, x = var_833_cast_fp16)[name = string("op_834_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_835_cast_fp16 = mul(x = y_31_cast_fp16, y = var_834_cast_fp16)[name = string("op_835_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_2_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26845184)))];
            tensor<fp16, [1, 6, 512]> f_5_cast_fp16 = mul(x = core_backbone_layers_2_post_ffn_norm_weight_to_fp16, y = var_835_cast_fp16)[name = string("f_5_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_47_cast_fp16 = add(x = x_41_cast_fp16, y = f_5_cast_fp16)[name = string("x_47_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_839_cast_fp16 = abs(x = x_47_cast_fp16)[name = string("op_839_cast_fp16")];
            tensor<int32, [1]> var_841_axes_0 = const()[name = string("op_841_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_841_cast_fp16 = reduce_max(axes = var_841_axes_0, keep_dims = var_32, x = var_839_cast_fp16)[name = string("op_841_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_31_cast_fp16 = maximum(x = var_841_cast_fp16, y = var_33_to_fp16)[name = string("s_31_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_33_cast_fp16 = real_div(x = x_47_cast_fp16, y = s_31_cast_fp16)[name = string("y_33_cast_fp16")];
            fp16 var_34_promoted_14_to_fp16 = const()[name = string("op_34_promoted_14_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_844_cast_fp16 = pow(x = y_33_cast_fp16, y = var_34_promoted_14_to_fp16)[name = string("op_844_cast_fp16")];
            tensor<int32, [1]> var_846_axes_0 = const()[name = string("op_846_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_846_keep_dims_0 = const()[name = string("op_846_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_846_cast_fp16 = reduce_mean(axes = var_846_axes_0, keep_dims = var_846_keep_dims_0, x = var_844_cast_fp16)[name = string("op_846_cast_fp16")];
            fp32 var_847_epsilon_0 = const()[name = string("op_847_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_847_cast_fp16 = inverse(epsilon = var_847_epsilon_0, x = s_31_cast_fp16)[name = string("op_847_cast_fp16")];
            fp16 var_848_to_fp16 = const()[name = string("op_848_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_849_cast_fp16 = mul(x = var_847_cast_fp16, y = var_848_to_fp16)[name = string("op_849_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_850_cast_fp16 = real_div(x = var_849_cast_fp16, y = s_31_cast_fp16)[name = string("op_850_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_851_cast_fp16 = add(x = var_846_cast_fp16, y = var_850_cast_fp16)[name = string("op_851_cast_fp16")];
            fp32 var_852_epsilon_0 = const()[name = string("op_852_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_852_cast_fp16 = rsqrt(epsilon = var_852_epsilon_0, x = var_851_cast_fp16)[name = string("op_852_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_853_cast_fp16 = mul(x = y_33_cast_fp16, y = var_852_cast_fp16)[name = string("op_853_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_3_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26846272)))];
            tensor<fp16, [1, 6, 512]> input_41_cast_fp16 = mul(x = core_backbone_layers_3_operator_norm_weight_to_fp16, y = var_853_cast_fp16)[name = string("input_41_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_3_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26847360)))];
            tensor<fp16, [1, 6, 1536]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_3_conv_in_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("linear_19_cast_fp16")];
            tensor<int32, [3]> bcx_5_perm_0 = const()[name = string("bcx_5_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_11_begin_0 = const()[name = string("b_11_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_11_end_0 = const()[name = string("b_11_end_0"), val = tensor<int32, [3]>([1, 512, 6])];
            tensor<bool, [3]> b_11_end_mask_0 = const()[name = string("b_11_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 6]> bcx_5_cast_fp16 = transpose(perm = bcx_5_perm_0, x = linear_19_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [1, 512, 6]> b_11_cast_fp16 = slice_by_index(begin = b_11_begin_0, end = b_11_end_0, end_mask = b_11_end_mask_0, x = bcx_5_cast_fp16)[name = string("b_11_cast_fp16")];
            tensor<int32, [3]> c_5_begin_0 = const()[name = string("c_5_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_5_end_0 = const()[name = string("c_5_end_0"), val = tensor<int32, [3]>([1, 1024, 6])];
            tensor<bool, [3]> c_5_end_mask_0 = const()[name = string("c_5_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 6]> c_5_cast_fp16 = slice_by_index(begin = c_5_begin_0, end = c_5_end_0, end_mask = c_5_end_mask_0, x = bcx_5_cast_fp16)[name = string("c_5_cast_fp16")];
            tensor<int32, [3]> xx_5_begin_0 = const()[name = string("xx_5_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_5_end_0 = const()[name = string("xx_5_end_0"), val = tensor<int32, [3]>([1, 1536, 6])];
            tensor<bool, [3]> xx_5_end_mask_0 = const()[name = string("xx_5_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> xx_5_cast_fp16 = slice_by_index(begin = xx_5_begin_0, end = xx_5_end_0, end_mask = xx_5_end_mask_0, x = bcx_5_cast_fp16)[name = string("xx_5_cast_fp16")];
            tensor<fp16, [1, 512, 6]> new_9_cast_fp16 = mul(x = b_11_cast_fp16, y = xx_5_cast_fp16)[name = string("new_9_cast_fp16")];
            tensor<int32, [4]> input_45_begin_0 = const()[name = string("input_45_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])];
            tensor<int32, [4]> input_45_end_0 = const()[name = string("input_45_end_0"), val = tensor<int32, [4]>([3, 1, 512, 6])];
            tensor<bool, [4]> input_45_end_mask_0 = const()[name = string("input_45_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_45_squeeze_mask_0 = const()[name = string("input_45_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_45_cast_fp16 = slice_by_index(begin = input_45_begin_0, end = input_45_end_0, end_mask = input_45_end_mask_0, squeeze_mask = input_45_squeeze_mask_0, x = conv)[name = string("input_45_cast_fp16")];
            tensor<int32, [6]> var_868_pad_0 = const()[name = string("op_868_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_868_mode_0 = const()[name = string("op_868_mode_0"), val = string("constant")];
            fp16 const_15_to_fp16 = const()[name = string("const_15_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 12]> var_868_cast_fp16 = pad(constant_val = const_15_to_fp16, mode = var_868_mode_0, pad = var_868_pad_0, x = new_9_cast_fp16)[name = string("op_868_cast_fp16")];
            tensor<int32, [6]> var_870_pad_0 = const()[name = string("op_870_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 6])];
            string var_870_mode_0 = const()[name = string("op_870_mode_0"), val = string("constant")];
            fp16 const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 12]> var_870_cast_fp16 = pad(constant_val = const_16_to_fp16, mode = var_870_mode_0, pad = var_870_pad_0, x = input_45_cast_fp16)[name = string("op_870_cast_fp16")];
            tensor<fp16, [1, 512, 12]> win_9_cast_fp16 = add(x = var_868_cast_fp16, y = var_870_cast_fp16)[name = string("win_9_cast_fp16")];
            tensor<int32, [3]> var_872_begin_0 = const()[name = string("op_872_begin_0"), val = tensor<int32, [3]>([0, 0, 6])];
            tensor<int32, [3]> var_872_end_0 = const()[name = string("op_872_end_0"), val = tensor<int32, [3]>([1, 512, 12])];
            tensor<bool, [3]> var_872_end_mask_0 = const()[name = string("op_872_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_872_cast_fp16 = slice_by_index(begin = var_872_begin_0, end = var_872_end_0, end_mask = var_872_end_mask_0, x = win_9_cast_fp16)[name = string("op_872_cast_fp16")];
            string var_878_pad_type_0 = const()[name = string("op_878_pad_type_0"), val = string("valid")];
            int32 var_878_groups_0 = const()[name = string("op_878_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_878_strides_0 = const()[name = string("op_878_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_878_pad_0 = const()[name = string("op_878_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_878_dilations_0 = const()[name = string("op_878_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_3_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28420288)))];
            tensor<fp16, [1, 512, 6]> var_878_cast_fp16 = conv(dilations = var_878_dilations_0, groups = var_878_groups_0, pad = var_878_pad_0, pad_type = var_878_pad_type_0, strides = var_878_strides_0, weight = core_backbone_layers_3_conv_conv_weight_to_fp16, x = win_9_cast_fp16)[name = string("op_878_cast_fp16")];
            tensor<fp16, [1, 512, 6]> var_880_cast_fp16 = mul(x = c_5_cast_fp16, y = var_878_cast_fp16)[name = string("op_880_cast_fp16")];
            tensor<int32, [3]> input_47_perm_0 = const()[name = string("input_47_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_3_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28427520)))];
            tensor<fp16, [1, 6, 512]> input_47_cast_fp16 = transpose(perm = input_47_perm_0, x = var_880_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [1, 6, 512]> linear_20_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_3_conv_out_proj_weight_to_fp16, x = input_47_cast_fp16)[name = string("linear_20_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_885_cast_fp16 = abs(x = linear_20_cast_fp16)[name = string("op_885_cast_fp16")];
            tensor<int32, [1]> var_887_axes_0 = const()[name = string("op_887_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_887_cast_fp16 = reduce_max(axes = var_887_axes_0, keep_dims = var_32, x = var_885_cast_fp16)[name = string("op_887_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_33_cast_fp16 = maximum(x = var_887_cast_fp16, y = var_33_to_fp16)[name = string("s_33_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_37_cast_fp16 = real_div(x = linear_20_cast_fp16, y = s_33_cast_fp16)[name = string("y_37_cast_fp16")];
            fp16 var_34_promoted_15_to_fp16 = const()[name = string("op_34_promoted_15_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_890_cast_fp16 = pow(x = y_37_cast_fp16, y = var_34_promoted_15_to_fp16)[name = string("op_890_cast_fp16")];
            tensor<int32, [1]> var_892_axes_0 = const()[name = string("op_892_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_892_keep_dims_0 = const()[name = string("op_892_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_892_cast_fp16 = reduce_mean(axes = var_892_axes_0, keep_dims = var_892_keep_dims_0, x = var_890_cast_fp16)[name = string("op_892_cast_fp16")];
            fp32 var_893_epsilon_0 = const()[name = string("op_893_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_893_cast_fp16 = inverse(epsilon = var_893_epsilon_0, x = s_33_cast_fp16)[name = string("op_893_cast_fp16")];
            fp16 var_894_to_fp16 = const()[name = string("op_894_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_895_cast_fp16 = mul(x = var_893_cast_fp16, y = var_894_to_fp16)[name = string("op_895_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_896_cast_fp16 = real_div(x = var_895_cast_fp16, y = s_33_cast_fp16)[name = string("op_896_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_897_cast_fp16 = add(x = var_892_cast_fp16, y = var_896_cast_fp16)[name = string("op_897_cast_fp16")];
            fp32 var_898_epsilon_0 = const()[name = string("op_898_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_898_cast_fp16 = rsqrt(epsilon = var_898_epsilon_0, x = var_897_cast_fp16)[name = string("op_898_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_899_cast_fp16 = mul(x = y_37_cast_fp16, y = var_898_cast_fp16)[name = string("op_899_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_3_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28951872)))];
            tensor<fp16, [1, 6, 512]> a_7_cast_fp16 = mul(x = core_backbone_layers_3_post_operator_norm_weight_to_fp16, y = var_899_cast_fp16)[name = string("a_7_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_51_cast_fp16 = add(x = x_47_cast_fp16, y = a_7_cast_fp16)[name = string("x_51_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_903_cast_fp16 = abs(x = x_51_cast_fp16)[name = string("op_903_cast_fp16")];
            tensor<int32, [1]> var_905_axes_0 = const()[name = string("op_905_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_905_cast_fp16 = reduce_max(axes = var_905_axes_0, keep_dims = var_32, x = var_903_cast_fp16)[name = string("op_905_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_35_cast_fp16 = maximum(x = var_905_cast_fp16, y = var_33_to_fp16)[name = string("s_35_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_39_cast_fp16 = real_div(x = x_51_cast_fp16, y = s_35_cast_fp16)[name = string("y_39_cast_fp16")];
            fp16 var_34_promoted_16_to_fp16 = const()[name = string("op_34_promoted_16_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_908_cast_fp16 = pow(x = y_39_cast_fp16, y = var_34_promoted_16_to_fp16)[name = string("op_908_cast_fp16")];
            tensor<int32, [1]> var_910_axes_0 = const()[name = string("op_910_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_910_keep_dims_0 = const()[name = string("op_910_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_910_cast_fp16 = reduce_mean(axes = var_910_axes_0, keep_dims = var_910_keep_dims_0, x = var_908_cast_fp16)[name = string("op_910_cast_fp16")];
            fp32 var_911_epsilon_0 = const()[name = string("op_911_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_911_cast_fp16 = inverse(epsilon = var_911_epsilon_0, x = s_35_cast_fp16)[name = string("op_911_cast_fp16")];
            fp16 var_912_to_fp16 = const()[name = string("op_912_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_913_cast_fp16 = mul(x = var_911_cast_fp16, y = var_912_to_fp16)[name = string("op_913_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_914_cast_fp16 = real_div(x = var_913_cast_fp16, y = s_35_cast_fp16)[name = string("op_914_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_915_cast_fp16 = add(x = var_910_cast_fp16, y = var_914_cast_fp16)[name = string("op_915_cast_fp16")];
            fp32 var_916_epsilon_0 = const()[name = string("op_916_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_916_cast_fp16 = rsqrt(epsilon = var_916_epsilon_0, x = var_915_cast_fp16)[name = string("op_916_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_917_cast_fp16 = mul(x = y_39_cast_fp16, y = var_916_cast_fp16)[name = string("op_917_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_3_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28952960)))];
            tensor<fp16, [1, 6, 512]> input_49_cast_fp16 = mul(x = core_backbone_layers_3_ffn_norm_weight_to_fp16, y = var_917_cast_fp16)[name = string("input_49_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_3_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28954048)))];
            tensor<fp16, [1, 6, 2304]> linear_21_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w1_weight_to_fp16, x = input_49_cast_fp16)[name = string("linear_21_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_3_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31313408)))];
            tensor<fp16, [1, 6, 2304]> linear_22_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w3_weight_to_fp16, x = input_49_cast_fp16)[name = string("linear_22_cast_fp16")];
            fp16 var_926_to_fp16 = const()[name = string("op_926_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 6, 2304]> var_927_cast_fp16 = mul(x = linear_21_cast_fp16, y = var_926_to_fp16)[name = string("op_927_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_930_cast_fp16 = tanh(x = var_927_cast_fp16)[name = string("op_930_cast_fp16")];
            fp16 var_931_to_fp16 = const()[name = string("op_931_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 6, 2304]> var_932_cast_fp16 = add(x = var_930_cast_fp16, y = var_931_to_fp16)[name = string("op_932_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_933_cast_fp16 = mul(x = var_927_cast_fp16, y = var_932_cast_fp16)[name = string("op_933_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> input_51_cast_fp16 = mul(x = var_933_cast_fp16, y = linear_22_cast_fp16)[name = string("input_51_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_3_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33672768)))];
            tensor<fp16, [1, 6, 512]> linear_23_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w2_weight_to_fp16, x = input_51_cast_fp16)[name = string("linear_23_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_938_cast_fp16 = abs(x = linear_23_cast_fp16)[name = string("op_938_cast_fp16")];
            tensor<int32, [1]> var_940_axes_0 = const()[name = string("op_940_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_940_cast_fp16 = reduce_max(axes = var_940_axes_0, keep_dims = var_32, x = var_938_cast_fp16)[name = string("op_940_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_37_cast_fp16 = maximum(x = var_940_cast_fp16, y = var_33_to_fp16)[name = string("s_37_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_41_cast_fp16 = real_div(x = linear_23_cast_fp16, y = s_37_cast_fp16)[name = string("y_41_cast_fp16")];
            fp16 var_34_promoted_17_to_fp16 = const()[name = string("op_34_promoted_17_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_943_cast_fp16 = pow(x = y_41_cast_fp16, y = var_34_promoted_17_to_fp16)[name = string("op_943_cast_fp16")];
            tensor<int32, [1]> var_945_axes_0 = const()[name = string("op_945_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_945_keep_dims_0 = const()[name = string("op_945_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_945_cast_fp16 = reduce_mean(axes = var_945_axes_0, keep_dims = var_945_keep_dims_0, x = var_943_cast_fp16)[name = string("op_945_cast_fp16")];
            fp32 var_946_epsilon_0 = const()[name = string("op_946_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_946_cast_fp16 = inverse(epsilon = var_946_epsilon_0, x = s_37_cast_fp16)[name = string("op_946_cast_fp16")];
            fp16 var_947_to_fp16 = const()[name = string("op_947_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_948_cast_fp16 = mul(x = var_946_cast_fp16, y = var_947_to_fp16)[name = string("op_948_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_949_cast_fp16 = real_div(x = var_948_cast_fp16, y = s_37_cast_fp16)[name = string("op_949_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_950_cast_fp16 = add(x = var_945_cast_fp16, y = var_949_cast_fp16)[name = string("op_950_cast_fp16")];
            fp32 var_951_epsilon_0 = const()[name = string("op_951_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_951_cast_fp16 = rsqrt(epsilon = var_951_epsilon_0, x = var_950_cast_fp16)[name = string("op_951_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_952_cast_fp16 = mul(x = y_41_cast_fp16, y = var_951_cast_fp16)[name = string("op_952_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_3_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36032128)))];
            tensor<fp16, [1, 6, 512]> f_7_cast_fp16 = mul(x = core_backbone_layers_3_post_ffn_norm_weight_to_fp16, y = var_952_cast_fp16)[name = string("f_7_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_57_cast_fp16 = add(x = x_51_cast_fp16, y = f_7_cast_fp16)[name = string("x_57_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_956_cast_fp16 = abs(x = x_57_cast_fp16)[name = string("op_956_cast_fp16")];
            tensor<int32, [1]> var_958_axes_0 = const()[name = string("op_958_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_958_cast_fp16 = reduce_max(axes = var_958_axes_0, keep_dims = var_32, x = var_956_cast_fp16)[name = string("op_958_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_39_cast_fp16 = maximum(x = var_958_cast_fp16, y = var_33_to_fp16)[name = string("s_39_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_43_cast_fp16 = real_div(x = x_57_cast_fp16, y = s_39_cast_fp16)[name = string("y_43_cast_fp16")];
            fp16 var_34_promoted_18_to_fp16 = const()[name = string("op_34_promoted_18_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_961_cast_fp16 = pow(x = y_43_cast_fp16, y = var_34_promoted_18_to_fp16)[name = string("op_961_cast_fp16")];
            tensor<int32, [1]> var_963_axes_0 = const()[name = string("op_963_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_963_keep_dims_0 = const()[name = string("op_963_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_963_cast_fp16 = reduce_mean(axes = var_963_axes_0, keep_dims = var_963_keep_dims_0, x = var_961_cast_fp16)[name = string("op_963_cast_fp16")];
            fp32 var_964_epsilon_0 = const()[name = string("op_964_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_964_cast_fp16 = inverse(epsilon = var_964_epsilon_0, x = s_39_cast_fp16)[name = string("op_964_cast_fp16")];
            fp16 var_965_to_fp16 = const()[name = string("op_965_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_966_cast_fp16 = mul(x = var_964_cast_fp16, y = var_965_to_fp16)[name = string("op_966_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_967_cast_fp16 = real_div(x = var_966_cast_fp16, y = s_39_cast_fp16)[name = string("op_967_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_968_cast_fp16 = add(x = var_963_cast_fp16, y = var_967_cast_fp16)[name = string("op_968_cast_fp16")];
            fp32 var_969_epsilon_0 = const()[name = string("op_969_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_969_cast_fp16 = rsqrt(epsilon = var_969_epsilon_0, x = var_968_cast_fp16)[name = string("op_969_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_970_cast_fp16 = mul(x = y_43_cast_fp16, y = var_969_cast_fp16)[name = string("op_970_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_4_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36033216)))];
            tensor<fp16, [1, 6, 512]> input_53_cast_fp16 = mul(x = core_backbone_layers_4_operator_norm_weight_to_fp16, y = var_970_cast_fp16)[name = string("input_53_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36034304)))];
            tensor<fp16, [1, 6, 512]> linear_24_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_q_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_24_cast_fp16")];
            tensor<int32, [4]> concat_8 = const()[name = string("concat_8"), val = tensor<int32, [4]>([1, 6, 8, 64])];
            tensor<fp16, [1, 6, 8, 64]> x_59_cast_fp16 = reshape(shape = concat_8, x = linear_24_cast_fp16)[name = string("x_59_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> var_977_cast_fp16 = abs(x = x_59_cast_fp16)[name = string("op_977_cast_fp16")];
            tensor<int32, [1]> var_979_axes_0 = const()[name = string("op_979_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 8, 1]> var_979_cast_fp16 = reduce_max(axes = var_979_axes_0, keep_dims = var_32, x = var_977_cast_fp16)[name = string("op_979_cast_fp16")];
            tensor<fp16, [1, 6, 8, 1]> s_41_cast_fp16 = maximum(x = var_979_cast_fp16, y = var_33_to_fp16)[name = string("s_41_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> y_45_cast_fp16 = real_div(x = x_59_cast_fp16, y = s_41_cast_fp16)[name = string("y_45_cast_fp16")];
            fp16 var_34_promoted_19_to_fp16 = const()[name = string("op_34_promoted_19_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 8, 64]> var_982_cast_fp16 = pow(x = y_45_cast_fp16, y = var_34_promoted_19_to_fp16)[name = string("op_982_cast_fp16")];
            tensor<int32, [1]> var_984_axes_0 = const()[name = string("op_984_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_984_keep_dims_0 = const()[name = string("op_984_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 8, 1]> var_984_cast_fp16 = reduce_mean(axes = var_984_axes_0, keep_dims = var_984_keep_dims_0, x = var_982_cast_fp16)[name = string("op_984_cast_fp16")];
            fp32 var_985_epsilon_0 = const()[name = string("op_985_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 8, 1]> var_985_cast_fp16 = inverse(epsilon = var_985_epsilon_0, x = s_41_cast_fp16)[name = string("op_985_cast_fp16")];
            fp16 var_986_to_fp16 = const()[name = string("op_986_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 8, 1]> var_987_cast_fp16 = mul(x = var_985_cast_fp16, y = var_986_to_fp16)[name = string("op_987_cast_fp16")];
            tensor<fp16, [1, 6, 8, 1]> var_988_cast_fp16 = real_div(x = var_987_cast_fp16, y = s_41_cast_fp16)[name = string("op_988_cast_fp16")];
            tensor<fp16, [1, 6, 8, 1]> var_989_cast_fp16 = add(x = var_984_cast_fp16, y = var_988_cast_fp16)[name = string("op_989_cast_fp16")];
            fp32 var_990_epsilon_0 = const()[name = string("op_990_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 8, 1]> var_990_cast_fp16 = rsqrt(epsilon = var_990_epsilon_0, x = var_989_cast_fp16)[name = string("op_990_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> var_991_cast_fp16 = mul(x = y_45_cast_fp16, y = var_990_cast_fp16)[name = string("op_991_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_4_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36558656)))];
            tensor<fp16, [1, 6, 8, 64]> var_992_cast_fp16 = mul(x = core_backbone_layers_4_self_attn_q_norm_weight_to_fp16, y = var_991_cast_fp16)[name = string("op_992_cast_fp16")];
            tensor<fp16, [128, 512]> core_backbone_layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36558848)))];
            tensor<fp16, [1, 6, 128]> linear_25_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_k_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_25_cast_fp16")];
            tensor<int32, [4]> concat_9 = const()[name = string("concat_9"), val = tensor<int32, [4]>([1, 6, 2, 64])];
            tensor<fp16, [1, 6, 2, 64]> x_61_cast_fp16 = reshape(shape = concat_9, x = linear_25_cast_fp16)[name = string("x_61_cast_fp16")];
            tensor<fp16, [1, 6, 2, 64]> var_999_cast_fp16 = abs(x = x_61_cast_fp16)[name = string("op_999_cast_fp16")];
            tensor<int32, [1]> var_1001_axes_0 = const()[name = string("op_1001_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 2, 1]> var_1001_cast_fp16 = reduce_max(axes = var_1001_axes_0, keep_dims = var_32, x = var_999_cast_fp16)[name = string("op_1001_cast_fp16")];
            tensor<fp16, [1, 6, 2, 1]> s_43_cast_fp16 = maximum(x = var_1001_cast_fp16, y = var_33_to_fp16)[name = string("s_43_cast_fp16")];
            tensor<fp16, [1, 6, 2, 64]> y_47_cast_fp16 = real_div(x = x_61_cast_fp16, y = s_43_cast_fp16)[name = string("y_47_cast_fp16")];
            fp16 var_34_promoted_20_to_fp16 = const()[name = string("op_34_promoted_20_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 2, 64]> var_1004_cast_fp16 = pow(x = y_47_cast_fp16, y = var_34_promoted_20_to_fp16)[name = string("op_1004_cast_fp16")];
            tensor<int32, [1]> var_1006_axes_0 = const()[name = string("op_1006_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1006_keep_dims_0 = const()[name = string("op_1006_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 2, 1]> var_1006_cast_fp16 = reduce_mean(axes = var_1006_axes_0, keep_dims = var_1006_keep_dims_0, x = var_1004_cast_fp16)[name = string("op_1006_cast_fp16")];
            fp32 var_1007_epsilon_0 = const()[name = string("op_1007_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 2, 1]> var_1007_cast_fp16 = inverse(epsilon = var_1007_epsilon_0, x = s_43_cast_fp16)[name = string("op_1007_cast_fp16")];
            fp16 var_1008_to_fp16 = const()[name = string("op_1008_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 2, 1]> var_1009_cast_fp16 = mul(x = var_1007_cast_fp16, y = var_1008_to_fp16)[name = string("op_1009_cast_fp16")];
            tensor<fp16, [1, 6, 2, 1]> var_1010_cast_fp16 = real_div(x = var_1009_cast_fp16, y = s_43_cast_fp16)[name = string("op_1010_cast_fp16")];
            tensor<fp16, [1, 6, 2, 1]> var_1011_cast_fp16 = add(x = var_1006_cast_fp16, y = var_1010_cast_fp16)[name = string("op_1011_cast_fp16")];
            fp32 var_1012_epsilon_0 = const()[name = string("op_1012_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 2, 1]> var_1012_cast_fp16 = rsqrt(epsilon = var_1012_epsilon_0, x = var_1011_cast_fp16)[name = string("op_1012_cast_fp16")];
            tensor<fp16, [1, 6, 2, 64]> var_1013_cast_fp16 = mul(x = y_47_cast_fp16, y = var_1012_cast_fp16)[name = string("op_1013_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_4_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36689984)))];
            tensor<fp16, [1, 6, 2, 64]> var_1014_cast_fp16 = mul(x = core_backbone_layers_4_self_attn_k_norm_weight_to_fp16, y = var_1013_cast_fp16)[name = string("op_1014_cast_fp16")];
            tensor<int32, [4]> new_11_perm_0 = const()[name = string("new_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [128, 512]> core_backbone_layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36690176)))];
            tensor<fp16, [1, 6, 128]> linear_26_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_v_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_26_cast_fp16")];
            tensor<int32, [4]> concat_10 = const()[name = string("concat_10"), val = tensor<int32, [4]>([1, 6, 2, 64])];
            tensor<fp16, [1, 6, 2, 64]> var_1019_cast_fp16 = reshape(shape = concat_10, x = linear_26_cast_fp16)[name = string("op_1019_cast_fp16")];
            tensor<int32, [4]> new_13_perm_0 = const()[name = string("new_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_57_begin_0 = const()[name = string("input_57_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
            tensor<int32, [5]> input_57_end_0 = const()[name = string("input_57_end_0"), val = tensor<int32, [5]>([2, 1, 2, 29, 64])];
            tensor<bool, [5]> input_57_end_mask_0 = const()[name = string("input_57_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_57_squeeze_mask_0 = const()[name = string("input_57_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_57_cast_fp16 = slice_by_index(begin = input_57_begin_0, end = input_57_end_0, end_mask = input_57_end_mask_0, squeeze_mask = input_57_squeeze_mask_0, x = k)[name = string("input_57_cast_fp16")];
            tensor<int32, [8]> var_1024_pad_0 = const()[name = string("op_1024_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_1024_mode_0 = const()[name = string("op_1024_mode_0"), val = string("constant")];
            fp16 const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 6, 64]> new_11_cast_fp16 = transpose(perm = new_11_perm_0, x = var_1014_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [1, 2, 35, 64]> var_1024_cast_fp16 = pad(constant_val = const_17_to_fp16, mode = var_1024_mode_0, pad = var_1024_pad_0, x = new_11_cast_fp16)[name = string("op_1024_cast_fp16")];
            tensor<int32, [8]> var_1026_pad_0 = const()[name = string("op_1026_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 6, 0, 0])];
            string var_1026_mode_0 = const()[name = string("op_1026_mode_0"), val = string("constant")];
            fp16 const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 35, 64]> var_1026_cast_fp16 = pad(constant_val = const_18_to_fp16, mode = var_1026_mode_0, pad = var_1026_pad_0, x = input_57_cast_fp16)[name = string("op_1026_cast_fp16")];
            tensor<fp16, [1, 2, 35, 64]> x_65_cast_fp16 = add(x = var_1024_cast_fp16, y = var_1026_cast_fp16)[name = string("x_65_cast_fp16")];
            tensor<int32, [5]> input_61_begin_0 = const()[name = string("input_61_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
            tensor<int32, [5]> input_61_end_0 = const()[name = string("input_61_end_0"), val = tensor<int32, [5]>([2, 1, 2, 29, 64])];
            tensor<bool, [5]> input_61_end_mask_0 = const()[name = string("input_61_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_61_squeeze_mask_0 = const()[name = string("input_61_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_61_cast_fp16 = slice_by_index(begin = input_61_begin_0, end = input_61_end_0, end_mask = input_61_end_mask_0, squeeze_mask = input_61_squeeze_mask_0, x = v)[name = string("input_61_cast_fp16")];
            tensor<int32, [8]> var_1031_pad_0 = const()[name = string("op_1031_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_1031_mode_0 = const()[name = string("op_1031_mode_0"), val = string("constant")];
            fp16 const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 6, 64]> new_13_cast_fp16 = transpose(perm = new_13_perm_0, x = var_1019_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [1, 2, 35, 64]> var_1031_cast_fp16 = pad(constant_val = const_19_to_fp16, mode = var_1031_mode_0, pad = var_1031_pad_0, x = new_13_cast_fp16)[name = string("op_1031_cast_fp16")];
            tensor<int32, [8]> var_1033_pad_0 = const()[name = string("op_1033_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 6, 0, 0])];
            string var_1033_mode_0 = const()[name = string("op_1033_mode_0"), val = string("constant")];
            fp16 const_20_to_fp16 = const()[name = string("const_20_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 35, 64]> var_1033_cast_fp16 = pad(constant_val = const_20_to_fp16, mode = var_1033_mode_0, pad = var_1033_pad_0, x = input_61_cast_fp16)[name = string("op_1033_cast_fp16")];
            tensor<fp16, [1, 2, 35, 64]> v_7_cast_fp16 = add(x = var_1031_cast_fp16, y = var_1033_cast_fp16)[name = string("v_7_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> var_1035_cast_fp16 = mul(x = var_992_cast_fp16, y = const_33_to_fp16)[name = string("op_1035_cast_fp16")];
            tensor<int32, [2]> var_1036_split_sizes_0 = const()[name = string("op_1036_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 const_38 = const()[name = string("const_38"), val = int32(3)];
            tensor<fp16, [1, 6, 8, 32]> var_1036_cast_fp16_0, tensor<fp16, [1, 6, 8, 32]> var_1036_cast_fp16_1 = split(axis = const_38, split_sizes = var_1036_split_sizes_0, x = var_992_cast_fp16)[name = string("op_1036_cast_fp16")];
            fp16 const_21_promoted_to_fp16 = const()[name = string("const_21_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 6, 8, 32]> var_1038_cast_fp16 = mul(x = var_1036_cast_fp16_1, y = const_21_promoted_to_fp16)[name = string("op_1038_cast_fp16")];
            bool var_1040_interleave_0 = const()[name = string("op_1040_interleave_0"), val = bool(false)];
            int32 const_39 = const()[name = string("const_39"), val = int32(3)];
            tensor<fp16, [1, 6, 8, 64]> var_1040_cast_fp16 = concat(axis = const_39, interleave = var_1040_interleave_0, values = (var_1038_cast_fp16, var_1036_cast_fp16_0))[name = string("op_1040_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> var_1041_cast_fp16 = mul(x = var_1040_cast_fp16, y = const_36_to_fp16)[name = string("op_1041_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> q_3_cast_fp16 = add(x = var_1035_cast_fp16, y = var_1041_cast_fp16)[name = string("q_3_cast_fp16")];
            tensor<fp16, [1, 2, 35, 64]> var_1043_cast_fp16 = mul(x = x_65_cast_fp16, y = const_1_to_fp16)[name = string("op_1043_cast_fp16")];
            tensor<int32, [2]> var_1044_split_sizes_0 = const()[name = string("op_1044_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 var_1044_axis_0 = const()[name = string("op_1044_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 2, 35, 32]> var_1044_cast_fp16_0, tensor<fp16, [1, 2, 35, 32]> var_1044_cast_fp16_1 = split(axis = var_1044_axis_0, split_sizes = var_1044_split_sizes_0, x = x_65_cast_fp16)[name = string("op_1044_cast_fp16")];
            fp16 const_22_promoted_to_fp16 = const()[name = string("const_22_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 2, 35, 32]> var_1046_cast_fp16 = mul(x = var_1044_cast_fp16_1, y = const_22_promoted_to_fp16)[name = string("op_1046_cast_fp16")];
            bool var_1048_interleave_0 = const()[name = string("op_1048_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 2, 35, 64]> var_1048_cast_fp16 = concat(axis = var_31, interleave = var_1048_interleave_0, values = (var_1046_cast_fp16, var_1044_cast_fp16_0))[name = string("op_1048_cast_fp16")];
            tensor<fp16, [1, 2, 35, 64]> var_1049_cast_fp16 = mul(x = var_1048_cast_fp16, y = const_2_to_fp16)[name = string("op_1049_cast_fp16")];
            tensor<fp16, [1, 2, 35, 64]> k_7_cast_fp16 = add(x = var_1043_cast_fp16, y = var_1049_cast_fp16)[name = string("k_7_cast_fp16")];
            tensor<int32, [1]> var_1051_axes_0 = const()[name = string("op_1051_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 35, 64]> var_1051_cast_fp16 = expand_dims(axes = var_1051_axes_0, x = k_7_cast_fp16)[name = string("op_1051_cast_fp16")];
            tensor<int32, [5]> var_1053_reps_0 = const()[name = string("op_1053_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 35, 64]> var_1053_cast_fp16 = tile(reps = var_1053_reps_0, x = var_1051_cast_fp16)[name = string("op_1053_cast_fp16")];
            tensor<int32, [4]> concat_12 = const()[name = string("concat_12"), val = tensor<int32, [4]>([1, 8, 35, 64])];
            tensor<fp16, [1, 8, 35, 64]> k_9_cast_fp16 = reshape(shape = concat_12, x = var_1053_cast_fp16)[name = string("k_9_cast_fp16")];
            tensor<int32, [1]> var_1055_axes_0 = const()[name = string("op_1055_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 35, 64]> var_1055_cast_fp16 = expand_dims(axes = var_1055_axes_0, x = v_7_cast_fp16)[name = string("op_1055_cast_fp16")];
            tensor<int32, [5]> var_1057_reps_0 = const()[name = string("op_1057_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 35, 64]> var_1057_cast_fp16 = tile(reps = var_1057_reps_0, x = var_1055_cast_fp16)[name = string("op_1057_cast_fp16")];
            tensor<int32, [4]> concat_14 = const()[name = string("concat_14"), val = tensor<int32, [4]>([1, 8, 35, 64])];
            tensor<fp16, [1, 8, 35, 64]> v_9_cast_fp16 = reshape(shape = concat_14, x = var_1057_cast_fp16)[name = string("v_9_cast_fp16")];
            tensor<int32, [4]> transpose_2_perm_0 = const()[name = string("transpose_2_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1060_transpose_x_1 = const()[name = string("op_1060_transpose_x_1"), val = bool(false)];
            bool var_1060_transpose_y_1 = const()[name = string("op_1060_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 6, 64]> transpose_2_cast_fp16 = transpose(perm = transpose_2_perm_0, x = q_3_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [1, 8, 6, 35]> var_1060_cast_fp16 = matmul(transpose_x = var_1060_transpose_x_1, transpose_y = var_1060_transpose_y_1, x = transpose_2_cast_fp16, y = k_9_cast_fp16)[name = string("op_1060_cast_fp16")];
            fp16 var_1061_to_fp16 = const()[name = string("op_1061_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 8, 6, 35]> s_45_cast_fp16 = mul(x = var_1060_cast_fp16, y = var_1061_to_fp16)[name = string("s_45_cast_fp16")];
            tensor<fp16, [1, 8, 6, 35]> x_67_cast_fp16 = add(x = s_45_cast_fp16, y = mask)[name = string("x_67_cast_fp16")];
            tensor<int32, [1]> var_1065_axes_0 = const()[name = string("op_1065_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 6, 1]> var_1065_cast_fp16 = reduce_max(axes = var_1065_axes_0, keep_dims = var_32, x = x_67_cast_fp16)[name = string("op_1065_cast_fp16")];
            tensor<fp16, [1, 8, 6, 35]> x_69_cast_fp16 = sub(x = x_67_cast_fp16, y = var_1065_cast_fp16)[name = string("x_69_cast_fp16")];
            tensor<fp16, [1, 8, 6, 35]> x_71_cast_fp16 = exp(x = x_69_cast_fp16)[name = string("x_71_cast_fp16")];
            tensor<int32, [1]> var_1069_axes_0 = const()[name = string("op_1069_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1069_keep_dims_0 = const()[name = string("op_1069_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 6, 1]> var_1069_cast_fp16 = reduce_sum(axes = var_1069_axes_0, keep_dims = var_1069_keep_dims_0, x = x_71_cast_fp16)[name = string("op_1069_cast_fp16")];
            tensor<fp16, [1, 8, 6, 35]> var_1070_cast_fp16 = real_div(x = x_71_cast_fp16, y = var_1069_cast_fp16)[name = string("op_1070_cast_fp16")];
            bool o_7_transpose_x_0 = const()[name = string("o_7_transpose_x_0"), val = bool(false)];
            bool o_7_transpose_y_0 = const()[name = string("o_7_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 6, 64]> o_7_cast_fp16 = matmul(transpose_x = o_7_transpose_x_0, transpose_y = o_7_transpose_y_0, x = var_1070_cast_fp16, y = v_9_cast_fp16)[name = string("o_7_cast_fp16")];
            tensor<int32, [4]> o_9_perm_0 = const()[name = string("o_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [8, 512]> core_backbone_layers_4_self_attn_gate_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_gate_proj_weight_to_fp16"), val = tensor<fp16, [8, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36821312)))];
            tensor<fp16, [8]> linear_27_bias_0_to_fp16 = const()[name = string("linear_27_bias_0_to_fp16"), val = tensor<fp16, [8]>([0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0])];
            tensor<fp16, [1, 6, 8]> linear_27_cast_fp16 = linear(bias = linear_27_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_gate_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_27_cast_fp16")];
            fp16 var_1076_to_fp16 = const()[name = string("op_1076_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 6, 8]> var_1077_cast_fp16 = mul(x = linear_27_cast_fp16, y = var_1076_to_fp16)[name = string("op_1077_cast_fp16")];
            tensor<fp16, [1, 6, 8]> var_1078_cast_fp16 = tanh(x = var_1077_cast_fp16)[name = string("op_1078_cast_fp16")];
            fp16 var_1079_to_fp16 = const()[name = string("op_1079_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 6, 8]> var_1080_cast_fp16 = add(x = var_1078_cast_fp16, y = var_1079_to_fp16)[name = string("op_1080_cast_fp16")];
            fp16 var_1081_to_fp16 = const()[name = string("op_1081_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 6, 8]> var_1082_cast_fp16 = mul(x = var_1080_cast_fp16, y = var_1081_to_fp16)[name = string("op_1082_cast_fp16")];
            tensor<int32, [1]> var_1083_axes_0 = const()[name = string("op_1083_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 8, 1]> var_1083_cast_fp16 = expand_dims(axes = var_1083_axes_0, x = var_1082_cast_fp16)[name = string("op_1083_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> o_9_cast_fp16 = transpose(perm = o_9_perm_0, x = o_7_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [1, 6, 8, 64]> o_11_cast_fp16 = mul(x = o_9_cast_fp16, y = var_1083_cast_fp16)[name = string("o_11_cast_fp16")];
            tensor<int32, [3]> concat_15 = const()[name = string("concat_15"), val = tensor<int32, [3]>([1, 6, 512])];
            tensor<fp16, [1, 6, 512]> input_63_cast_fp16 = reshape(shape = concat_15, x = o_11_cast_fp16)[name = string("input_63_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36829568)))];
            tensor<fp16, [1, 6, 512]> linear_28_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_o_proj_weight_to_fp16, x = input_63_cast_fp16)[name = string("linear_28_cast_fp16")];
            tensor<int32, [4]> var_1090_begin_0 = const()[name = string("op_1090_begin_0"), val = tensor<int32, [4]>([0, 0, 6, 0])];
            tensor<int32, [4]> var_1090_end_0 = const()[name = string("op_1090_end_0"), val = tensor<int32, [4]>([1, 2, 35, 64])];
            tensor<bool, [4]> var_1090_end_mask_0 = const()[name = string("op_1090_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_1090_cast_fp16 = slice_by_index(begin = var_1090_begin_0, end = var_1090_end_0, end_mask = var_1090_end_mask_0, x = x_65_cast_fp16)[name = string("op_1090_cast_fp16")];
            tensor<int32, [4]> var_1093_begin_0 = const()[name = string("op_1093_begin_0"), val = tensor<int32, [4]>([0, 0, 6, 0])];
            tensor<int32, [4]> var_1093_end_0 = const()[name = string("op_1093_end_0"), val = tensor<int32, [4]>([1, 2, 35, 64])];
            tensor<bool, [4]> var_1093_end_mask_0 = const()[name = string("op_1093_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_1093_cast_fp16 = slice_by_index(begin = var_1093_begin_0, end = var_1093_end_0, end_mask = var_1093_end_mask_0, x = v_7_cast_fp16)[name = string("op_1093_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1095_cast_fp16 = abs(x = linear_28_cast_fp16)[name = string("op_1095_cast_fp16")];
            tensor<int32, [1]> var_1097_axes_0 = const()[name = string("op_1097_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1097_cast_fp16 = reduce_max(axes = var_1097_axes_0, keep_dims = var_32, x = var_1095_cast_fp16)[name = string("op_1097_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_47_cast_fp16 = maximum(x = var_1097_cast_fp16, y = var_33_to_fp16)[name = string("s_47_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_49_cast_fp16 = real_div(x = linear_28_cast_fp16, y = s_47_cast_fp16)[name = string("y_49_cast_fp16")];
            fp16 var_34_promoted_21_to_fp16 = const()[name = string("op_34_promoted_21_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1100_cast_fp16 = pow(x = y_49_cast_fp16, y = var_34_promoted_21_to_fp16)[name = string("op_1100_cast_fp16")];
            tensor<int32, [1]> var_1102_axes_0 = const()[name = string("op_1102_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1102_keep_dims_0 = const()[name = string("op_1102_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1102_cast_fp16 = reduce_mean(axes = var_1102_axes_0, keep_dims = var_1102_keep_dims_0, x = var_1100_cast_fp16)[name = string("op_1102_cast_fp16")];
            fp32 var_1103_epsilon_0 = const()[name = string("op_1103_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1103_cast_fp16 = inverse(epsilon = var_1103_epsilon_0, x = s_47_cast_fp16)[name = string("op_1103_cast_fp16")];
            fp16 var_1104_to_fp16 = const()[name = string("op_1104_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1105_cast_fp16 = mul(x = var_1103_cast_fp16, y = var_1104_to_fp16)[name = string("op_1105_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1106_cast_fp16 = real_div(x = var_1105_cast_fp16, y = s_47_cast_fp16)[name = string("op_1106_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1107_cast_fp16 = add(x = var_1102_cast_fp16, y = var_1106_cast_fp16)[name = string("op_1107_cast_fp16")];
            fp32 var_1108_epsilon_0 = const()[name = string("op_1108_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1108_cast_fp16 = rsqrt(epsilon = var_1108_epsilon_0, x = var_1107_cast_fp16)[name = string("op_1108_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1109_cast_fp16 = mul(x = y_49_cast_fp16, y = var_1108_cast_fp16)[name = string("op_1109_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_4_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37353920)))];
            tensor<fp16, [1, 6, 512]> a_9_cast_fp16 = mul(x = core_backbone_layers_4_post_operator_norm_weight_to_fp16, y = var_1109_cast_fp16)[name = string("a_9_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_77_cast_fp16 = add(x = x_57_cast_fp16, y = a_9_cast_fp16)[name = string("x_77_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1113_cast_fp16 = abs(x = x_77_cast_fp16)[name = string("op_1113_cast_fp16")];
            tensor<int32, [1]> var_1115_axes_0 = const()[name = string("op_1115_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1115_cast_fp16 = reduce_max(axes = var_1115_axes_0, keep_dims = var_32, x = var_1113_cast_fp16)[name = string("op_1115_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_49_cast_fp16 = maximum(x = var_1115_cast_fp16, y = var_33_to_fp16)[name = string("s_49_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_51_cast_fp16 = real_div(x = x_77_cast_fp16, y = s_49_cast_fp16)[name = string("y_51_cast_fp16")];
            fp16 var_34_promoted_22_to_fp16 = const()[name = string("op_34_promoted_22_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1118_cast_fp16 = pow(x = y_51_cast_fp16, y = var_34_promoted_22_to_fp16)[name = string("op_1118_cast_fp16")];
            tensor<int32, [1]> var_1120_axes_0 = const()[name = string("op_1120_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1120_keep_dims_0 = const()[name = string("op_1120_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1120_cast_fp16 = reduce_mean(axes = var_1120_axes_0, keep_dims = var_1120_keep_dims_0, x = var_1118_cast_fp16)[name = string("op_1120_cast_fp16")];
            fp32 var_1121_epsilon_0 = const()[name = string("op_1121_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1121_cast_fp16 = inverse(epsilon = var_1121_epsilon_0, x = s_49_cast_fp16)[name = string("op_1121_cast_fp16")];
            fp16 var_1122_to_fp16 = const()[name = string("op_1122_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1123_cast_fp16 = mul(x = var_1121_cast_fp16, y = var_1122_to_fp16)[name = string("op_1123_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1124_cast_fp16 = real_div(x = var_1123_cast_fp16, y = s_49_cast_fp16)[name = string("op_1124_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1125_cast_fp16 = add(x = var_1120_cast_fp16, y = var_1124_cast_fp16)[name = string("op_1125_cast_fp16")];
            fp32 var_1126_epsilon_0 = const()[name = string("op_1126_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1126_cast_fp16 = rsqrt(epsilon = var_1126_epsilon_0, x = var_1125_cast_fp16)[name = string("op_1126_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1127_cast_fp16 = mul(x = y_51_cast_fp16, y = var_1126_cast_fp16)[name = string("op_1127_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_4_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37355008)))];
            tensor<fp16, [1, 6, 512]> input_65_cast_fp16 = mul(x = core_backbone_layers_4_ffn_norm_weight_to_fp16, y = var_1127_cast_fp16)[name = string("input_65_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_4_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37356096)))];
            tensor<fp16, [1, 6, 2304]> linear_29_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w1_weight_to_fp16, x = input_65_cast_fp16)[name = string("linear_29_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_4_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39715456)))];
            tensor<fp16, [1, 6, 2304]> linear_30_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w3_weight_to_fp16, x = input_65_cast_fp16)[name = string("linear_30_cast_fp16")];
            fp16 var_1136_to_fp16 = const()[name = string("op_1136_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 6, 2304]> var_1137_cast_fp16 = mul(x = linear_29_cast_fp16, y = var_1136_to_fp16)[name = string("op_1137_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_1140_cast_fp16 = tanh(x = var_1137_cast_fp16)[name = string("op_1140_cast_fp16")];
            fp16 var_1141_to_fp16 = const()[name = string("op_1141_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 6, 2304]> var_1142_cast_fp16 = add(x = var_1140_cast_fp16, y = var_1141_to_fp16)[name = string("op_1142_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_1143_cast_fp16 = mul(x = var_1137_cast_fp16, y = var_1142_cast_fp16)[name = string("op_1143_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> input_67_cast_fp16 = mul(x = var_1143_cast_fp16, y = linear_30_cast_fp16)[name = string("input_67_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_4_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42074816)))];
            tensor<fp16, [1, 6, 512]> linear_31_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w2_weight_to_fp16, x = input_67_cast_fp16)[name = string("linear_31_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1148_cast_fp16 = abs(x = linear_31_cast_fp16)[name = string("op_1148_cast_fp16")];
            tensor<int32, [1]> var_1150_axes_0 = const()[name = string("op_1150_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1150_cast_fp16 = reduce_max(axes = var_1150_axes_0, keep_dims = var_32, x = var_1148_cast_fp16)[name = string("op_1150_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_51_cast_fp16 = maximum(x = var_1150_cast_fp16, y = var_33_to_fp16)[name = string("s_51_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_53_cast_fp16 = real_div(x = linear_31_cast_fp16, y = s_51_cast_fp16)[name = string("y_53_cast_fp16")];
            fp16 var_34_promoted_23_to_fp16 = const()[name = string("op_34_promoted_23_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1153_cast_fp16 = pow(x = y_53_cast_fp16, y = var_34_promoted_23_to_fp16)[name = string("op_1153_cast_fp16")];
            tensor<int32, [1]> var_1155_axes_0 = const()[name = string("op_1155_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1155_keep_dims_0 = const()[name = string("op_1155_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1155_cast_fp16 = reduce_mean(axes = var_1155_axes_0, keep_dims = var_1155_keep_dims_0, x = var_1153_cast_fp16)[name = string("op_1155_cast_fp16")];
            fp32 var_1156_epsilon_0 = const()[name = string("op_1156_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1156_cast_fp16 = inverse(epsilon = var_1156_epsilon_0, x = s_51_cast_fp16)[name = string("op_1156_cast_fp16")];
            fp16 var_1157_to_fp16 = const()[name = string("op_1157_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1158_cast_fp16 = mul(x = var_1156_cast_fp16, y = var_1157_to_fp16)[name = string("op_1158_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1159_cast_fp16 = real_div(x = var_1158_cast_fp16, y = s_51_cast_fp16)[name = string("op_1159_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1160_cast_fp16 = add(x = var_1155_cast_fp16, y = var_1159_cast_fp16)[name = string("op_1160_cast_fp16")];
            fp32 var_1161_epsilon_0 = const()[name = string("op_1161_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1161_cast_fp16 = rsqrt(epsilon = var_1161_epsilon_0, x = var_1160_cast_fp16)[name = string("op_1161_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1162_cast_fp16 = mul(x = y_53_cast_fp16, y = var_1161_cast_fp16)[name = string("op_1162_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_4_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44434176)))];
            tensor<fp16, [1, 6, 512]> f_9_cast_fp16 = mul(x = core_backbone_layers_4_post_ffn_norm_weight_to_fp16, y = var_1162_cast_fp16)[name = string("f_9_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_83_cast_fp16 = add(x = x_77_cast_fp16, y = f_9_cast_fp16)[name = string("x_83_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1166_cast_fp16 = abs(x = x_83_cast_fp16)[name = string("op_1166_cast_fp16")];
            tensor<int32, [1]> var_1168_axes_0 = const()[name = string("op_1168_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1168_cast_fp16 = reduce_max(axes = var_1168_axes_0, keep_dims = var_32, x = var_1166_cast_fp16)[name = string("op_1168_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_53_cast_fp16 = maximum(x = var_1168_cast_fp16, y = var_33_to_fp16)[name = string("s_53_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_55_cast_fp16 = real_div(x = x_83_cast_fp16, y = s_53_cast_fp16)[name = string("y_55_cast_fp16")];
            fp16 var_34_promoted_24_to_fp16 = const()[name = string("op_34_promoted_24_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1171_cast_fp16 = pow(x = y_55_cast_fp16, y = var_34_promoted_24_to_fp16)[name = string("op_1171_cast_fp16")];
            tensor<int32, [1]> var_1173_axes_0 = const()[name = string("op_1173_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1173_keep_dims_0 = const()[name = string("op_1173_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1173_cast_fp16 = reduce_mean(axes = var_1173_axes_0, keep_dims = var_1173_keep_dims_0, x = var_1171_cast_fp16)[name = string("op_1173_cast_fp16")];
            fp32 var_1174_epsilon_0 = const()[name = string("op_1174_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1174_cast_fp16 = inverse(epsilon = var_1174_epsilon_0, x = s_53_cast_fp16)[name = string("op_1174_cast_fp16")];
            fp16 var_1175_to_fp16 = const()[name = string("op_1175_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1176_cast_fp16 = mul(x = var_1174_cast_fp16, y = var_1175_to_fp16)[name = string("op_1176_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1177_cast_fp16 = real_div(x = var_1176_cast_fp16, y = s_53_cast_fp16)[name = string("op_1177_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1178_cast_fp16 = add(x = var_1173_cast_fp16, y = var_1177_cast_fp16)[name = string("op_1178_cast_fp16")];
            fp32 var_1179_epsilon_0 = const()[name = string("op_1179_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1179_cast_fp16 = rsqrt(epsilon = var_1179_epsilon_0, x = var_1178_cast_fp16)[name = string("op_1179_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1180_cast_fp16 = mul(x = y_55_cast_fp16, y = var_1179_cast_fp16)[name = string("op_1180_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_5_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44435264)))];
            tensor<fp16, [1, 6, 512]> input_69_cast_fp16 = mul(x = core_backbone_layers_5_operator_norm_weight_to_fp16, y = var_1180_cast_fp16)[name = string("input_69_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_5_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44436352)))];
            tensor<fp16, [1, 6, 1536]> linear_32_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_5_conv_in_proj_weight_to_fp16, x = input_69_cast_fp16)[name = string("linear_32_cast_fp16")];
            tensor<int32, [3]> bcx_7_perm_0 = const()[name = string("bcx_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_17_begin_0 = const()[name = string("b_17_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_17_end_0 = const()[name = string("b_17_end_0"), val = tensor<int32, [3]>([1, 512, 6])];
            tensor<bool, [3]> b_17_end_mask_0 = const()[name = string("b_17_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 6]> bcx_7_cast_fp16 = transpose(perm = bcx_7_perm_0, x = linear_32_cast_fp16)[name = string("transpose_13")];
            tensor<fp16, [1, 512, 6]> b_17_cast_fp16 = slice_by_index(begin = b_17_begin_0, end = b_17_end_0, end_mask = b_17_end_mask_0, x = bcx_7_cast_fp16)[name = string("b_17_cast_fp16")];
            tensor<int32, [3]> c_7_begin_0 = const()[name = string("c_7_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_7_end_0 = const()[name = string("c_7_end_0"), val = tensor<int32, [3]>([1, 1024, 6])];
            tensor<bool, [3]> c_7_end_mask_0 = const()[name = string("c_7_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 6]> c_7_cast_fp16 = slice_by_index(begin = c_7_begin_0, end = c_7_end_0, end_mask = c_7_end_mask_0, x = bcx_7_cast_fp16)[name = string("c_7_cast_fp16")];
            tensor<int32, [3]> xx_7_begin_0 = const()[name = string("xx_7_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_7_end_0 = const()[name = string("xx_7_end_0"), val = tensor<int32, [3]>([1, 1536, 6])];
            tensor<bool, [3]> xx_7_end_mask_0 = const()[name = string("xx_7_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> xx_7_cast_fp16 = slice_by_index(begin = xx_7_begin_0, end = xx_7_end_0, end_mask = xx_7_end_mask_0, x = bcx_7_cast_fp16)[name = string("xx_7_cast_fp16")];
            tensor<fp16, [1, 512, 6]> new_15_cast_fp16 = mul(x = b_17_cast_fp16, y = xx_7_cast_fp16)[name = string("new_15_cast_fp16")];
            tensor<int32, [4]> input_73_begin_0 = const()[name = string("input_73_begin_0"), val = tensor<int32, [4]>([3, 0, 0, 0])];
            tensor<int32, [4]> input_73_end_0 = const()[name = string("input_73_end_0"), val = tensor<int32, [4]>([4, 1, 512, 6])];
            tensor<bool, [4]> input_73_end_mask_0 = const()[name = string("input_73_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_73_squeeze_mask_0 = const()[name = string("input_73_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_73_cast_fp16 = slice_by_index(begin = input_73_begin_0, end = input_73_end_0, end_mask = input_73_end_mask_0, squeeze_mask = input_73_squeeze_mask_0, x = conv)[name = string("input_73_cast_fp16")];
            tensor<int32, [6]> var_1195_pad_0 = const()[name = string("op_1195_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_1195_mode_0 = const()[name = string("op_1195_mode_0"), val = string("constant")];
            fp16 const_23_to_fp16 = const()[name = string("const_23_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 12]> var_1195_cast_fp16 = pad(constant_val = const_23_to_fp16, mode = var_1195_mode_0, pad = var_1195_pad_0, x = new_15_cast_fp16)[name = string("op_1195_cast_fp16")];
            tensor<int32, [6]> var_1197_pad_0 = const()[name = string("op_1197_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 6])];
            string var_1197_mode_0 = const()[name = string("op_1197_mode_0"), val = string("constant")];
            fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 12]> var_1197_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = var_1197_mode_0, pad = var_1197_pad_0, x = input_73_cast_fp16)[name = string("op_1197_cast_fp16")];
            tensor<fp16, [1, 512, 12]> win_13_cast_fp16 = add(x = var_1195_cast_fp16, y = var_1197_cast_fp16)[name = string("win_13_cast_fp16")];
            tensor<int32, [3]> var_1199_begin_0 = const()[name = string("op_1199_begin_0"), val = tensor<int32, [3]>([0, 0, 6])];
            tensor<int32, [3]> var_1199_end_0 = const()[name = string("op_1199_end_0"), val = tensor<int32, [3]>([1, 512, 12])];
            tensor<bool, [3]> var_1199_end_mask_0 = const()[name = string("op_1199_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_1199_cast_fp16 = slice_by_index(begin = var_1199_begin_0, end = var_1199_end_0, end_mask = var_1199_end_mask_0, x = win_13_cast_fp16)[name = string("op_1199_cast_fp16")];
            string var_1205_pad_type_0 = const()[name = string("op_1205_pad_type_0"), val = string("valid")];
            int32 var_1205_groups_0 = const()[name = string("op_1205_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_1205_strides_0 = const()[name = string("op_1205_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1205_pad_0 = const()[name = string("op_1205_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1205_dilations_0 = const()[name = string("op_1205_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_5_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_5_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46009280)))];
            tensor<fp16, [1, 512, 6]> var_1205_cast_fp16 = conv(dilations = var_1205_dilations_0, groups = var_1205_groups_0, pad = var_1205_pad_0, pad_type = var_1205_pad_type_0, strides = var_1205_strides_0, weight = core_backbone_layers_5_conv_conv_weight_to_fp16, x = win_13_cast_fp16)[name = string("op_1205_cast_fp16")];
            tensor<fp16, [1, 512, 6]> var_1207_cast_fp16 = mul(x = c_7_cast_fp16, y = var_1205_cast_fp16)[name = string("op_1207_cast_fp16")];
            tensor<int32, [3]> input_75_perm_0 = const()[name = string("input_75_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_5_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46016512)))];
            tensor<fp16, [1, 6, 512]> input_75_cast_fp16 = transpose(perm = input_75_perm_0, x = var_1207_cast_fp16)[name = string("transpose_12")];
            tensor<fp16, [1, 6, 512]> linear_33_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_5_conv_out_proj_weight_to_fp16, x = input_75_cast_fp16)[name = string("linear_33_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1212_cast_fp16 = abs(x = linear_33_cast_fp16)[name = string("op_1212_cast_fp16")];
            tensor<int32, [1]> var_1214_axes_0 = const()[name = string("op_1214_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1214_cast_fp16 = reduce_max(axes = var_1214_axes_0, keep_dims = var_32, x = var_1212_cast_fp16)[name = string("op_1214_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_55_cast_fp16 = maximum(x = var_1214_cast_fp16, y = var_33_to_fp16)[name = string("s_55_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_59_cast_fp16 = real_div(x = linear_33_cast_fp16, y = s_55_cast_fp16)[name = string("y_59_cast_fp16")];
            fp16 var_34_promoted_25_to_fp16 = const()[name = string("op_34_promoted_25_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1217_cast_fp16 = pow(x = y_59_cast_fp16, y = var_34_promoted_25_to_fp16)[name = string("op_1217_cast_fp16")];
            tensor<int32, [1]> var_1219_axes_0 = const()[name = string("op_1219_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1219_keep_dims_0 = const()[name = string("op_1219_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1219_cast_fp16 = reduce_mean(axes = var_1219_axes_0, keep_dims = var_1219_keep_dims_0, x = var_1217_cast_fp16)[name = string("op_1219_cast_fp16")];
            fp32 var_1220_epsilon_0 = const()[name = string("op_1220_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1220_cast_fp16 = inverse(epsilon = var_1220_epsilon_0, x = s_55_cast_fp16)[name = string("op_1220_cast_fp16")];
            fp16 var_1221_to_fp16 = const()[name = string("op_1221_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1222_cast_fp16 = mul(x = var_1220_cast_fp16, y = var_1221_to_fp16)[name = string("op_1222_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1223_cast_fp16 = real_div(x = var_1222_cast_fp16, y = s_55_cast_fp16)[name = string("op_1223_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1224_cast_fp16 = add(x = var_1219_cast_fp16, y = var_1223_cast_fp16)[name = string("op_1224_cast_fp16")];
            fp32 var_1225_epsilon_0 = const()[name = string("op_1225_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1225_cast_fp16 = rsqrt(epsilon = var_1225_epsilon_0, x = var_1224_cast_fp16)[name = string("op_1225_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1226_cast_fp16 = mul(x = y_59_cast_fp16, y = var_1225_cast_fp16)[name = string("op_1226_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_5_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46540864)))];
            tensor<fp16, [1, 6, 512]> a_11_cast_fp16 = mul(x = core_backbone_layers_5_post_operator_norm_weight_to_fp16, y = var_1226_cast_fp16)[name = string("a_11_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_87_cast_fp16 = add(x = x_83_cast_fp16, y = a_11_cast_fp16)[name = string("x_87_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1230_cast_fp16 = abs(x = x_87_cast_fp16)[name = string("op_1230_cast_fp16")];
            tensor<int32, [1]> var_1232_axes_0 = const()[name = string("op_1232_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1232_cast_fp16 = reduce_max(axes = var_1232_axes_0, keep_dims = var_32, x = var_1230_cast_fp16)[name = string("op_1232_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_57_cast_fp16 = maximum(x = var_1232_cast_fp16, y = var_33_to_fp16)[name = string("s_57_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_61_cast_fp16 = real_div(x = x_87_cast_fp16, y = s_57_cast_fp16)[name = string("y_61_cast_fp16")];
            fp16 var_34_promoted_26_to_fp16 = const()[name = string("op_34_promoted_26_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1235_cast_fp16 = pow(x = y_61_cast_fp16, y = var_34_promoted_26_to_fp16)[name = string("op_1235_cast_fp16")];
            tensor<int32, [1]> var_1237_axes_0 = const()[name = string("op_1237_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1237_keep_dims_0 = const()[name = string("op_1237_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1237_cast_fp16 = reduce_mean(axes = var_1237_axes_0, keep_dims = var_1237_keep_dims_0, x = var_1235_cast_fp16)[name = string("op_1237_cast_fp16")];
            fp32 var_1238_epsilon_0 = const()[name = string("op_1238_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1238_cast_fp16 = inverse(epsilon = var_1238_epsilon_0, x = s_57_cast_fp16)[name = string("op_1238_cast_fp16")];
            fp16 var_1239_to_fp16 = const()[name = string("op_1239_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1240_cast_fp16 = mul(x = var_1238_cast_fp16, y = var_1239_to_fp16)[name = string("op_1240_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1241_cast_fp16 = real_div(x = var_1240_cast_fp16, y = s_57_cast_fp16)[name = string("op_1241_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1242_cast_fp16 = add(x = var_1237_cast_fp16, y = var_1241_cast_fp16)[name = string("op_1242_cast_fp16")];
            fp32 var_1243_epsilon_0 = const()[name = string("op_1243_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1243_cast_fp16 = rsqrt(epsilon = var_1243_epsilon_0, x = var_1242_cast_fp16)[name = string("op_1243_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1244_cast_fp16 = mul(x = y_61_cast_fp16, y = var_1243_cast_fp16)[name = string("op_1244_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_5_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46541952)))];
            tensor<fp16, [1, 6, 512]> input_77_cast_fp16 = mul(x = core_backbone_layers_5_ffn_norm_weight_to_fp16, y = var_1244_cast_fp16)[name = string("input_77_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_5_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46543040)))];
            tensor<fp16, [1, 6, 2304]> linear_34_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w1_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_34_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_5_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48902400)))];
            tensor<fp16, [1, 6, 2304]> linear_35_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w3_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_35_cast_fp16")];
            fp16 var_1253_to_fp16 = const()[name = string("op_1253_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 6, 2304]> var_1254_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_1253_to_fp16)[name = string("op_1254_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_1257_cast_fp16 = tanh(x = var_1254_cast_fp16)[name = string("op_1257_cast_fp16")];
            fp16 var_1258_to_fp16 = const()[name = string("op_1258_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 6, 2304]> var_1259_cast_fp16 = add(x = var_1257_cast_fp16, y = var_1258_to_fp16)[name = string("op_1259_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_1260_cast_fp16 = mul(x = var_1254_cast_fp16, y = var_1259_cast_fp16)[name = string("op_1260_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> input_79_cast_fp16 = mul(x = var_1260_cast_fp16, y = linear_35_cast_fp16)[name = string("input_79_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_5_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51261760)))];
            tensor<fp16, [1, 6, 512]> linear_36_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w2_weight_to_fp16, x = input_79_cast_fp16)[name = string("linear_36_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1265_cast_fp16 = abs(x = linear_36_cast_fp16)[name = string("op_1265_cast_fp16")];
            tensor<int32, [1]> var_1267_axes_0 = const()[name = string("op_1267_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1267_cast_fp16 = reduce_max(axes = var_1267_axes_0, keep_dims = var_32, x = var_1265_cast_fp16)[name = string("op_1267_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_59_cast_fp16 = maximum(x = var_1267_cast_fp16, y = var_33_to_fp16)[name = string("s_59_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_63_cast_fp16 = real_div(x = linear_36_cast_fp16, y = s_59_cast_fp16)[name = string("y_63_cast_fp16")];
            fp16 var_34_promoted_27_to_fp16 = const()[name = string("op_34_promoted_27_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1270_cast_fp16 = pow(x = y_63_cast_fp16, y = var_34_promoted_27_to_fp16)[name = string("op_1270_cast_fp16")];
            tensor<int32, [1]> var_1272_axes_0 = const()[name = string("op_1272_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1272_keep_dims_0 = const()[name = string("op_1272_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1272_cast_fp16 = reduce_mean(axes = var_1272_axes_0, keep_dims = var_1272_keep_dims_0, x = var_1270_cast_fp16)[name = string("op_1272_cast_fp16")];
            fp32 var_1273_epsilon_0 = const()[name = string("op_1273_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1273_cast_fp16 = inverse(epsilon = var_1273_epsilon_0, x = s_59_cast_fp16)[name = string("op_1273_cast_fp16")];
            fp16 var_1274_to_fp16 = const()[name = string("op_1274_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1275_cast_fp16 = mul(x = var_1273_cast_fp16, y = var_1274_to_fp16)[name = string("op_1275_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1276_cast_fp16 = real_div(x = var_1275_cast_fp16, y = s_59_cast_fp16)[name = string("op_1276_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1277_cast_fp16 = add(x = var_1272_cast_fp16, y = var_1276_cast_fp16)[name = string("op_1277_cast_fp16")];
            fp32 var_1278_epsilon_0 = const()[name = string("op_1278_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1278_cast_fp16 = rsqrt(epsilon = var_1278_epsilon_0, x = var_1277_cast_fp16)[name = string("op_1278_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1279_cast_fp16 = mul(x = y_63_cast_fp16, y = var_1278_cast_fp16)[name = string("op_1279_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_5_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53621120)))];
            tensor<fp16, [1, 6, 512]> f_11_cast_fp16 = mul(x = core_backbone_layers_5_post_ffn_norm_weight_to_fp16, y = var_1279_cast_fp16)[name = string("f_11_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_93_cast_fp16 = add(x = x_87_cast_fp16, y = f_11_cast_fp16)[name = string("x_93_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1283_cast_fp16 = abs(x = x_93_cast_fp16)[name = string("op_1283_cast_fp16")];
            tensor<int32, [1]> var_1285_axes_0 = const()[name = string("op_1285_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1285_cast_fp16 = reduce_max(axes = var_1285_axes_0, keep_dims = var_32, x = var_1283_cast_fp16)[name = string("op_1285_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_61_cast_fp16 = maximum(x = var_1285_cast_fp16, y = var_33_to_fp16)[name = string("s_61_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_65_cast_fp16 = real_div(x = x_93_cast_fp16, y = s_61_cast_fp16)[name = string("y_65_cast_fp16")];
            fp16 var_34_promoted_28_to_fp16 = const()[name = string("op_34_promoted_28_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1288_cast_fp16 = pow(x = y_65_cast_fp16, y = var_34_promoted_28_to_fp16)[name = string("op_1288_cast_fp16")];
            tensor<int32, [1]> var_1290_axes_0 = const()[name = string("op_1290_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1290_keep_dims_0 = const()[name = string("op_1290_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1290_cast_fp16 = reduce_mean(axes = var_1290_axes_0, keep_dims = var_1290_keep_dims_0, x = var_1288_cast_fp16)[name = string("op_1290_cast_fp16")];
            fp32 var_1291_epsilon_0 = const()[name = string("op_1291_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1291_cast_fp16 = inverse(epsilon = var_1291_epsilon_0, x = s_61_cast_fp16)[name = string("op_1291_cast_fp16")];
            fp16 var_1292_to_fp16 = const()[name = string("op_1292_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1293_cast_fp16 = mul(x = var_1291_cast_fp16, y = var_1292_to_fp16)[name = string("op_1293_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1294_cast_fp16 = real_div(x = var_1293_cast_fp16, y = s_61_cast_fp16)[name = string("op_1294_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1295_cast_fp16 = add(x = var_1290_cast_fp16, y = var_1294_cast_fp16)[name = string("op_1295_cast_fp16")];
            fp32 var_1296_epsilon_0 = const()[name = string("op_1296_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1296_cast_fp16 = rsqrt(epsilon = var_1296_epsilon_0, x = var_1295_cast_fp16)[name = string("op_1296_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1297_cast_fp16 = mul(x = y_65_cast_fp16, y = var_1296_cast_fp16)[name = string("op_1297_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_6_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53622208)))];
            tensor<fp16, [1, 6, 512]> input_81_cast_fp16 = mul(x = core_backbone_layers_6_operator_norm_weight_to_fp16, y = var_1297_cast_fp16)[name = string("input_81_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53623296)))];
            tensor<fp16, [1, 6, 512]> linear_37_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_q_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_37_cast_fp16")];
            tensor<int32, [4]> concat_16 = const()[name = string("concat_16"), val = tensor<int32, [4]>([1, 6, 8, 64])];
            tensor<fp16, [1, 6, 8, 64]> x_95_cast_fp16 = reshape(shape = concat_16, x = linear_37_cast_fp16)[name = string("x_95_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> var_1304_cast_fp16 = abs(x = x_95_cast_fp16)[name = string("op_1304_cast_fp16")];
            tensor<int32, [1]> var_1306_axes_0 = const()[name = string("op_1306_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 8, 1]> var_1306_cast_fp16 = reduce_max(axes = var_1306_axes_0, keep_dims = var_32, x = var_1304_cast_fp16)[name = string("op_1306_cast_fp16")];
            tensor<fp16, [1, 6, 8, 1]> s_63_cast_fp16 = maximum(x = var_1306_cast_fp16, y = var_33_to_fp16)[name = string("s_63_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> y_67_cast_fp16 = real_div(x = x_95_cast_fp16, y = s_63_cast_fp16)[name = string("y_67_cast_fp16")];
            fp16 var_34_promoted_29_to_fp16 = const()[name = string("op_34_promoted_29_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 8, 64]> var_1309_cast_fp16 = pow(x = y_67_cast_fp16, y = var_34_promoted_29_to_fp16)[name = string("op_1309_cast_fp16")];
            tensor<int32, [1]> var_1311_axes_0 = const()[name = string("op_1311_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1311_keep_dims_0 = const()[name = string("op_1311_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 8, 1]> var_1311_cast_fp16 = reduce_mean(axes = var_1311_axes_0, keep_dims = var_1311_keep_dims_0, x = var_1309_cast_fp16)[name = string("op_1311_cast_fp16")];
            fp32 var_1312_epsilon_0 = const()[name = string("op_1312_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 8, 1]> var_1312_cast_fp16 = inverse(epsilon = var_1312_epsilon_0, x = s_63_cast_fp16)[name = string("op_1312_cast_fp16")];
            fp16 var_1313_to_fp16 = const()[name = string("op_1313_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 8, 1]> var_1314_cast_fp16 = mul(x = var_1312_cast_fp16, y = var_1313_to_fp16)[name = string("op_1314_cast_fp16")];
            tensor<fp16, [1, 6, 8, 1]> var_1315_cast_fp16 = real_div(x = var_1314_cast_fp16, y = s_63_cast_fp16)[name = string("op_1315_cast_fp16")];
            tensor<fp16, [1, 6, 8, 1]> var_1316_cast_fp16 = add(x = var_1311_cast_fp16, y = var_1315_cast_fp16)[name = string("op_1316_cast_fp16")];
            fp32 var_1317_epsilon_0 = const()[name = string("op_1317_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 8, 1]> var_1317_cast_fp16 = rsqrt(epsilon = var_1317_epsilon_0, x = var_1316_cast_fp16)[name = string("op_1317_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> var_1318_cast_fp16 = mul(x = y_67_cast_fp16, y = var_1317_cast_fp16)[name = string("op_1318_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_6_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54147648)))];
            tensor<fp16, [1, 6, 8, 64]> var_1319_cast_fp16 = mul(x = core_backbone_layers_6_self_attn_q_norm_weight_to_fp16, y = var_1318_cast_fp16)[name = string("op_1319_cast_fp16")];
            tensor<fp16, [128, 512]> core_backbone_layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54147840)))];
            tensor<fp16, [1, 6, 128]> linear_38_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_k_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_38_cast_fp16")];
            tensor<int32, [4]> concat_17 = const()[name = string("concat_17"), val = tensor<int32, [4]>([1, 6, 2, 64])];
            tensor<fp16, [1, 6, 2, 64]> x_97_cast_fp16 = reshape(shape = concat_17, x = linear_38_cast_fp16)[name = string("x_97_cast_fp16")];
            tensor<fp16, [1, 6, 2, 64]> var_1326_cast_fp16 = abs(x = x_97_cast_fp16)[name = string("op_1326_cast_fp16")];
            tensor<int32, [1]> var_1328_axes_0 = const()[name = string("op_1328_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 2, 1]> var_1328_cast_fp16 = reduce_max(axes = var_1328_axes_0, keep_dims = var_32, x = var_1326_cast_fp16)[name = string("op_1328_cast_fp16")];
            tensor<fp16, [1, 6, 2, 1]> s_65_cast_fp16 = maximum(x = var_1328_cast_fp16, y = var_33_to_fp16)[name = string("s_65_cast_fp16")];
            tensor<fp16, [1, 6, 2, 64]> y_69_cast_fp16 = real_div(x = x_97_cast_fp16, y = s_65_cast_fp16)[name = string("y_69_cast_fp16")];
            fp16 var_34_promoted_30_to_fp16 = const()[name = string("op_34_promoted_30_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 2, 64]> var_1331_cast_fp16 = pow(x = y_69_cast_fp16, y = var_34_promoted_30_to_fp16)[name = string("op_1331_cast_fp16")];
            tensor<int32, [1]> var_1333_axes_0 = const()[name = string("op_1333_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1333_keep_dims_0 = const()[name = string("op_1333_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 2, 1]> var_1333_cast_fp16 = reduce_mean(axes = var_1333_axes_0, keep_dims = var_1333_keep_dims_0, x = var_1331_cast_fp16)[name = string("op_1333_cast_fp16")];
            fp32 var_1334_epsilon_0 = const()[name = string("op_1334_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 2, 1]> var_1334_cast_fp16 = inverse(epsilon = var_1334_epsilon_0, x = s_65_cast_fp16)[name = string("op_1334_cast_fp16")];
            fp16 var_1335_to_fp16 = const()[name = string("op_1335_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 2, 1]> var_1336_cast_fp16 = mul(x = var_1334_cast_fp16, y = var_1335_to_fp16)[name = string("op_1336_cast_fp16")];
            tensor<fp16, [1, 6, 2, 1]> var_1337_cast_fp16 = real_div(x = var_1336_cast_fp16, y = s_65_cast_fp16)[name = string("op_1337_cast_fp16")];
            tensor<fp16, [1, 6, 2, 1]> var_1338_cast_fp16 = add(x = var_1333_cast_fp16, y = var_1337_cast_fp16)[name = string("op_1338_cast_fp16")];
            fp32 var_1339_epsilon_0 = const()[name = string("op_1339_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 2, 1]> var_1339_cast_fp16 = rsqrt(epsilon = var_1339_epsilon_0, x = var_1338_cast_fp16)[name = string("op_1339_cast_fp16")];
            tensor<fp16, [1, 6, 2, 64]> var_1340_cast_fp16 = mul(x = y_69_cast_fp16, y = var_1339_cast_fp16)[name = string("op_1340_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_6_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54278976)))];
            tensor<fp16, [1, 6, 2, 64]> var_1341_cast_fp16 = mul(x = core_backbone_layers_6_self_attn_k_norm_weight_to_fp16, y = var_1340_cast_fp16)[name = string("op_1341_cast_fp16")];
            tensor<int32, [4]> new_17_perm_0 = const()[name = string("new_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [128, 512]> core_backbone_layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54279168)))];
            tensor<fp16, [1, 6, 128]> linear_39_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_v_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_39_cast_fp16")];
            tensor<int32, [4]> concat_18 = const()[name = string("concat_18"), val = tensor<int32, [4]>([1, 6, 2, 64])];
            tensor<fp16, [1, 6, 2, 64]> var_1346_cast_fp16 = reshape(shape = concat_18, x = linear_39_cast_fp16)[name = string("op_1346_cast_fp16")];
            tensor<int32, [4]> new_19_perm_0 = const()[name = string("new_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_85_begin_0 = const()[name = string("input_85_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])];
            tensor<int32, [5]> input_85_end_0 = const()[name = string("input_85_end_0"), val = tensor<int32, [5]>([3, 1, 2, 29, 64])];
            tensor<bool, [5]> input_85_end_mask_0 = const()[name = string("input_85_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_85_squeeze_mask_0 = const()[name = string("input_85_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_85_cast_fp16 = slice_by_index(begin = input_85_begin_0, end = input_85_end_0, end_mask = input_85_end_mask_0, squeeze_mask = input_85_squeeze_mask_0, x = k)[name = string("input_85_cast_fp16")];
            tensor<int32, [8]> var_1351_pad_0 = const()[name = string("op_1351_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_1351_mode_0 = const()[name = string("op_1351_mode_0"), val = string("constant")];
            fp16 const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 6, 64]> new_17_cast_fp16 = transpose(perm = new_17_perm_0, x = var_1341_cast_fp16)[name = string("transpose_11")];
            tensor<fp16, [1, 2, 35, 64]> var_1351_cast_fp16 = pad(constant_val = const_25_to_fp16, mode = var_1351_mode_0, pad = var_1351_pad_0, x = new_17_cast_fp16)[name = string("op_1351_cast_fp16")];
            tensor<int32, [8]> var_1353_pad_0 = const()[name = string("op_1353_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 6, 0, 0])];
            string var_1353_mode_0 = const()[name = string("op_1353_mode_0"), val = string("constant")];
            fp16 const_26_to_fp16 = const()[name = string("const_26_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 35, 64]> var_1353_cast_fp16 = pad(constant_val = const_26_to_fp16, mode = var_1353_mode_0, pad = var_1353_pad_0, x = input_85_cast_fp16)[name = string("op_1353_cast_fp16")];
            tensor<fp16, [1, 2, 35, 64]> x_101_cast_fp16 = add(x = var_1351_cast_fp16, y = var_1353_cast_fp16)[name = string("x_101_cast_fp16")];
            tensor<int32, [5]> input_89_begin_0 = const()[name = string("input_89_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])];
            tensor<int32, [5]> input_89_end_0 = const()[name = string("input_89_end_0"), val = tensor<int32, [5]>([3, 1, 2, 29, 64])];
            tensor<bool, [5]> input_89_end_mask_0 = const()[name = string("input_89_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_89_squeeze_mask_0 = const()[name = string("input_89_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_89_cast_fp16 = slice_by_index(begin = input_89_begin_0, end = input_89_end_0, end_mask = input_89_end_mask_0, squeeze_mask = input_89_squeeze_mask_0, x = v)[name = string("input_89_cast_fp16")];
            tensor<int32, [8]> var_1358_pad_0 = const()[name = string("op_1358_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_1358_mode_0 = const()[name = string("op_1358_mode_0"), val = string("constant")];
            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 6, 64]> new_19_cast_fp16 = transpose(perm = new_19_perm_0, x = var_1346_cast_fp16)[name = string("transpose_10")];
            tensor<fp16, [1, 2, 35, 64]> var_1358_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = var_1358_mode_0, pad = var_1358_pad_0, x = new_19_cast_fp16)[name = string("op_1358_cast_fp16")];
            tensor<int32, [8]> var_1360_pad_0 = const()[name = string("op_1360_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 6, 0, 0])];
            string var_1360_mode_0 = const()[name = string("op_1360_mode_0"), val = string("constant")];
            fp16 const_28_to_fp16 = const()[name = string("const_28_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 35, 64]> var_1360_cast_fp16 = pad(constant_val = const_28_to_fp16, mode = var_1360_mode_0, pad = var_1360_pad_0, x = input_89_cast_fp16)[name = string("op_1360_cast_fp16")];
            tensor<fp16, [1, 2, 35, 64]> v_11_cast_fp16 = add(x = var_1358_cast_fp16, y = var_1360_cast_fp16)[name = string("v_11_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> var_1362_cast_fp16 = mul(x = var_1319_cast_fp16, y = const_33_to_fp16)[name = string("op_1362_cast_fp16")];
            tensor<int32, [2]> var_1363_split_sizes_0 = const()[name = string("op_1363_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 const_42 = const()[name = string("const_42"), val = int32(3)];
            tensor<fp16, [1, 6, 8, 32]> var_1363_cast_fp16_0, tensor<fp16, [1, 6, 8, 32]> var_1363_cast_fp16_1 = split(axis = const_42, split_sizes = var_1363_split_sizes_0, x = var_1319_cast_fp16)[name = string("op_1363_cast_fp16")];
            fp16 const_29_promoted_to_fp16 = const()[name = string("const_29_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 6, 8, 32]> var_1365_cast_fp16 = mul(x = var_1363_cast_fp16_1, y = const_29_promoted_to_fp16)[name = string("op_1365_cast_fp16")];
            bool var_1367_interleave_0 = const()[name = string("op_1367_interleave_0"), val = bool(false)];
            int32 const_43 = const()[name = string("const_43"), val = int32(3)];
            tensor<fp16, [1, 6, 8, 64]> var_1367_cast_fp16 = concat(axis = const_43, interleave = var_1367_interleave_0, values = (var_1365_cast_fp16, var_1363_cast_fp16_0))[name = string("op_1367_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> var_1368_cast_fp16 = mul(x = var_1367_cast_fp16, y = const_36_to_fp16)[name = string("op_1368_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> q_cast_fp16 = add(x = var_1362_cast_fp16, y = var_1368_cast_fp16)[name = string("q_cast_fp16")];
            tensor<fp16, [1, 2, 35, 64]> var_1370_cast_fp16 = mul(x = x_101_cast_fp16, y = const_1_to_fp16)[name = string("op_1370_cast_fp16")];
            tensor<int32, [2]> var_1371_split_sizes_0 = const()[name = string("op_1371_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 var_1371_axis_0 = const()[name = string("op_1371_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 2, 35, 32]> var_1371_cast_fp16_0, tensor<fp16, [1, 2, 35, 32]> var_1371_cast_fp16_1 = split(axis = var_1371_axis_0, split_sizes = var_1371_split_sizes_0, x = x_101_cast_fp16)[name = string("op_1371_cast_fp16")];
            fp16 const_30_promoted_to_fp16 = const()[name = string("const_30_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 2, 35, 32]> var_1373_cast_fp16 = mul(x = var_1371_cast_fp16_1, y = const_30_promoted_to_fp16)[name = string("op_1373_cast_fp16")];
            bool var_1375_interleave_0 = const()[name = string("op_1375_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 2, 35, 64]> var_1375_cast_fp16 = concat(axis = var_31, interleave = var_1375_interleave_0, values = (var_1373_cast_fp16, var_1371_cast_fp16_0))[name = string("op_1375_cast_fp16")];
            tensor<fp16, [1, 2, 35, 64]> var_1376_cast_fp16 = mul(x = var_1375_cast_fp16, y = const_2_to_fp16)[name = string("op_1376_cast_fp16")];
            tensor<fp16, [1, 2, 35, 64]> k_11_cast_fp16 = add(x = var_1370_cast_fp16, y = var_1376_cast_fp16)[name = string("k_11_cast_fp16")];
            tensor<int32, [1]> var_1378_axes_0 = const()[name = string("op_1378_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 35, 64]> var_1378_cast_fp16 = expand_dims(axes = var_1378_axes_0, x = k_11_cast_fp16)[name = string("op_1378_cast_fp16")];
            tensor<int32, [5]> var_1380_reps_0 = const()[name = string("op_1380_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 35, 64]> var_1380_cast_fp16 = tile(reps = var_1380_reps_0, x = var_1378_cast_fp16)[name = string("op_1380_cast_fp16")];
            tensor<int32, [4]> concat_20 = const()[name = string("concat_20"), val = tensor<int32, [4]>([1, 8, 35, 64])];
            tensor<fp16, [1, 8, 35, 64]> k_cast_fp16 = reshape(shape = concat_20, x = var_1380_cast_fp16)[name = string("k_cast_fp16")];
            tensor<int32, [1]> var_1382_axes_0 = const()[name = string("op_1382_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 35, 64]> var_1382_cast_fp16 = expand_dims(axes = var_1382_axes_0, x = v_11_cast_fp16)[name = string("op_1382_cast_fp16")];
            tensor<int32, [5]> var_1384_reps_0 = const()[name = string("op_1384_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 35, 64]> var_1384_cast_fp16 = tile(reps = var_1384_reps_0, x = var_1382_cast_fp16)[name = string("op_1384_cast_fp16")];
            tensor<int32, [4]> concat_22 = const()[name = string("concat_22"), val = tensor<int32, [4]>([1, 8, 35, 64])];
            tensor<fp16, [1, 8, 35, 64]> v_cast_fp16 = reshape(shape = concat_22, x = var_1384_cast_fp16)[name = string("v_cast_fp16")];
            tensor<int32, [4]> transpose_4_perm_0 = const()[name = string("transpose_4_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1387_transpose_x_1 = const()[name = string("op_1387_transpose_x_1"), val = bool(false)];
            bool var_1387_transpose_y_1 = const()[name = string("op_1387_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 6, 64]> transpose_4_cast_fp16 = transpose(perm = transpose_4_perm_0, x = q_cast_fp16)[name = string("transpose_9")];
            tensor<fp16, [1, 8, 6, 35]> var_1387_cast_fp16 = matmul(transpose_x = var_1387_transpose_x_1, transpose_y = var_1387_transpose_y_1, x = transpose_4_cast_fp16, y = k_cast_fp16)[name = string("op_1387_cast_fp16")];
            fp16 var_1388_to_fp16 = const()[name = string("op_1388_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 8, 6, 35]> s_67_cast_fp16 = mul(x = var_1387_cast_fp16, y = var_1388_to_fp16)[name = string("s_67_cast_fp16")];
            tensor<fp16, [1, 8, 6, 35]> x_103_cast_fp16 = add(x = s_67_cast_fp16, y = mask)[name = string("x_103_cast_fp16")];
            tensor<int32, [1]> var_1392_axes_0 = const()[name = string("op_1392_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 6, 1]> var_1392_cast_fp16 = reduce_max(axes = var_1392_axes_0, keep_dims = var_32, x = x_103_cast_fp16)[name = string("op_1392_cast_fp16")];
            tensor<fp16, [1, 8, 6, 35]> x_105_cast_fp16 = sub(x = x_103_cast_fp16, y = var_1392_cast_fp16)[name = string("x_105_cast_fp16")];
            tensor<fp16, [1, 8, 6, 35]> x_107_cast_fp16 = exp(x = x_105_cast_fp16)[name = string("x_107_cast_fp16")];
            tensor<int32, [1]> var_1396_axes_0 = const()[name = string("op_1396_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1396_keep_dims_0 = const()[name = string("op_1396_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 6, 1]> var_1396_cast_fp16 = reduce_sum(axes = var_1396_axes_0, keep_dims = var_1396_keep_dims_0, x = x_107_cast_fp16)[name = string("op_1396_cast_fp16")];
            tensor<fp16, [1, 8, 6, 35]> var_1397_cast_fp16 = real_div(x = x_107_cast_fp16, y = var_1396_cast_fp16)[name = string("op_1397_cast_fp16")];
            bool o_13_transpose_x_0 = const()[name = string("o_13_transpose_x_0"), val = bool(false)];
            bool o_13_transpose_y_0 = const()[name = string("o_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 6, 64]> o_13_cast_fp16 = matmul(transpose_x = o_13_transpose_x_0, transpose_y = o_13_transpose_y_0, x = var_1397_cast_fp16, y = v_cast_fp16)[name = string("o_13_cast_fp16")];
            tensor<int32, [4]> o_15_perm_0 = const()[name = string("o_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [8, 512]> core_backbone_layers_6_self_attn_gate_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_gate_proj_weight_to_fp16"), val = tensor<fp16, [8, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54410304)))];
            tensor<fp16, [8]> linear_40_bias_0_to_fp16 = const()[name = string("linear_40_bias_0_to_fp16"), val = tensor<fp16, [8]>([0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0])];
            tensor<fp16, [1, 6, 8]> linear_40_cast_fp16 = linear(bias = linear_40_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_gate_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_40_cast_fp16")];
            fp16 var_1403_to_fp16 = const()[name = string("op_1403_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 6, 8]> var_1404_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_1403_to_fp16)[name = string("op_1404_cast_fp16")];
            tensor<fp16, [1, 6, 8]> var_1405_cast_fp16 = tanh(x = var_1404_cast_fp16)[name = string("op_1405_cast_fp16")];
            fp16 var_1406_to_fp16 = const()[name = string("op_1406_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 6, 8]> var_1407_cast_fp16 = add(x = var_1405_cast_fp16, y = var_1406_to_fp16)[name = string("op_1407_cast_fp16")];
            fp16 var_1408_to_fp16 = const()[name = string("op_1408_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 6, 8]> var_1409_cast_fp16 = mul(x = var_1407_cast_fp16, y = var_1408_to_fp16)[name = string("op_1409_cast_fp16")];
            tensor<int32, [1]> var_1410_axes_0 = const()[name = string("op_1410_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 8, 1]> var_1410_cast_fp16 = expand_dims(axes = var_1410_axes_0, x = var_1409_cast_fp16)[name = string("op_1410_cast_fp16")];
            tensor<fp16, [1, 6, 8, 64]> o_15_cast_fp16 = transpose(perm = o_15_perm_0, x = o_13_cast_fp16)[name = string("transpose_8")];
            tensor<fp16, [1, 6, 8, 64]> o_cast_fp16 = mul(x = o_15_cast_fp16, y = var_1410_cast_fp16)[name = string("o_cast_fp16")];
            tensor<int32, [3]> concat_23 = const()[name = string("concat_23"), val = tensor<int32, [3]>([1, 6, 512])];
            tensor<fp16, [1, 6, 512]> input_91_cast_fp16 = reshape(shape = concat_23, x = o_cast_fp16)[name = string("input_91_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54418560)))];
            tensor<fp16, [1, 6, 512]> linear_41_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_o_proj_weight_to_fp16, x = input_91_cast_fp16)[name = string("linear_41_cast_fp16")];
            tensor<int32, [4]> var_1417_begin_0 = const()[name = string("op_1417_begin_0"), val = tensor<int32, [4]>([0, 0, 6, 0])];
            tensor<int32, [4]> var_1417_end_0 = const()[name = string("op_1417_end_0"), val = tensor<int32, [4]>([1, 2, 35, 64])];
            tensor<bool, [4]> var_1417_end_mask_0 = const()[name = string("op_1417_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_1417_cast_fp16 = slice_by_index(begin = var_1417_begin_0, end = var_1417_end_0, end_mask = var_1417_end_mask_0, x = x_101_cast_fp16)[name = string("op_1417_cast_fp16")];
            tensor<int32, [4]> var_1420_begin_0 = const()[name = string("op_1420_begin_0"), val = tensor<int32, [4]>([0, 0, 6, 0])];
            tensor<int32, [4]> var_1420_end_0 = const()[name = string("op_1420_end_0"), val = tensor<int32, [4]>([1, 2, 35, 64])];
            tensor<bool, [4]> var_1420_end_mask_0 = const()[name = string("op_1420_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_1420_cast_fp16 = slice_by_index(begin = var_1420_begin_0, end = var_1420_end_0, end_mask = var_1420_end_mask_0, x = v_11_cast_fp16)[name = string("op_1420_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1422_cast_fp16 = abs(x = linear_41_cast_fp16)[name = string("op_1422_cast_fp16")];
            tensor<int32, [1]> var_1424_axes_0 = const()[name = string("op_1424_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1424_cast_fp16 = reduce_max(axes = var_1424_axes_0, keep_dims = var_32, x = var_1422_cast_fp16)[name = string("op_1424_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_69_cast_fp16 = maximum(x = var_1424_cast_fp16, y = var_33_to_fp16)[name = string("s_69_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_71_cast_fp16 = real_div(x = linear_41_cast_fp16, y = s_69_cast_fp16)[name = string("y_71_cast_fp16")];
            fp16 var_34_promoted_31_to_fp16 = const()[name = string("op_34_promoted_31_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1427_cast_fp16 = pow(x = y_71_cast_fp16, y = var_34_promoted_31_to_fp16)[name = string("op_1427_cast_fp16")];
            tensor<int32, [1]> var_1429_axes_0 = const()[name = string("op_1429_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1429_keep_dims_0 = const()[name = string("op_1429_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1429_cast_fp16 = reduce_mean(axes = var_1429_axes_0, keep_dims = var_1429_keep_dims_0, x = var_1427_cast_fp16)[name = string("op_1429_cast_fp16")];
            fp32 var_1430_epsilon_0 = const()[name = string("op_1430_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1430_cast_fp16 = inverse(epsilon = var_1430_epsilon_0, x = s_69_cast_fp16)[name = string("op_1430_cast_fp16")];
            fp16 var_1431_to_fp16 = const()[name = string("op_1431_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1432_cast_fp16 = mul(x = var_1430_cast_fp16, y = var_1431_to_fp16)[name = string("op_1432_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1433_cast_fp16 = real_div(x = var_1432_cast_fp16, y = s_69_cast_fp16)[name = string("op_1433_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1434_cast_fp16 = add(x = var_1429_cast_fp16, y = var_1433_cast_fp16)[name = string("op_1434_cast_fp16")];
            fp32 var_1435_epsilon_0 = const()[name = string("op_1435_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1435_cast_fp16 = rsqrt(epsilon = var_1435_epsilon_0, x = var_1434_cast_fp16)[name = string("op_1435_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1436_cast_fp16 = mul(x = y_71_cast_fp16, y = var_1435_cast_fp16)[name = string("op_1436_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_6_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54942912)))];
            tensor<fp16, [1, 6, 512]> a_13_cast_fp16 = mul(x = core_backbone_layers_6_post_operator_norm_weight_to_fp16, y = var_1436_cast_fp16)[name = string("a_13_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_113_cast_fp16 = add(x = x_93_cast_fp16, y = a_13_cast_fp16)[name = string("x_113_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1440_cast_fp16 = abs(x = x_113_cast_fp16)[name = string("op_1440_cast_fp16")];
            tensor<int32, [1]> var_1442_axes_0 = const()[name = string("op_1442_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1442_cast_fp16 = reduce_max(axes = var_1442_axes_0, keep_dims = var_32, x = var_1440_cast_fp16)[name = string("op_1442_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_71_cast_fp16 = maximum(x = var_1442_cast_fp16, y = var_33_to_fp16)[name = string("s_71_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_73_cast_fp16 = real_div(x = x_113_cast_fp16, y = s_71_cast_fp16)[name = string("y_73_cast_fp16")];
            fp16 var_34_promoted_32_to_fp16 = const()[name = string("op_34_promoted_32_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1445_cast_fp16 = pow(x = y_73_cast_fp16, y = var_34_promoted_32_to_fp16)[name = string("op_1445_cast_fp16")];
            tensor<int32, [1]> var_1447_axes_0 = const()[name = string("op_1447_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1447_keep_dims_0 = const()[name = string("op_1447_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1447_cast_fp16 = reduce_mean(axes = var_1447_axes_0, keep_dims = var_1447_keep_dims_0, x = var_1445_cast_fp16)[name = string("op_1447_cast_fp16")];
            fp32 var_1448_epsilon_0 = const()[name = string("op_1448_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1448_cast_fp16 = inverse(epsilon = var_1448_epsilon_0, x = s_71_cast_fp16)[name = string("op_1448_cast_fp16")];
            fp16 var_1449_to_fp16 = const()[name = string("op_1449_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1450_cast_fp16 = mul(x = var_1448_cast_fp16, y = var_1449_to_fp16)[name = string("op_1450_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1451_cast_fp16 = real_div(x = var_1450_cast_fp16, y = s_71_cast_fp16)[name = string("op_1451_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1452_cast_fp16 = add(x = var_1447_cast_fp16, y = var_1451_cast_fp16)[name = string("op_1452_cast_fp16")];
            fp32 var_1453_epsilon_0 = const()[name = string("op_1453_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1453_cast_fp16 = rsqrt(epsilon = var_1453_epsilon_0, x = var_1452_cast_fp16)[name = string("op_1453_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1454_cast_fp16 = mul(x = y_73_cast_fp16, y = var_1453_cast_fp16)[name = string("op_1454_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_6_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54944000)))];
            tensor<fp16, [1, 6, 512]> input_93_cast_fp16 = mul(x = core_backbone_layers_6_ffn_norm_weight_to_fp16, y = var_1454_cast_fp16)[name = string("input_93_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_6_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54945088)))];
            tensor<fp16, [1, 6, 2304]> linear_42_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w1_weight_to_fp16, x = input_93_cast_fp16)[name = string("linear_42_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_6_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57304448)))];
            tensor<fp16, [1, 6, 2304]> linear_43_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w3_weight_to_fp16, x = input_93_cast_fp16)[name = string("linear_43_cast_fp16")];
            fp16 var_1463_to_fp16 = const()[name = string("op_1463_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 6, 2304]> var_1464_cast_fp16 = mul(x = linear_42_cast_fp16, y = var_1463_to_fp16)[name = string("op_1464_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_1467_cast_fp16 = tanh(x = var_1464_cast_fp16)[name = string("op_1467_cast_fp16")];
            fp16 var_1468_to_fp16 = const()[name = string("op_1468_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 6, 2304]> var_1469_cast_fp16 = add(x = var_1467_cast_fp16, y = var_1468_to_fp16)[name = string("op_1469_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_1470_cast_fp16 = mul(x = var_1464_cast_fp16, y = var_1469_cast_fp16)[name = string("op_1470_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> input_95_cast_fp16 = mul(x = var_1470_cast_fp16, y = linear_43_cast_fp16)[name = string("input_95_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_6_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59663808)))];
            tensor<fp16, [1, 6, 512]> linear_44_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w2_weight_to_fp16, x = input_95_cast_fp16)[name = string("linear_44_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1475_cast_fp16 = abs(x = linear_44_cast_fp16)[name = string("op_1475_cast_fp16")];
            tensor<int32, [1]> var_1477_axes_0 = const()[name = string("op_1477_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1477_cast_fp16 = reduce_max(axes = var_1477_axes_0, keep_dims = var_32, x = var_1475_cast_fp16)[name = string("op_1477_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_73_cast_fp16 = maximum(x = var_1477_cast_fp16, y = var_33_to_fp16)[name = string("s_73_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_75_cast_fp16 = real_div(x = linear_44_cast_fp16, y = s_73_cast_fp16)[name = string("y_75_cast_fp16")];
            fp16 var_34_promoted_33_to_fp16 = const()[name = string("op_34_promoted_33_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1480_cast_fp16 = pow(x = y_75_cast_fp16, y = var_34_promoted_33_to_fp16)[name = string("op_1480_cast_fp16")];
            tensor<int32, [1]> var_1482_axes_0 = const()[name = string("op_1482_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1482_keep_dims_0 = const()[name = string("op_1482_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1482_cast_fp16 = reduce_mean(axes = var_1482_axes_0, keep_dims = var_1482_keep_dims_0, x = var_1480_cast_fp16)[name = string("op_1482_cast_fp16")];
            fp32 var_1483_epsilon_0 = const()[name = string("op_1483_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1483_cast_fp16 = inverse(epsilon = var_1483_epsilon_0, x = s_73_cast_fp16)[name = string("op_1483_cast_fp16")];
            fp16 var_1484_to_fp16 = const()[name = string("op_1484_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1485_cast_fp16 = mul(x = var_1483_cast_fp16, y = var_1484_to_fp16)[name = string("op_1485_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1486_cast_fp16 = real_div(x = var_1485_cast_fp16, y = s_73_cast_fp16)[name = string("op_1486_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1487_cast_fp16 = add(x = var_1482_cast_fp16, y = var_1486_cast_fp16)[name = string("op_1487_cast_fp16")];
            fp32 var_1488_epsilon_0 = const()[name = string("op_1488_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1488_cast_fp16 = rsqrt(epsilon = var_1488_epsilon_0, x = var_1487_cast_fp16)[name = string("op_1488_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1489_cast_fp16 = mul(x = y_75_cast_fp16, y = var_1488_cast_fp16)[name = string("op_1489_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_6_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62023168)))];
            tensor<fp16, [1, 6, 512]> f_13_cast_fp16 = mul(x = core_backbone_layers_6_post_ffn_norm_weight_to_fp16, y = var_1489_cast_fp16)[name = string("f_13_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_119_cast_fp16 = add(x = x_113_cast_fp16, y = f_13_cast_fp16)[name = string("x_119_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1493_cast_fp16 = abs(x = x_119_cast_fp16)[name = string("op_1493_cast_fp16")];
            tensor<int32, [1]> var_1495_axes_0 = const()[name = string("op_1495_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1495_cast_fp16 = reduce_max(axes = var_1495_axes_0, keep_dims = var_32, x = var_1493_cast_fp16)[name = string("op_1495_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_75_cast_fp16 = maximum(x = var_1495_cast_fp16, y = var_33_to_fp16)[name = string("s_75_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_77_cast_fp16 = real_div(x = x_119_cast_fp16, y = s_75_cast_fp16)[name = string("y_77_cast_fp16")];
            fp16 var_34_promoted_34_to_fp16 = const()[name = string("op_34_promoted_34_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1498_cast_fp16 = pow(x = y_77_cast_fp16, y = var_34_promoted_34_to_fp16)[name = string("op_1498_cast_fp16")];
            tensor<int32, [1]> var_1500_axes_0 = const()[name = string("op_1500_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1500_keep_dims_0 = const()[name = string("op_1500_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1500_cast_fp16 = reduce_mean(axes = var_1500_axes_0, keep_dims = var_1500_keep_dims_0, x = var_1498_cast_fp16)[name = string("op_1500_cast_fp16")];
            fp32 var_1501_epsilon_0 = const()[name = string("op_1501_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1501_cast_fp16 = inverse(epsilon = var_1501_epsilon_0, x = s_75_cast_fp16)[name = string("op_1501_cast_fp16")];
            fp16 var_1502_to_fp16 = const()[name = string("op_1502_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1503_cast_fp16 = mul(x = var_1501_cast_fp16, y = var_1502_to_fp16)[name = string("op_1503_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1504_cast_fp16 = real_div(x = var_1503_cast_fp16, y = s_75_cast_fp16)[name = string("op_1504_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1505_cast_fp16 = add(x = var_1500_cast_fp16, y = var_1504_cast_fp16)[name = string("op_1505_cast_fp16")];
            fp32 var_1506_epsilon_0 = const()[name = string("op_1506_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1506_cast_fp16 = rsqrt(epsilon = var_1506_epsilon_0, x = var_1505_cast_fp16)[name = string("op_1506_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1507_cast_fp16 = mul(x = y_77_cast_fp16, y = var_1506_cast_fp16)[name = string("op_1507_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_7_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62024256)))];
            tensor<fp16, [1, 6, 512]> input_97_cast_fp16 = mul(x = core_backbone_layers_7_operator_norm_weight_to_fp16, y = var_1507_cast_fp16)[name = string("input_97_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_7_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62025344)))];
            tensor<fp16, [1, 6, 1536]> linear_45_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_7_conv_in_proj_weight_to_fp16, x = input_97_cast_fp16)[name = string("linear_45_cast_fp16")];
            tensor<int32, [3]> bcx_perm_0 = const()[name = string("bcx_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_23_begin_0 = const()[name = string("b_23_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_23_end_0 = const()[name = string("b_23_end_0"), val = tensor<int32, [3]>([1, 512, 6])];
            tensor<bool, [3]> b_23_end_mask_0 = const()[name = string("b_23_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 6]> bcx_cast_fp16 = transpose(perm = bcx_perm_0, x = linear_45_cast_fp16)[name = string("transpose_7")];
            tensor<fp16, [1, 512, 6]> b_23_cast_fp16 = slice_by_index(begin = b_23_begin_0, end = b_23_end_0, end_mask = b_23_end_mask_0, x = bcx_cast_fp16)[name = string("b_23_cast_fp16")];
            tensor<int32, [3]> c_begin_0 = const()[name = string("c_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_end_0 = const()[name = string("c_end_0"), val = tensor<int32, [3]>([1, 1024, 6])];
            tensor<bool, [3]> c_end_mask_0 = const()[name = string("c_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 6]> c_cast_fp16 = slice_by_index(begin = c_begin_0, end = c_end_0, end_mask = c_end_mask_0, x = bcx_cast_fp16)[name = string("c_cast_fp16")];
            tensor<int32, [3]> xx_begin_0 = const()[name = string("xx_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_end_0 = const()[name = string("xx_end_0"), val = tensor<int32, [3]>([1, 1536, 6])];
            tensor<bool, [3]> xx_end_mask_0 = const()[name = string("xx_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> xx_cast_fp16 = slice_by_index(begin = xx_begin_0, end = xx_end_0, end_mask = xx_end_mask_0, x = bcx_cast_fp16)[name = string("xx_cast_fp16")];
            tensor<fp16, [1, 512, 6]> new_cast_fp16 = mul(x = b_23_cast_fp16, y = xx_cast_fp16)[name = string("new_cast_fp16")];
            tensor<int32, [4]> input_101_begin_0 = const()[name = string("input_101_begin_0"), val = tensor<int32, [4]>([4, 0, 0, 0])];
            tensor<int32, [4]> input_101_end_0 = const()[name = string("input_101_end_0"), val = tensor<int32, [4]>([5, 1, 512, 6])];
            tensor<bool, [4]> input_101_end_mask_0 = const()[name = string("input_101_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_101_squeeze_mask_0 = const()[name = string("input_101_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_101_cast_fp16 = slice_by_index(begin = input_101_begin_0, end = input_101_end_0, end_mask = input_101_end_mask_0, squeeze_mask = input_101_squeeze_mask_0, x = conv)[name = string("input_101_cast_fp16")];
            tensor<int32, [6]> var_1522_pad_0 = const()[name = string("op_1522_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_1522_mode_0 = const()[name = string("op_1522_mode_0"), val = string("constant")];
            fp16 const_31_to_fp16 = const()[name = string("const_31_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 12]> var_1522_cast_fp16 = pad(constant_val = const_31_to_fp16, mode = var_1522_mode_0, pad = var_1522_pad_0, x = new_cast_fp16)[name = string("op_1522_cast_fp16")];
            tensor<int32, [6]> var_1524_pad_0 = const()[name = string("op_1524_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 6])];
            string var_1524_mode_0 = const()[name = string("op_1524_mode_0"), val = string("constant")];
            fp16 const_32_to_fp16 = const()[name = string("const_32_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 12]> var_1524_cast_fp16 = pad(constant_val = const_32_to_fp16, mode = var_1524_mode_0, pad = var_1524_pad_0, x = input_101_cast_fp16)[name = string("op_1524_cast_fp16")];
            tensor<fp16, [1, 512, 12]> win_17_cast_fp16 = add(x = var_1522_cast_fp16, y = var_1524_cast_fp16)[name = string("win_17_cast_fp16")];
            tensor<int32, [3]> var_1526_begin_0 = const()[name = string("op_1526_begin_0"), val = tensor<int32, [3]>([0, 0, 6])];
            tensor<int32, [3]> var_1526_end_0 = const()[name = string("op_1526_end_0"), val = tensor<int32, [3]>([1, 512, 12])];
            tensor<bool, [3]> var_1526_end_mask_0 = const()[name = string("op_1526_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_1526_cast_fp16 = slice_by_index(begin = var_1526_begin_0, end = var_1526_end_0, end_mask = var_1526_end_mask_0, x = win_17_cast_fp16)[name = string("op_1526_cast_fp16")];
            string var_1532_pad_type_0 = const()[name = string("op_1532_pad_type_0"), val = string("valid")];
            int32 var_1532_groups_0 = const()[name = string("op_1532_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_1532_strides_0 = const()[name = string("op_1532_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1532_pad_0 = const()[name = string("op_1532_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1532_dilations_0 = const()[name = string("op_1532_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_7_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63598272)))];
            tensor<fp16, [1, 512, 6]> var_1532_cast_fp16 = conv(dilations = var_1532_dilations_0, groups = var_1532_groups_0, pad = var_1532_pad_0, pad_type = var_1532_pad_type_0, strides = var_1532_strides_0, weight = core_backbone_layers_7_conv_conv_weight_to_fp16, x = win_17_cast_fp16)[name = string("op_1532_cast_fp16")];
            tensor<fp16, [1, 512, 6]> var_1534_cast_fp16 = mul(x = c_cast_fp16, y = var_1532_cast_fp16)[name = string("op_1534_cast_fp16")];
            tensor<int32, [3]> input_103_perm_0 = const()[name = string("input_103_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_7_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63605504)))];
            tensor<fp16, [1, 6, 512]> input_103_cast_fp16 = transpose(perm = input_103_perm_0, x = var_1534_cast_fp16)[name = string("transpose_6")];
            tensor<fp16, [1, 6, 512]> linear_46_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_7_conv_out_proj_weight_to_fp16, x = input_103_cast_fp16)[name = string("linear_46_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1539_cast_fp16 = abs(x = linear_46_cast_fp16)[name = string("op_1539_cast_fp16")];
            tensor<int32, [1]> var_1541_axes_0 = const()[name = string("op_1541_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1541_cast_fp16 = reduce_max(axes = var_1541_axes_0, keep_dims = var_32, x = var_1539_cast_fp16)[name = string("op_1541_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_77_cast_fp16 = maximum(x = var_1541_cast_fp16, y = var_33_to_fp16)[name = string("s_77_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_81_cast_fp16 = real_div(x = linear_46_cast_fp16, y = s_77_cast_fp16)[name = string("y_81_cast_fp16")];
            fp16 var_34_promoted_35_to_fp16 = const()[name = string("op_34_promoted_35_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1544_cast_fp16 = pow(x = y_81_cast_fp16, y = var_34_promoted_35_to_fp16)[name = string("op_1544_cast_fp16")];
            tensor<int32, [1]> var_1546_axes_0 = const()[name = string("op_1546_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1546_keep_dims_0 = const()[name = string("op_1546_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1546_cast_fp16 = reduce_mean(axes = var_1546_axes_0, keep_dims = var_1546_keep_dims_0, x = var_1544_cast_fp16)[name = string("op_1546_cast_fp16")];
            fp32 var_1547_epsilon_0 = const()[name = string("op_1547_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1547_cast_fp16 = inverse(epsilon = var_1547_epsilon_0, x = s_77_cast_fp16)[name = string("op_1547_cast_fp16")];
            fp16 var_1548_to_fp16 = const()[name = string("op_1548_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1549_cast_fp16 = mul(x = var_1547_cast_fp16, y = var_1548_to_fp16)[name = string("op_1549_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1550_cast_fp16 = real_div(x = var_1549_cast_fp16, y = s_77_cast_fp16)[name = string("op_1550_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1551_cast_fp16 = add(x = var_1546_cast_fp16, y = var_1550_cast_fp16)[name = string("op_1551_cast_fp16")];
            fp32 var_1552_epsilon_0 = const()[name = string("op_1552_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1552_cast_fp16 = rsqrt(epsilon = var_1552_epsilon_0, x = var_1551_cast_fp16)[name = string("op_1552_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1553_cast_fp16 = mul(x = y_81_cast_fp16, y = var_1552_cast_fp16)[name = string("op_1553_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_7_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64129856)))];
            tensor<fp16, [1, 6, 512]> a_cast_fp16 = mul(x = core_backbone_layers_7_post_operator_norm_weight_to_fp16, y = var_1553_cast_fp16)[name = string("a_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_123_cast_fp16 = add(x = x_119_cast_fp16, y = a_cast_fp16)[name = string("x_123_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1557_cast_fp16 = abs(x = x_123_cast_fp16)[name = string("op_1557_cast_fp16")];
            tensor<int32, [1]> var_1559_axes_0 = const()[name = string("op_1559_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1559_cast_fp16 = reduce_max(axes = var_1559_axes_0, keep_dims = var_32, x = var_1557_cast_fp16)[name = string("op_1559_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_79_cast_fp16 = maximum(x = var_1559_cast_fp16, y = var_33_to_fp16)[name = string("s_79_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_83_cast_fp16 = real_div(x = x_123_cast_fp16, y = s_79_cast_fp16)[name = string("y_83_cast_fp16")];
            fp16 var_34_promoted_36_to_fp16 = const()[name = string("op_34_promoted_36_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1562_cast_fp16 = pow(x = y_83_cast_fp16, y = var_34_promoted_36_to_fp16)[name = string("op_1562_cast_fp16")];
            tensor<int32, [1]> var_1564_axes_0 = const()[name = string("op_1564_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1564_keep_dims_0 = const()[name = string("op_1564_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1564_cast_fp16 = reduce_mean(axes = var_1564_axes_0, keep_dims = var_1564_keep_dims_0, x = var_1562_cast_fp16)[name = string("op_1564_cast_fp16")];
            fp32 var_1565_epsilon_0 = const()[name = string("op_1565_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1565_cast_fp16 = inverse(epsilon = var_1565_epsilon_0, x = s_79_cast_fp16)[name = string("op_1565_cast_fp16")];
            fp16 var_1566_to_fp16 = const()[name = string("op_1566_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1567_cast_fp16 = mul(x = var_1565_cast_fp16, y = var_1566_to_fp16)[name = string("op_1567_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1568_cast_fp16 = real_div(x = var_1567_cast_fp16, y = s_79_cast_fp16)[name = string("op_1568_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1569_cast_fp16 = add(x = var_1564_cast_fp16, y = var_1568_cast_fp16)[name = string("op_1569_cast_fp16")];
            fp32 var_1570_epsilon_0 = const()[name = string("op_1570_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1570_cast_fp16 = rsqrt(epsilon = var_1570_epsilon_0, x = var_1569_cast_fp16)[name = string("op_1570_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1571_cast_fp16 = mul(x = y_83_cast_fp16, y = var_1570_cast_fp16)[name = string("op_1571_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_7_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64130944)))];
            tensor<fp16, [1, 6, 512]> input_105_cast_fp16 = mul(x = core_backbone_layers_7_ffn_norm_weight_to_fp16, y = var_1571_cast_fp16)[name = string("input_105_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_7_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64132032)))];
            tensor<fp16, [1, 6, 2304]> linear_47_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w1_weight_to_fp16, x = input_105_cast_fp16)[name = string("linear_47_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_7_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66491392)))];
            tensor<fp16, [1, 6, 2304]> linear_48_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w3_weight_to_fp16, x = input_105_cast_fp16)[name = string("linear_48_cast_fp16")];
            fp16 var_1580_to_fp16 = const()[name = string("op_1580_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 6, 2304]> var_1581_cast_fp16 = mul(x = linear_47_cast_fp16, y = var_1580_to_fp16)[name = string("op_1581_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_1584_cast_fp16 = tanh(x = var_1581_cast_fp16)[name = string("op_1584_cast_fp16")];
            fp16 var_1585_to_fp16 = const()[name = string("op_1585_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 6, 2304]> var_1586_cast_fp16 = add(x = var_1584_cast_fp16, y = var_1585_to_fp16)[name = string("op_1586_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> var_1587_cast_fp16 = mul(x = var_1581_cast_fp16, y = var_1586_cast_fp16)[name = string("op_1587_cast_fp16")];
            tensor<fp16, [1, 6, 2304]> input_cast_fp16 = mul(x = var_1587_cast_fp16, y = linear_48_cast_fp16)[name = string("input_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_7_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68850752)))];
            tensor<fp16, [1, 6, 512]> linear_49_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w2_weight_to_fp16, x = input_cast_fp16)[name = string("linear_49_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1592_cast_fp16 = abs(x = linear_49_cast_fp16)[name = string("op_1592_cast_fp16")];
            tensor<int32, [1]> var_1594_axes_0 = const()[name = string("op_1594_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1594_cast_fp16 = reduce_max(axes = var_1594_axes_0, keep_dims = var_32, x = var_1592_cast_fp16)[name = string("op_1594_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_81_cast_fp16 = maximum(x = var_1594_cast_fp16, y = var_33_to_fp16)[name = string("s_81_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_85_cast_fp16 = real_div(x = linear_49_cast_fp16, y = s_81_cast_fp16)[name = string("y_85_cast_fp16")];
            fp16 var_34_promoted_37_to_fp16 = const()[name = string("op_34_promoted_37_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1597_cast_fp16 = pow(x = y_85_cast_fp16, y = var_34_promoted_37_to_fp16)[name = string("op_1597_cast_fp16")];
            tensor<int32, [1]> var_1599_axes_0 = const()[name = string("op_1599_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1599_keep_dims_0 = const()[name = string("op_1599_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1599_cast_fp16 = reduce_mean(axes = var_1599_axes_0, keep_dims = var_1599_keep_dims_0, x = var_1597_cast_fp16)[name = string("op_1599_cast_fp16")];
            fp32 var_1600_epsilon_0 = const()[name = string("op_1600_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1600_cast_fp16 = inverse(epsilon = var_1600_epsilon_0, x = s_81_cast_fp16)[name = string("op_1600_cast_fp16")];
            fp16 var_1601_to_fp16 = const()[name = string("op_1601_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1602_cast_fp16 = mul(x = var_1600_cast_fp16, y = var_1601_to_fp16)[name = string("op_1602_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1603_cast_fp16 = real_div(x = var_1602_cast_fp16, y = s_81_cast_fp16)[name = string("op_1603_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1604_cast_fp16 = add(x = var_1599_cast_fp16, y = var_1603_cast_fp16)[name = string("op_1604_cast_fp16")];
            fp32 var_1605_epsilon_0 = const()[name = string("op_1605_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1605_cast_fp16 = rsqrt(epsilon = var_1605_epsilon_0, x = var_1604_cast_fp16)[name = string("op_1605_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1606_cast_fp16 = mul(x = y_85_cast_fp16, y = var_1605_cast_fp16)[name = string("op_1606_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_7_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71210112)))];
            tensor<fp16, [1, 6, 512]> f_cast_fp16 = mul(x = core_backbone_layers_7_post_ffn_norm_weight_to_fp16, y = var_1606_cast_fp16)[name = string("f_cast_fp16")];
            tensor<fp16, [1, 6, 512]> x_cast_fp16 = add(x = x_123_cast_fp16, y = f_cast_fp16)[name = string("x_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1610_cast_fp16 = abs(x = x_cast_fp16)[name = string("op_1610_cast_fp16")];
            tensor<int32, [1]> var_1612_axes_0 = const()[name = string("op_1612_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 6, 1]> var_1612_cast_fp16 = reduce_max(axes = var_1612_axes_0, keep_dims = var_32, x = var_1610_cast_fp16)[name = string("op_1612_cast_fp16")];
            tensor<fp16, [1, 6, 1]> s_cast_fp16 = maximum(x = var_1612_cast_fp16, y = var_33_to_fp16)[name = string("s_cast_fp16")];
            tensor<fp16, [1, 6, 512]> y_cast_fp16 = real_div(x = x_cast_fp16, y = s_cast_fp16)[name = string("y_cast_fp16")];
            fp16 var_34_promoted_38_to_fp16 = const()[name = string("op_34_promoted_38_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 6, 512]> var_1615_cast_fp16 = pow(x = y_cast_fp16, y = var_34_promoted_38_to_fp16)[name = string("op_1615_cast_fp16")];
            tensor<int32, [1]> var_1617_axes_0 = const()[name = string("op_1617_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1617_keep_dims_0 = const()[name = string("op_1617_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 6, 1]> var_1617_cast_fp16 = reduce_mean(axes = var_1617_axes_0, keep_dims = var_1617_keep_dims_0, x = var_1615_cast_fp16)[name = string("op_1617_cast_fp16")];
            fp32 var_1618_epsilon_0 = const()[name = string("op_1618_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 6, 1]> var_1618_cast_fp16 = inverse(epsilon = var_1618_epsilon_0, x = s_cast_fp16)[name = string("op_1618_cast_fp16")];
            fp16 var_1619_to_fp16 = const()[name = string("op_1619_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 6, 1]> var_1620_cast_fp16 = mul(x = var_1618_cast_fp16, y = var_1619_to_fp16)[name = string("op_1620_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1621_cast_fp16 = real_div(x = var_1620_cast_fp16, y = s_cast_fp16)[name = string("op_1621_cast_fp16")];
            tensor<fp16, [1, 6, 1]> var_1622_cast_fp16 = add(x = var_1617_cast_fp16, y = var_1621_cast_fp16)[name = string("op_1622_cast_fp16")];
            fp32 var_1623_epsilon_0 = const()[name = string("op_1623_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 6, 1]> var_1623_cast_fp16 = rsqrt(epsilon = var_1623_epsilon_0, x = var_1622_cast_fp16)[name = string("op_1623_cast_fp16")];
            tensor<fp16, [1, 6, 512]> var_1624_cast_fp16 = mul(x = y_cast_fp16, y = var_1623_cast_fp16)[name = string("op_1624_cast_fp16")];
            tensor<fp16, [512]> core_backbone_embedding_norm_weight_to_fp16 = const()[name = string("core_backbone_embedding_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71211200)))];
            tensor<fp16, [1, 6, 512]> hidden = mul(x = core_backbone_embedding_norm_weight_to_fp16, y = var_1624_cast_fp16)[name = string("op_1625_cast_fp16")];
            int32 var_1627_axis_0 = const()[name = string("op_1627_axis_0"), val = int32(0)];
            tensor<fp16, [5, 1, 512, 6]> conv_out = stack(axis = var_1627_axis_0, values = (var_428_cast_fp16, var_545_cast_fp16, var_872_cast_fp16, var_1199_cast_fp16, var_1526_cast_fp16))[name = string("op_1627_cast_fp16")];
            int32 var_1629_axis_0 = const()[name = string("op_1629_axis_0"), val = int32(0)];
            tensor<fp16, [3, 1, 2, 29, 64]> k_out = stack(axis = var_1629_axis_0, values = (var_763_cast_fp16, var_1090_cast_fp16, var_1417_cast_fp16))[name = string("op_1629_cast_fp16")];
            int32 var_1631_axis_0 = const()[name = string("op_1631_axis_0"), val = int32(0)];
            tensor<fp16, [3, 1, 2, 29, 64]> v_out = stack(axis = var_1631_axis_0, values = (var_766_cast_fp16, var_1093_cast_fp16, var_1420_cast_fp16))[name = string("op_1631_cast_fp16")];
        } -> (hidden, conv_out, k_out, v_out);
    func f25<ios18>(tensor<fp16, [5, 1, 512, 6]> conv, tensor<fp16, [3, 1, 2, 29, 64]> k, tensor<fp16, [1, 25, 48]> lat48, tensor<fp16, [1, 1, 150, 179]> mask, tensor<fp16, [3, 1, 2, 29, 64]> v) {
            tensor<fp16, [512, 48]> proj_weight_to_fp16 = const()[name = string("proj_weight_to_fp16"), val = tensor<fp16, [512, 48]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            tensor<fp16, [512]> proj_bias_to_fp16 = const()[name = string("proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49280)))];
            tensor<fp16, [1, 25, 512]> linear_0_cast_fp16 = linear(bias = proj_bias_to_fp16, weight = proj_weight_to_fp16, x = lat48)[name = string("linear_0_cast_fp16")];
            bool x_1_transpose_x_0 = const()[name = string("x_1_transpose_x_0"), val = bool(false)];
            bool x_1_transpose_y_0 = const()[name = string("x_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [150, 25]> const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = tensor<fp16, [150, 25]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71232512)))];
            tensor<fp16, [1, 150, 512]> x_1_cast_fp16 = matmul(transpose_x = x_1_transpose_x_0, transpose_y = x_1_transpose_y_0, x = const_0_to_fp16, y = linear_0_cast_fp16)[name = string("x_1_cast_fp16")];
            int32 var_32 = const()[name = string("op_32"), val = int32(-1)];
            bool var_33 = const()[name = string("op_33"), val = bool(true)];
            tensor<fp16, [1, 150, 512]> var_396_cast_fp16 = abs(x = x_1_cast_fp16)[name = string("op_396_cast_fp16")];
            tensor<int32, [1]> var_398_axes_0 = const()[name = string("op_398_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_398_cast_fp16 = reduce_max(axes = var_398_axes_0, keep_dims = var_33, x = var_396_cast_fp16)[name = string("op_398_cast_fp16")];
            fp16 var_34_to_fp16 = const()[name = string("op_34_to_fp16"), val = fp16(0x1.a38p-14)];
            tensor<fp16, [1, 150, 1]> s_1_cast_fp16 = maximum(x = var_398_cast_fp16, y = var_34_to_fp16)[name = string("s_1_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_1_cast_fp16 = real_div(x = x_1_cast_fp16, y = s_1_cast_fp16)[name = string("y_1_cast_fp16")];
            fp16 var_35_promoted_to_fp16 = const()[name = string("op_35_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_401_cast_fp16 = pow(x = y_1_cast_fp16, y = var_35_promoted_to_fp16)[name = string("op_401_cast_fp16")];
            tensor<int32, [1]> var_403_axes_0 = const()[name = string("op_403_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_403_keep_dims_0 = const()[name = string("op_403_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_403_cast_fp16 = reduce_mean(axes = var_403_axes_0, keep_dims = var_403_keep_dims_0, x = var_401_cast_fp16)[name = string("op_403_cast_fp16")];
            fp32 var_404_epsilon_0 = const()[name = string("op_404_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_404_cast_fp16 = inverse(epsilon = var_404_epsilon_0, x = s_1_cast_fp16)[name = string("op_404_cast_fp16")];
            fp16 var_405_to_fp16 = const()[name = string("op_405_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_406_cast_fp16 = mul(x = var_404_cast_fp16, y = var_405_to_fp16)[name = string("op_406_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_407_cast_fp16 = real_div(x = var_406_cast_fp16, y = s_1_cast_fp16)[name = string("op_407_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_408_cast_fp16 = add(x = var_403_cast_fp16, y = var_407_cast_fp16)[name = string("op_408_cast_fp16")];
            fp32 var_409_epsilon_0 = const()[name = string("op_409_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_409_cast_fp16 = rsqrt(epsilon = var_409_epsilon_0, x = var_408_cast_fp16)[name = string("op_409_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_410_cast_fp16 = mul(x = y_1_cast_fp16, y = var_409_cast_fp16)[name = string("op_410_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_0_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50368)))];
            tensor<fp16, [1, 150, 512]> input_1_cast_fp16 = mul(x = core_backbone_layers_0_operator_norm_weight_to_fp16, y = var_410_cast_fp16)[name = string("input_1_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_0_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51456)))];
            tensor<fp16, [1536]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1624384)))];
            tensor<fp16, [1, 150, 1536]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_0_conv_in_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<int32, [3]> bcx_1_perm_0 = const()[name = string("bcx_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_1_begin_0 = const()[name = string("b_1_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_1_end_0 = const()[name = string("b_1_end_0"), val = tensor<int32, [3]>([1, 512, 150])];
            tensor<bool, [3]> b_1_end_mask_0 = const()[name = string("b_1_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 150]> bcx_1_cast_fp16 = transpose(perm = bcx_1_perm_0, x = linear_1_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [1, 512, 150]> b_1_cast_fp16 = slice_by_index(begin = b_1_begin_0, end = b_1_end_0, end_mask = b_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("b_1_cast_fp16")];
            tensor<int32, [3]> c_1_begin_0 = const()[name = string("c_1_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_1_end_0 = const()[name = string("c_1_end_0"), val = tensor<int32, [3]>([1, 1024, 150])];
            tensor<bool, [3]> c_1_end_mask_0 = const()[name = string("c_1_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 150]> c_1_cast_fp16 = slice_by_index(begin = c_1_begin_0, end = c_1_end_0, end_mask = c_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("c_1_cast_fp16")];
            tensor<int32, [3]> xx_1_begin_0 = const()[name = string("xx_1_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_1_end_0 = const()[name = string("xx_1_end_0"), val = tensor<int32, [3]>([1, 1536, 150])];
            tensor<bool, [3]> xx_1_end_mask_0 = const()[name = string("xx_1_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 150]> xx_1_cast_fp16 = slice_by_index(begin = xx_1_begin_0, end = xx_1_end_0, end_mask = xx_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("xx_1_cast_fp16")];
            tensor<fp16, [1, 512, 150]> new_1_cast_fp16 = mul(x = b_1_cast_fp16, y = xx_1_cast_fp16)[name = string("new_1_cast_fp16")];
            tensor<int32, [4]> input_5_begin_0 = const()[name = string("input_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> input_5_end_0 = const()[name = string("input_5_end_0"), val = tensor<int32, [4]>([1, 1, 512, 6])];
            tensor<bool, [4]> input_5_end_mask_0 = const()[name = string("input_5_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_5_squeeze_mask_0 = const()[name = string("input_5_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_5_cast_fp16 = slice_by_index(begin = input_5_begin_0, end = input_5_end_0, end_mask = input_5_end_mask_0, squeeze_mask = input_5_squeeze_mask_0, x = conv)[name = string("input_5_cast_fp16")];
            tensor<int32, [6]> var_425_pad_0 = const()[name = string("op_425_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_425_mode_0 = const()[name = string("op_425_mode_0"), val = string("constant")];
            fp16 const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 156]> var_425_cast_fp16 = pad(constant_val = const_5_to_fp16, mode = var_425_mode_0, pad = var_425_pad_0, x = new_1_cast_fp16)[name = string("op_425_cast_fp16")];
            tensor<int32, [6]> var_427_pad_0 = const()[name = string("op_427_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 150])];
            string var_427_mode_0 = const()[name = string("op_427_mode_0"), val = string("constant")];
            fp16 const_6_to_fp16 = const()[name = string("const_6_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 156]> var_427_cast_fp16 = pad(constant_val = const_6_to_fp16, mode = var_427_mode_0, pad = var_427_pad_0, x = input_5_cast_fp16)[name = string("op_427_cast_fp16")];
            tensor<fp16, [1, 512, 156]> win_1_cast_fp16 = add(x = var_425_cast_fp16, y = var_427_cast_fp16)[name = string("win_1_cast_fp16")];
            tensor<int32, [3]> var_429_begin_0 = const()[name = string("op_429_begin_0"), val = tensor<int32, [3]>([0, 0, 150])];
            tensor<int32, [3]> var_429_end_0 = const()[name = string("op_429_end_0"), val = tensor<int32, [3]>([1, 512, 156])];
            tensor<bool, [3]> var_429_end_mask_0 = const()[name = string("op_429_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_429_cast_fp16 = slice_by_index(begin = var_429_begin_0, end = var_429_end_0, end_mask = var_429_end_mask_0, x = win_1_cast_fp16)[name = string("op_429_cast_fp16")];
            string var_435_pad_type_0 = const()[name = string("op_435_pad_type_0"), val = string("valid")];
            int32 var_435_groups_0 = const()[name = string("op_435_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_435_strides_0 = const()[name = string("op_435_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_435_pad_0 = const()[name = string("op_435_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_435_dilations_0 = const()[name = string("op_435_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_0_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1627520)))];
            tensor<fp16, [1, 512, 150]> var_435_cast_fp16 = conv(dilations = var_435_dilations_0, groups = var_435_groups_0, pad = var_435_pad_0, pad_type = var_435_pad_type_0, strides = var_435_strides_0, weight = core_backbone_layers_0_conv_conv_weight_to_fp16, x = win_1_cast_fp16)[name = string("op_435_cast_fp16")];
            tensor<fp16, [1, 512, 150]> var_437_cast_fp16 = mul(x = c_1_cast_fp16, y = var_435_cast_fp16)[name = string("op_437_cast_fp16")];
            tensor<int32, [3]> input_7_perm_0 = const()[name = string("input_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_0_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1634752)))];
            tensor<fp16, [512]> linear_2_bias_0_to_fp16 = const()[name = string("linear_2_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2159104)))];
            tensor<fp16, [1, 150, 512]> input_7_cast_fp16 = transpose(perm = input_7_perm_0, x = var_437_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [1, 150, 512]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_0_conv_out_proj_weight_to_fp16, x = input_7_cast_fp16)[name = string("linear_2_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_442_cast_fp16 = abs(x = linear_2_cast_fp16)[name = string("op_442_cast_fp16")];
            tensor<int32, [1]> var_444_axes_0 = const()[name = string("op_444_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_444_cast_fp16 = reduce_max(axes = var_444_axes_0, keep_dims = var_33, x = var_442_cast_fp16)[name = string("op_444_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_3_cast_fp16 = maximum(x = var_444_cast_fp16, y = var_34_to_fp16)[name = string("s_3_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_5_cast_fp16 = real_div(x = linear_2_cast_fp16, y = s_3_cast_fp16)[name = string("y_5_cast_fp16")];
            fp16 var_35_promoted_1_to_fp16 = const()[name = string("op_35_promoted_1_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_447_cast_fp16 = pow(x = y_5_cast_fp16, y = var_35_promoted_1_to_fp16)[name = string("op_447_cast_fp16")];
            tensor<int32, [1]> var_449_axes_0 = const()[name = string("op_449_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_449_keep_dims_0 = const()[name = string("op_449_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_449_cast_fp16 = reduce_mean(axes = var_449_axes_0, keep_dims = var_449_keep_dims_0, x = var_447_cast_fp16)[name = string("op_449_cast_fp16")];
            fp32 var_450_epsilon_0 = const()[name = string("op_450_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_450_cast_fp16 = inverse(epsilon = var_450_epsilon_0, x = s_3_cast_fp16)[name = string("op_450_cast_fp16")];
            fp16 var_451_to_fp16 = const()[name = string("op_451_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_452_cast_fp16 = mul(x = var_450_cast_fp16, y = var_451_to_fp16)[name = string("op_452_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_453_cast_fp16 = real_div(x = var_452_cast_fp16, y = s_3_cast_fp16)[name = string("op_453_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_454_cast_fp16 = add(x = var_449_cast_fp16, y = var_453_cast_fp16)[name = string("op_454_cast_fp16")];
            fp32 var_455_epsilon_0 = const()[name = string("op_455_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_455_cast_fp16 = rsqrt(epsilon = var_455_epsilon_0, x = var_454_cast_fp16)[name = string("op_455_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_456_cast_fp16 = mul(x = y_5_cast_fp16, y = var_455_cast_fp16)[name = string("op_456_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_0_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2160192)))];
            tensor<fp16, [1, 150, 512]> a_1_cast_fp16 = mul(x = core_backbone_layers_0_post_operator_norm_weight_to_fp16, y = var_456_cast_fp16)[name = string("a_1_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_5_cast_fp16 = add(x = x_1_cast_fp16, y = a_1_cast_fp16)[name = string("x_5_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_460_cast_fp16 = abs(x = x_5_cast_fp16)[name = string("op_460_cast_fp16")];
            tensor<int32, [1]> var_462_axes_0 = const()[name = string("op_462_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_462_cast_fp16 = reduce_max(axes = var_462_axes_0, keep_dims = var_33, x = var_460_cast_fp16)[name = string("op_462_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_5_cast_fp16 = maximum(x = var_462_cast_fp16, y = var_34_to_fp16)[name = string("s_5_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_7_cast_fp16 = real_div(x = x_5_cast_fp16, y = s_5_cast_fp16)[name = string("y_7_cast_fp16")];
            fp16 var_35_promoted_2_to_fp16 = const()[name = string("op_35_promoted_2_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_465_cast_fp16 = pow(x = y_7_cast_fp16, y = var_35_promoted_2_to_fp16)[name = string("op_465_cast_fp16")];
            tensor<int32, [1]> var_467_axes_0 = const()[name = string("op_467_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_467_keep_dims_0 = const()[name = string("op_467_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_467_cast_fp16 = reduce_mean(axes = var_467_axes_0, keep_dims = var_467_keep_dims_0, x = var_465_cast_fp16)[name = string("op_467_cast_fp16")];
            fp32 var_468_epsilon_0 = const()[name = string("op_468_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_468_cast_fp16 = inverse(epsilon = var_468_epsilon_0, x = s_5_cast_fp16)[name = string("op_468_cast_fp16")];
            fp16 var_469_to_fp16 = const()[name = string("op_469_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_470_cast_fp16 = mul(x = var_468_cast_fp16, y = var_469_to_fp16)[name = string("op_470_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_471_cast_fp16 = real_div(x = var_470_cast_fp16, y = s_5_cast_fp16)[name = string("op_471_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_472_cast_fp16 = add(x = var_467_cast_fp16, y = var_471_cast_fp16)[name = string("op_472_cast_fp16")];
            fp32 var_473_epsilon_0 = const()[name = string("op_473_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_473_cast_fp16 = rsqrt(epsilon = var_473_epsilon_0, x = var_472_cast_fp16)[name = string("op_473_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_474_cast_fp16 = mul(x = y_7_cast_fp16, y = var_473_cast_fp16)[name = string("op_474_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_0_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2161280)))];
            tensor<fp16, [1, 150, 512]> input_9_cast_fp16 = mul(x = core_backbone_layers_0_ffn_norm_weight_to_fp16, y = var_474_cast_fp16)[name = string("input_9_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_0_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2162368)))];
            tensor<fp16, [2304]> linear_3_bias_0_to_fp16 = const()[name = string("linear_3_bias_0_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4521728)))];
            tensor<fp16, [1, 150, 2304]> linear_3_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w1_weight_to_fp16, x = input_9_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_0_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4526400)))];
            tensor<fp16, [1, 150, 2304]> linear_4_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w3_weight_to_fp16, x = input_9_cast_fp16)[name = string("linear_4_cast_fp16")];
            fp16 var_483_to_fp16 = const()[name = string("op_483_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 150, 2304]> var_484_cast_fp16 = mul(x = linear_3_cast_fp16, y = var_483_to_fp16)[name = string("op_484_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_487_cast_fp16 = tanh(x = var_484_cast_fp16)[name = string("op_487_cast_fp16")];
            fp16 var_488_to_fp16 = const()[name = string("op_488_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 150, 2304]> var_489_cast_fp16 = add(x = var_487_cast_fp16, y = var_488_to_fp16)[name = string("op_489_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_490_cast_fp16 = mul(x = var_484_cast_fp16, y = var_489_cast_fp16)[name = string("op_490_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> input_11_cast_fp16 = mul(x = var_490_cast_fp16, y = linear_4_cast_fp16)[name = string("input_11_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_0_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6885760)))];
            tensor<fp16, [1, 150, 512]> linear_5_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w2_weight_to_fp16, x = input_11_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_495_cast_fp16 = abs(x = linear_5_cast_fp16)[name = string("op_495_cast_fp16")];
            tensor<int32, [1]> var_497_axes_0 = const()[name = string("op_497_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_497_cast_fp16 = reduce_max(axes = var_497_axes_0, keep_dims = var_33, x = var_495_cast_fp16)[name = string("op_497_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_7_cast_fp16 = maximum(x = var_497_cast_fp16, y = var_34_to_fp16)[name = string("s_7_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_9_cast_fp16 = real_div(x = linear_5_cast_fp16, y = s_7_cast_fp16)[name = string("y_9_cast_fp16")];
            fp16 var_35_promoted_3_to_fp16 = const()[name = string("op_35_promoted_3_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_500_cast_fp16 = pow(x = y_9_cast_fp16, y = var_35_promoted_3_to_fp16)[name = string("op_500_cast_fp16")];
            tensor<int32, [1]> var_502_axes_0 = const()[name = string("op_502_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_502_keep_dims_0 = const()[name = string("op_502_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_502_cast_fp16 = reduce_mean(axes = var_502_axes_0, keep_dims = var_502_keep_dims_0, x = var_500_cast_fp16)[name = string("op_502_cast_fp16")];
            fp32 var_503_epsilon_0 = const()[name = string("op_503_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_503_cast_fp16 = inverse(epsilon = var_503_epsilon_0, x = s_7_cast_fp16)[name = string("op_503_cast_fp16")];
            fp16 var_504_to_fp16 = const()[name = string("op_504_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_505_cast_fp16 = mul(x = var_503_cast_fp16, y = var_504_to_fp16)[name = string("op_505_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_506_cast_fp16 = real_div(x = var_505_cast_fp16, y = s_7_cast_fp16)[name = string("op_506_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_507_cast_fp16 = add(x = var_502_cast_fp16, y = var_506_cast_fp16)[name = string("op_507_cast_fp16")];
            fp32 var_508_epsilon_0 = const()[name = string("op_508_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_508_cast_fp16 = rsqrt(epsilon = var_508_epsilon_0, x = var_507_cast_fp16)[name = string("op_508_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_509_cast_fp16 = mul(x = y_9_cast_fp16, y = var_508_cast_fp16)[name = string("op_509_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_0_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9245120)))];
            tensor<fp16, [1, 150, 512]> f_1_cast_fp16 = mul(x = core_backbone_layers_0_post_ffn_norm_weight_to_fp16, y = var_509_cast_fp16)[name = string("f_1_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_11_cast_fp16 = add(x = x_5_cast_fp16, y = f_1_cast_fp16)[name = string("x_11_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_513_cast_fp16 = abs(x = x_11_cast_fp16)[name = string("op_513_cast_fp16")];
            tensor<int32, [1]> var_515_axes_0 = const()[name = string("op_515_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_515_cast_fp16 = reduce_max(axes = var_515_axes_0, keep_dims = var_33, x = var_513_cast_fp16)[name = string("op_515_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_9_cast_fp16 = maximum(x = var_515_cast_fp16, y = var_34_to_fp16)[name = string("s_9_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_11_cast_fp16 = real_div(x = x_11_cast_fp16, y = s_9_cast_fp16)[name = string("y_11_cast_fp16")];
            fp16 var_35_promoted_4_to_fp16 = const()[name = string("op_35_promoted_4_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_518_cast_fp16 = pow(x = y_11_cast_fp16, y = var_35_promoted_4_to_fp16)[name = string("op_518_cast_fp16")];
            tensor<int32, [1]> var_520_axes_0 = const()[name = string("op_520_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_520_keep_dims_0 = const()[name = string("op_520_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_520_cast_fp16 = reduce_mean(axes = var_520_axes_0, keep_dims = var_520_keep_dims_0, x = var_518_cast_fp16)[name = string("op_520_cast_fp16")];
            fp32 var_521_epsilon_0 = const()[name = string("op_521_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_521_cast_fp16 = inverse(epsilon = var_521_epsilon_0, x = s_9_cast_fp16)[name = string("op_521_cast_fp16")];
            fp16 var_522_to_fp16 = const()[name = string("op_522_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_523_cast_fp16 = mul(x = var_521_cast_fp16, y = var_522_to_fp16)[name = string("op_523_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_524_cast_fp16 = real_div(x = var_523_cast_fp16, y = s_9_cast_fp16)[name = string("op_524_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_525_cast_fp16 = add(x = var_520_cast_fp16, y = var_524_cast_fp16)[name = string("op_525_cast_fp16")];
            fp32 var_526_epsilon_0 = const()[name = string("op_526_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_526_cast_fp16 = rsqrt(epsilon = var_526_epsilon_0, x = var_525_cast_fp16)[name = string("op_526_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_527_cast_fp16 = mul(x = y_11_cast_fp16, y = var_526_cast_fp16)[name = string("op_527_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_1_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9246208)))];
            tensor<fp16, [1, 150, 512]> input_13_cast_fp16 = mul(x = core_backbone_layers_1_operator_norm_weight_to_fp16, y = var_527_cast_fp16)[name = string("input_13_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_1_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9247296)))];
            tensor<fp16, [1, 150, 1536]> linear_6_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_1_conv_in_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("linear_6_cast_fp16")];
            tensor<int32, [3]> bcx_3_perm_0 = const()[name = string("bcx_3_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_5_begin_0 = const()[name = string("b_5_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_5_end_0 = const()[name = string("b_5_end_0"), val = tensor<int32, [3]>([1, 512, 150])];
            tensor<bool, [3]> b_5_end_mask_0 = const()[name = string("b_5_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 150]> bcx_3_cast_fp16 = transpose(perm = bcx_3_perm_0, x = linear_6_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [1, 512, 150]> b_5_cast_fp16 = slice_by_index(begin = b_5_begin_0, end = b_5_end_0, end_mask = b_5_end_mask_0, x = bcx_3_cast_fp16)[name = string("b_5_cast_fp16")];
            tensor<int32, [3]> c_3_begin_0 = const()[name = string("c_3_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_3_end_0 = const()[name = string("c_3_end_0"), val = tensor<int32, [3]>([1, 1024, 150])];
            tensor<bool, [3]> c_3_end_mask_0 = const()[name = string("c_3_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 150]> c_3_cast_fp16 = slice_by_index(begin = c_3_begin_0, end = c_3_end_0, end_mask = c_3_end_mask_0, x = bcx_3_cast_fp16)[name = string("c_3_cast_fp16")];
            tensor<int32, [3]> xx_3_begin_0 = const()[name = string("xx_3_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_3_end_0 = const()[name = string("xx_3_end_0"), val = tensor<int32, [3]>([1, 1536, 150])];
            tensor<bool, [3]> xx_3_end_mask_0 = const()[name = string("xx_3_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 150]> xx_3_cast_fp16 = slice_by_index(begin = xx_3_begin_0, end = xx_3_end_0, end_mask = xx_3_end_mask_0, x = bcx_3_cast_fp16)[name = string("xx_3_cast_fp16")];
            tensor<fp16, [1, 512, 150]> new_3_cast_fp16 = mul(x = b_5_cast_fp16, y = xx_3_cast_fp16)[name = string("new_3_cast_fp16")];
            tensor<int32, [4]> input_17_begin_0 = const()[name = string("input_17_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])];
            tensor<int32, [4]> input_17_end_0 = const()[name = string("input_17_end_0"), val = tensor<int32, [4]>([2, 1, 512, 6])];
            tensor<bool, [4]> input_17_end_mask_0 = const()[name = string("input_17_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_17_squeeze_mask_0 = const()[name = string("input_17_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_17_cast_fp16 = slice_by_index(begin = input_17_begin_0, end = input_17_end_0, end_mask = input_17_end_mask_0, squeeze_mask = input_17_squeeze_mask_0, x = conv)[name = string("input_17_cast_fp16")];
            tensor<int32, [6]> var_542_pad_0 = const()[name = string("op_542_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_542_mode_0 = const()[name = string("op_542_mode_0"), val = string("constant")];
            fp16 const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 156]> var_542_cast_fp16 = pad(constant_val = const_7_to_fp16, mode = var_542_mode_0, pad = var_542_pad_0, x = new_3_cast_fp16)[name = string("op_542_cast_fp16")];
            tensor<int32, [6]> var_544_pad_0 = const()[name = string("op_544_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 150])];
            string var_544_mode_0 = const()[name = string("op_544_mode_0"), val = string("constant")];
            fp16 const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 156]> var_544_cast_fp16 = pad(constant_val = const_8_to_fp16, mode = var_544_mode_0, pad = var_544_pad_0, x = input_17_cast_fp16)[name = string("op_544_cast_fp16")];
            tensor<fp16, [1, 512, 156]> win_5_cast_fp16 = add(x = var_542_cast_fp16, y = var_544_cast_fp16)[name = string("win_5_cast_fp16")];
            tensor<int32, [3]> var_546_begin_0 = const()[name = string("op_546_begin_0"), val = tensor<int32, [3]>([0, 0, 150])];
            tensor<int32, [3]> var_546_end_0 = const()[name = string("op_546_end_0"), val = tensor<int32, [3]>([1, 512, 156])];
            tensor<bool, [3]> var_546_end_mask_0 = const()[name = string("op_546_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_546_cast_fp16 = slice_by_index(begin = var_546_begin_0, end = var_546_end_0, end_mask = var_546_end_mask_0, x = win_5_cast_fp16)[name = string("op_546_cast_fp16")];
            string var_552_pad_type_0 = const()[name = string("op_552_pad_type_0"), val = string("valid")];
            int32 var_552_groups_0 = const()[name = string("op_552_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_552_strides_0 = const()[name = string("op_552_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_552_pad_0 = const()[name = string("op_552_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_552_dilations_0 = const()[name = string("op_552_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_1_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10820224)))];
            tensor<fp16, [1, 512, 150]> var_552_cast_fp16 = conv(dilations = var_552_dilations_0, groups = var_552_groups_0, pad = var_552_pad_0, pad_type = var_552_pad_type_0, strides = var_552_strides_0, weight = core_backbone_layers_1_conv_conv_weight_to_fp16, x = win_5_cast_fp16)[name = string("op_552_cast_fp16")];
            tensor<fp16, [1, 512, 150]> var_554_cast_fp16 = mul(x = c_3_cast_fp16, y = var_552_cast_fp16)[name = string("op_554_cast_fp16")];
            tensor<int32, [3]> input_19_perm_0 = const()[name = string("input_19_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_1_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10827456)))];
            tensor<fp16, [1, 150, 512]> input_19_cast_fp16 = transpose(perm = input_19_perm_0, x = var_554_cast_fp16)[name = string("transpose_24")];
            tensor<fp16, [1, 150, 512]> linear_7_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_1_conv_out_proj_weight_to_fp16, x = input_19_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_559_cast_fp16 = abs(x = linear_7_cast_fp16)[name = string("op_559_cast_fp16")];
            tensor<int32, [1]> var_561_axes_0 = const()[name = string("op_561_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_561_cast_fp16 = reduce_max(axes = var_561_axes_0, keep_dims = var_33, x = var_559_cast_fp16)[name = string("op_561_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_11_cast_fp16 = maximum(x = var_561_cast_fp16, y = var_34_to_fp16)[name = string("s_11_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_15_cast_fp16 = real_div(x = linear_7_cast_fp16, y = s_11_cast_fp16)[name = string("y_15_cast_fp16")];
            fp16 var_35_promoted_5_to_fp16 = const()[name = string("op_35_promoted_5_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_564_cast_fp16 = pow(x = y_15_cast_fp16, y = var_35_promoted_5_to_fp16)[name = string("op_564_cast_fp16")];
            tensor<int32, [1]> var_566_axes_0 = const()[name = string("op_566_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_566_keep_dims_0 = const()[name = string("op_566_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_566_cast_fp16 = reduce_mean(axes = var_566_axes_0, keep_dims = var_566_keep_dims_0, x = var_564_cast_fp16)[name = string("op_566_cast_fp16")];
            fp32 var_567_epsilon_0 = const()[name = string("op_567_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_567_cast_fp16 = inverse(epsilon = var_567_epsilon_0, x = s_11_cast_fp16)[name = string("op_567_cast_fp16")];
            fp16 var_568_to_fp16 = const()[name = string("op_568_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_569_cast_fp16 = mul(x = var_567_cast_fp16, y = var_568_to_fp16)[name = string("op_569_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_570_cast_fp16 = real_div(x = var_569_cast_fp16, y = s_11_cast_fp16)[name = string("op_570_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_571_cast_fp16 = add(x = var_566_cast_fp16, y = var_570_cast_fp16)[name = string("op_571_cast_fp16")];
            fp32 var_572_epsilon_0 = const()[name = string("op_572_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_572_cast_fp16 = rsqrt(epsilon = var_572_epsilon_0, x = var_571_cast_fp16)[name = string("op_572_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_573_cast_fp16 = mul(x = y_15_cast_fp16, y = var_572_cast_fp16)[name = string("op_573_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_1_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11351808)))];
            tensor<fp16, [1, 150, 512]> a_3_cast_fp16 = mul(x = core_backbone_layers_1_post_operator_norm_weight_to_fp16, y = var_573_cast_fp16)[name = string("a_3_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_15_cast_fp16 = add(x = x_11_cast_fp16, y = a_3_cast_fp16)[name = string("x_15_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_577_cast_fp16 = abs(x = x_15_cast_fp16)[name = string("op_577_cast_fp16")];
            tensor<int32, [1]> var_579_axes_0 = const()[name = string("op_579_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_579_cast_fp16 = reduce_max(axes = var_579_axes_0, keep_dims = var_33, x = var_577_cast_fp16)[name = string("op_579_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_13_cast_fp16 = maximum(x = var_579_cast_fp16, y = var_34_to_fp16)[name = string("s_13_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_17_cast_fp16 = real_div(x = x_15_cast_fp16, y = s_13_cast_fp16)[name = string("y_17_cast_fp16")];
            fp16 var_35_promoted_6_to_fp16 = const()[name = string("op_35_promoted_6_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_582_cast_fp16 = pow(x = y_17_cast_fp16, y = var_35_promoted_6_to_fp16)[name = string("op_582_cast_fp16")];
            tensor<int32, [1]> var_584_axes_0 = const()[name = string("op_584_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_584_keep_dims_0 = const()[name = string("op_584_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_584_cast_fp16 = reduce_mean(axes = var_584_axes_0, keep_dims = var_584_keep_dims_0, x = var_582_cast_fp16)[name = string("op_584_cast_fp16")];
            fp32 var_585_epsilon_0 = const()[name = string("op_585_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_585_cast_fp16 = inverse(epsilon = var_585_epsilon_0, x = s_13_cast_fp16)[name = string("op_585_cast_fp16")];
            fp16 var_586_to_fp16 = const()[name = string("op_586_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_587_cast_fp16 = mul(x = var_585_cast_fp16, y = var_586_to_fp16)[name = string("op_587_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_588_cast_fp16 = real_div(x = var_587_cast_fp16, y = s_13_cast_fp16)[name = string("op_588_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_589_cast_fp16 = add(x = var_584_cast_fp16, y = var_588_cast_fp16)[name = string("op_589_cast_fp16")];
            fp32 var_590_epsilon_0 = const()[name = string("op_590_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_590_cast_fp16 = rsqrt(epsilon = var_590_epsilon_0, x = var_589_cast_fp16)[name = string("op_590_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_591_cast_fp16 = mul(x = y_17_cast_fp16, y = var_590_cast_fp16)[name = string("op_591_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_1_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11352896)))];
            tensor<fp16, [1, 150, 512]> input_21_cast_fp16 = mul(x = core_backbone_layers_1_ffn_norm_weight_to_fp16, y = var_591_cast_fp16)[name = string("input_21_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_1_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11353984)))];
            tensor<fp16, [1, 150, 2304]> linear_8_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w1_weight_to_fp16, x = input_21_cast_fp16)[name = string("linear_8_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_1_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13713344)))];
            tensor<fp16, [1, 150, 2304]> linear_9_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w3_weight_to_fp16, x = input_21_cast_fp16)[name = string("linear_9_cast_fp16")];
            fp16 var_600_to_fp16 = const()[name = string("op_600_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 150, 2304]> var_601_cast_fp16 = mul(x = linear_8_cast_fp16, y = var_600_to_fp16)[name = string("op_601_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_604_cast_fp16 = tanh(x = var_601_cast_fp16)[name = string("op_604_cast_fp16")];
            fp16 var_605_to_fp16 = const()[name = string("op_605_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 150, 2304]> var_606_cast_fp16 = add(x = var_604_cast_fp16, y = var_605_to_fp16)[name = string("op_606_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_607_cast_fp16 = mul(x = var_601_cast_fp16, y = var_606_cast_fp16)[name = string("op_607_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> input_23_cast_fp16 = mul(x = var_607_cast_fp16, y = linear_9_cast_fp16)[name = string("input_23_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_1_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16072704)))];
            tensor<fp16, [1, 150, 512]> linear_10_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w2_weight_to_fp16, x = input_23_cast_fp16)[name = string("linear_10_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_612_cast_fp16 = abs(x = linear_10_cast_fp16)[name = string("op_612_cast_fp16")];
            tensor<int32, [1]> var_614_axes_0 = const()[name = string("op_614_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_614_cast_fp16 = reduce_max(axes = var_614_axes_0, keep_dims = var_33, x = var_612_cast_fp16)[name = string("op_614_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_15_cast_fp16 = maximum(x = var_614_cast_fp16, y = var_34_to_fp16)[name = string("s_15_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_19_cast_fp16 = real_div(x = linear_10_cast_fp16, y = s_15_cast_fp16)[name = string("y_19_cast_fp16")];
            fp16 var_35_promoted_7_to_fp16 = const()[name = string("op_35_promoted_7_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_617_cast_fp16 = pow(x = y_19_cast_fp16, y = var_35_promoted_7_to_fp16)[name = string("op_617_cast_fp16")];
            tensor<int32, [1]> var_619_axes_0 = const()[name = string("op_619_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_619_keep_dims_0 = const()[name = string("op_619_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_619_cast_fp16 = reduce_mean(axes = var_619_axes_0, keep_dims = var_619_keep_dims_0, x = var_617_cast_fp16)[name = string("op_619_cast_fp16")];
            fp32 var_620_epsilon_0 = const()[name = string("op_620_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_620_cast_fp16 = inverse(epsilon = var_620_epsilon_0, x = s_15_cast_fp16)[name = string("op_620_cast_fp16")];
            fp16 var_621_to_fp16 = const()[name = string("op_621_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_622_cast_fp16 = mul(x = var_620_cast_fp16, y = var_621_to_fp16)[name = string("op_622_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_623_cast_fp16 = real_div(x = var_622_cast_fp16, y = s_15_cast_fp16)[name = string("op_623_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_624_cast_fp16 = add(x = var_619_cast_fp16, y = var_623_cast_fp16)[name = string("op_624_cast_fp16")];
            fp32 var_625_epsilon_0 = const()[name = string("op_625_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_625_cast_fp16 = rsqrt(epsilon = var_625_epsilon_0, x = var_624_cast_fp16)[name = string("op_625_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_626_cast_fp16 = mul(x = y_19_cast_fp16, y = var_625_cast_fp16)[name = string("op_626_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_1_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18432064)))];
            tensor<fp16, [1, 150, 512]> f_3_cast_fp16 = mul(x = core_backbone_layers_1_post_ffn_norm_weight_to_fp16, y = var_626_cast_fp16)[name = string("f_3_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_21_cast_fp16 = add(x = x_15_cast_fp16, y = f_3_cast_fp16)[name = string("x_21_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_630_cast_fp16 = abs(x = x_21_cast_fp16)[name = string("op_630_cast_fp16")];
            tensor<int32, [1]> var_632_axes_0 = const()[name = string("op_632_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_632_cast_fp16 = reduce_max(axes = var_632_axes_0, keep_dims = var_33, x = var_630_cast_fp16)[name = string("op_632_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_17_cast_fp16 = maximum(x = var_632_cast_fp16, y = var_34_to_fp16)[name = string("s_17_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_21_cast_fp16 = real_div(x = x_21_cast_fp16, y = s_17_cast_fp16)[name = string("y_21_cast_fp16")];
            fp16 var_35_promoted_8_to_fp16 = const()[name = string("op_35_promoted_8_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_635_cast_fp16 = pow(x = y_21_cast_fp16, y = var_35_promoted_8_to_fp16)[name = string("op_635_cast_fp16")];
            tensor<int32, [1]> var_637_axes_0 = const()[name = string("op_637_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_637_keep_dims_0 = const()[name = string("op_637_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_637_cast_fp16 = reduce_mean(axes = var_637_axes_0, keep_dims = var_637_keep_dims_0, x = var_635_cast_fp16)[name = string("op_637_cast_fp16")];
            fp32 var_638_epsilon_0 = const()[name = string("op_638_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_638_cast_fp16 = inverse(epsilon = var_638_epsilon_0, x = s_17_cast_fp16)[name = string("op_638_cast_fp16")];
            fp16 var_639_to_fp16 = const()[name = string("op_639_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_640_cast_fp16 = mul(x = var_638_cast_fp16, y = var_639_to_fp16)[name = string("op_640_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_641_cast_fp16 = real_div(x = var_640_cast_fp16, y = s_17_cast_fp16)[name = string("op_641_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_642_cast_fp16 = add(x = var_637_cast_fp16, y = var_641_cast_fp16)[name = string("op_642_cast_fp16")];
            fp32 var_643_epsilon_0 = const()[name = string("op_643_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_643_cast_fp16 = rsqrt(epsilon = var_643_epsilon_0, x = var_642_cast_fp16)[name = string("op_643_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_644_cast_fp16 = mul(x = y_21_cast_fp16, y = var_643_cast_fp16)[name = string("op_644_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_2_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18433152)))];
            tensor<fp16, [1, 150, 512]> input_25_cast_fp16 = mul(x = core_backbone_layers_2_operator_norm_weight_to_fp16, y = var_644_cast_fp16)[name = string("input_25_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18434240)))];
            tensor<fp16, [1, 150, 512]> linear_11_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_q_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_11_cast_fp16")];
            tensor<int32, [4]> concat_0 = const()[name = string("concat_0"), val = tensor<int32, [4]>([1, 150, 8, 64])];
            tensor<fp16, [1, 150, 8, 64]> x_23_cast_fp16 = reshape(shape = concat_0, x = linear_11_cast_fp16)[name = string("x_23_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> var_651_cast_fp16 = abs(x = x_23_cast_fp16)[name = string("op_651_cast_fp16")];
            tensor<int32, [1]> var_653_axes_0 = const()[name = string("op_653_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 8, 1]> var_653_cast_fp16 = reduce_max(axes = var_653_axes_0, keep_dims = var_33, x = var_651_cast_fp16)[name = string("op_653_cast_fp16")];
            tensor<fp16, [1, 150, 8, 1]> s_19_cast_fp16 = maximum(x = var_653_cast_fp16, y = var_34_to_fp16)[name = string("s_19_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> y_23_cast_fp16 = real_div(x = x_23_cast_fp16, y = s_19_cast_fp16)[name = string("y_23_cast_fp16")];
            fp16 var_35_promoted_9_to_fp16 = const()[name = string("op_35_promoted_9_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 8, 64]> var_656_cast_fp16 = pow(x = y_23_cast_fp16, y = var_35_promoted_9_to_fp16)[name = string("op_656_cast_fp16")];
            tensor<int32, [1]> var_658_axes_0 = const()[name = string("op_658_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_658_keep_dims_0 = const()[name = string("op_658_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 8, 1]> var_658_cast_fp16 = reduce_mean(axes = var_658_axes_0, keep_dims = var_658_keep_dims_0, x = var_656_cast_fp16)[name = string("op_658_cast_fp16")];
            fp32 var_659_epsilon_0 = const()[name = string("op_659_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 8, 1]> var_659_cast_fp16 = inverse(epsilon = var_659_epsilon_0, x = s_19_cast_fp16)[name = string("op_659_cast_fp16")];
            fp16 var_660_to_fp16 = const()[name = string("op_660_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 8, 1]> var_661_cast_fp16 = mul(x = var_659_cast_fp16, y = var_660_to_fp16)[name = string("op_661_cast_fp16")];
            tensor<fp16, [1, 150, 8, 1]> var_662_cast_fp16 = real_div(x = var_661_cast_fp16, y = s_19_cast_fp16)[name = string("op_662_cast_fp16")];
            tensor<fp16, [1, 150, 8, 1]> var_663_cast_fp16 = add(x = var_658_cast_fp16, y = var_662_cast_fp16)[name = string("op_663_cast_fp16")];
            fp32 var_664_epsilon_0 = const()[name = string("op_664_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 8, 1]> var_664_cast_fp16 = rsqrt(epsilon = var_664_epsilon_0, x = var_663_cast_fp16)[name = string("op_664_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> var_665_cast_fp16 = mul(x = y_23_cast_fp16, y = var_664_cast_fp16)[name = string("op_665_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_2_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18958592)))];
            tensor<fp16, [1, 150, 8, 64]> var_666_cast_fp16 = mul(x = core_backbone_layers_2_self_attn_q_norm_weight_to_fp16, y = var_665_cast_fp16)[name = string("op_666_cast_fp16")];
            tensor<fp16, [128, 512]> core_backbone_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18958784)))];
            tensor<fp16, [128]> linear_12_bias_0_to_fp16 = const()[name = string("linear_12_bias_0_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19089920)))];
            tensor<fp16, [1, 150, 128]> linear_12_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_k_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_12_cast_fp16")];
            tensor<int32, [4]> concat_1 = const()[name = string("concat_1"), val = tensor<int32, [4]>([1, 150, 2, 64])];
            tensor<fp16, [1, 150, 2, 64]> x_25_cast_fp16 = reshape(shape = concat_1, x = linear_12_cast_fp16)[name = string("x_25_cast_fp16")];
            tensor<fp16, [1, 150, 2, 64]> var_673_cast_fp16 = abs(x = x_25_cast_fp16)[name = string("op_673_cast_fp16")];
            tensor<int32, [1]> var_675_axes_0 = const()[name = string("op_675_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 2, 1]> var_675_cast_fp16 = reduce_max(axes = var_675_axes_0, keep_dims = var_33, x = var_673_cast_fp16)[name = string("op_675_cast_fp16")];
            tensor<fp16, [1, 150, 2, 1]> s_21_cast_fp16 = maximum(x = var_675_cast_fp16, y = var_34_to_fp16)[name = string("s_21_cast_fp16")];
            tensor<fp16, [1, 150, 2, 64]> y_25_cast_fp16 = real_div(x = x_25_cast_fp16, y = s_21_cast_fp16)[name = string("y_25_cast_fp16")];
            fp16 var_35_promoted_10_to_fp16 = const()[name = string("op_35_promoted_10_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 2, 64]> var_678_cast_fp16 = pow(x = y_25_cast_fp16, y = var_35_promoted_10_to_fp16)[name = string("op_678_cast_fp16")];
            tensor<int32, [1]> var_680_axes_0 = const()[name = string("op_680_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_680_keep_dims_0 = const()[name = string("op_680_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 2, 1]> var_680_cast_fp16 = reduce_mean(axes = var_680_axes_0, keep_dims = var_680_keep_dims_0, x = var_678_cast_fp16)[name = string("op_680_cast_fp16")];
            fp32 var_681_epsilon_0 = const()[name = string("op_681_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 2, 1]> var_681_cast_fp16 = inverse(epsilon = var_681_epsilon_0, x = s_21_cast_fp16)[name = string("op_681_cast_fp16")];
            fp16 var_682_to_fp16 = const()[name = string("op_682_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 2, 1]> var_683_cast_fp16 = mul(x = var_681_cast_fp16, y = var_682_to_fp16)[name = string("op_683_cast_fp16")];
            tensor<fp16, [1, 150, 2, 1]> var_684_cast_fp16 = real_div(x = var_683_cast_fp16, y = s_21_cast_fp16)[name = string("op_684_cast_fp16")];
            tensor<fp16, [1, 150, 2, 1]> var_685_cast_fp16 = add(x = var_680_cast_fp16, y = var_684_cast_fp16)[name = string("op_685_cast_fp16")];
            fp32 var_686_epsilon_0 = const()[name = string("op_686_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 2, 1]> var_686_cast_fp16 = rsqrt(epsilon = var_686_epsilon_0, x = var_685_cast_fp16)[name = string("op_686_cast_fp16")];
            tensor<fp16, [1, 150, 2, 64]> var_687_cast_fp16 = mul(x = y_25_cast_fp16, y = var_686_cast_fp16)[name = string("op_687_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_2_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19090240)))];
            tensor<fp16, [1, 150, 2, 64]> var_688_cast_fp16 = mul(x = core_backbone_layers_2_self_attn_k_norm_weight_to_fp16, y = var_687_cast_fp16)[name = string("op_688_cast_fp16")];
            tensor<int32, [4]> new_5_perm_0 = const()[name = string("new_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [128, 512]> core_backbone_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19090432)))];
            tensor<fp16, [1, 150, 128]> linear_13_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_v_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_13_cast_fp16")];
            tensor<int32, [4]> concat_2 = const()[name = string("concat_2"), val = tensor<int32, [4]>([1, 150, 2, 64])];
            tensor<fp16, [1, 150, 2, 64]> var_693_cast_fp16 = reshape(shape = concat_2, x = linear_13_cast_fp16)[name = string("op_693_cast_fp16")];
            tensor<int32, [4]> new_7_perm_0 = const()[name = string("new_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_29_begin_0 = const()[name = string("input_29_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
            tensor<int32, [5]> input_29_end_0 = const()[name = string("input_29_end_0"), val = tensor<int32, [5]>([1, 1, 2, 29, 64])];
            tensor<bool, [5]> input_29_end_mask_0 = const()[name = string("input_29_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_29_squeeze_mask_0 = const()[name = string("input_29_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_29_cast_fp16 = slice_by_index(begin = input_29_begin_0, end = input_29_end_0, end_mask = input_29_end_mask_0, squeeze_mask = input_29_squeeze_mask_0, x = k)[name = string("input_29_cast_fp16")];
            tensor<int32, [8]> var_698_pad_0 = const()[name = string("op_698_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_698_mode_0 = const()[name = string("op_698_mode_0"), val = string("constant")];
            fp16 const_9_to_fp16 = const()[name = string("const_9_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 150, 64]> new_5_cast_fp16 = transpose(perm = new_5_perm_0, x = var_688_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [1, 2, 179, 64]> var_698_cast_fp16 = pad(constant_val = const_9_to_fp16, mode = var_698_mode_0, pad = var_698_pad_0, x = new_5_cast_fp16)[name = string("op_698_cast_fp16")];
            tensor<int32, [8]> var_700_pad_0 = const()[name = string("op_700_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 150, 0, 0])];
            string var_700_mode_0 = const()[name = string("op_700_mode_0"), val = string("constant")];
            fp16 const_10_to_fp16 = const()[name = string("const_10_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 179, 64]> var_700_cast_fp16 = pad(constant_val = const_10_to_fp16, mode = var_700_mode_0, pad = var_700_pad_0, x = input_29_cast_fp16)[name = string("op_700_cast_fp16")];
            tensor<fp16, [1, 2, 179, 64]> x_29_cast_fp16 = add(x = var_698_cast_fp16, y = var_700_cast_fp16)[name = string("x_29_cast_fp16")];
            tensor<int32, [5]> input_33_begin_0 = const()[name = string("input_33_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
            tensor<int32, [5]> input_33_end_0 = const()[name = string("input_33_end_0"), val = tensor<int32, [5]>([1, 1, 2, 29, 64])];
            tensor<bool, [5]> input_33_end_mask_0 = const()[name = string("input_33_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_33_squeeze_mask_0 = const()[name = string("input_33_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_33_cast_fp16 = slice_by_index(begin = input_33_begin_0, end = input_33_end_0, end_mask = input_33_end_mask_0, squeeze_mask = input_33_squeeze_mask_0, x = v)[name = string("input_33_cast_fp16")];
            tensor<int32, [8]> var_705_pad_0 = const()[name = string("op_705_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_705_mode_0 = const()[name = string("op_705_mode_0"), val = string("constant")];
            fp16 const_11_to_fp16 = const()[name = string("const_11_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 150, 64]> new_7_cast_fp16 = transpose(perm = new_7_perm_0, x = var_693_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [1, 2, 179, 64]> var_705_cast_fp16 = pad(constant_val = const_11_to_fp16, mode = var_705_mode_0, pad = var_705_pad_0, x = new_7_cast_fp16)[name = string("op_705_cast_fp16")];
            tensor<int32, [8]> var_707_pad_0 = const()[name = string("op_707_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 150, 0, 0])];
            string var_707_mode_0 = const()[name = string("op_707_mode_0"), val = string("constant")];
            fp16 const_12_to_fp16 = const()[name = string("const_12_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 179, 64]> var_707_cast_fp16 = pad(constant_val = const_12_to_fp16, mode = var_707_mode_0, pad = var_707_pad_0, x = input_33_cast_fp16)[name = string("op_707_cast_fp16")];
            tensor<fp16, [1, 2, 179, 64]> v_3_cast_fp16 = add(x = var_705_cast_fp16, y = var_707_cast_fp16)[name = string("v_3_cast_fp16")];
            tensor<fp16, [1, 150, 1, 64]> const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = tensor<fp16, [1, 150, 1, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71240128)))];
            tensor<fp16, [1, 150, 8, 64]> var_709_cast_fp16 = mul(x = var_666_cast_fp16, y = const_33_to_fp16)[name = string("op_709_cast_fp16")];
            tensor<int32, [2]> var_710_split_sizes_0 = const()[name = string("op_710_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 const_34 = const()[name = string("const_34"), val = int32(3)];
            tensor<fp16, [1, 150, 8, 32]> var_710_cast_fp16_0, tensor<fp16, [1, 150, 8, 32]> var_710_cast_fp16_1 = split(axis = const_34, split_sizes = var_710_split_sizes_0, x = var_666_cast_fp16)[name = string("op_710_cast_fp16")];
            fp16 const_13_promoted_to_fp16 = const()[name = string("const_13_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 150, 8, 32]> var_712_cast_fp16 = mul(x = var_710_cast_fp16_1, y = const_13_promoted_to_fp16)[name = string("op_712_cast_fp16")];
            bool var_714_interleave_0 = const()[name = string("op_714_interleave_0"), val = bool(false)];
            int32 const_35 = const()[name = string("const_35"), val = int32(3)];
            tensor<fp16, [1, 150, 8, 64]> var_714_cast_fp16 = concat(axis = const_35, interleave = var_714_interleave_0, values = (var_712_cast_fp16, var_710_cast_fp16_0))[name = string("op_714_cast_fp16")];
            tensor<fp16, [1, 150, 1, 64]> const_36_to_fp16 = const()[name = string("const_36_to_fp16"), val = tensor<fp16, [1, 150, 1, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71259392)))];
            tensor<fp16, [1, 150, 8, 64]> var_715_cast_fp16 = mul(x = var_714_cast_fp16, y = const_36_to_fp16)[name = string("op_715_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> q_1_cast_fp16 = add(x = var_709_cast_fp16, y = var_715_cast_fp16)[name = string("q_1_cast_fp16")];
            tensor<fp16, [1, 1, 179, 64]> const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = tensor<fp16, [1, 1, 179, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71278656)))];
            tensor<fp16, [1, 2, 179, 64]> var_717_cast_fp16 = mul(x = x_29_cast_fp16, y = const_1_to_fp16)[name = string("op_717_cast_fp16")];
            tensor<int32, [2]> var_718_split_sizes_0 = const()[name = string("op_718_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 var_718_axis_0 = const()[name = string("op_718_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 2, 179, 32]> var_718_cast_fp16_0, tensor<fp16, [1, 2, 179, 32]> var_718_cast_fp16_1 = split(axis = var_718_axis_0, split_sizes = var_718_split_sizes_0, x = x_29_cast_fp16)[name = string("op_718_cast_fp16")];
            fp16 const_14_promoted_to_fp16 = const()[name = string("const_14_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 2, 179, 32]> var_720_cast_fp16 = mul(x = var_718_cast_fp16_1, y = const_14_promoted_to_fp16)[name = string("op_720_cast_fp16")];
            bool var_722_interleave_0 = const()[name = string("op_722_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 2, 179, 64]> var_722_cast_fp16 = concat(axis = var_32, interleave = var_722_interleave_0, values = (var_720_cast_fp16, var_718_cast_fp16_0))[name = string("op_722_cast_fp16")];
            tensor<fp16, [1, 1, 179, 64]> const_2_to_fp16 = const()[name = string("const_2_to_fp16"), val = tensor<fp16, [1, 1, 179, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71301632)))];
            tensor<fp16, [1, 2, 179, 64]> var_723_cast_fp16 = mul(x = var_722_cast_fp16, y = const_2_to_fp16)[name = string("op_723_cast_fp16")];
            tensor<fp16, [1, 2, 179, 64]> k_3_cast_fp16 = add(x = var_717_cast_fp16, y = var_723_cast_fp16)[name = string("k_3_cast_fp16")];
            tensor<int32, [1]> var_725_axes_0 = const()[name = string("op_725_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 179, 64]> var_725_cast_fp16 = expand_dims(axes = var_725_axes_0, x = k_3_cast_fp16)[name = string("op_725_cast_fp16")];
            tensor<int32, [5]> var_727_reps_0 = const()[name = string("op_727_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 179, 64]> var_727_cast_fp16 = tile(reps = var_727_reps_0, x = var_725_cast_fp16)[name = string("op_727_cast_fp16")];
            tensor<int32, [4]> concat_4 = const()[name = string("concat_4"), val = tensor<int32, [4]>([1, 8, 179, 64])];
            tensor<fp16, [1, 8, 179, 64]> k_5_cast_fp16 = reshape(shape = concat_4, x = var_727_cast_fp16)[name = string("k_5_cast_fp16")];
            tensor<int32, [1]> var_729_axes_0 = const()[name = string("op_729_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 179, 64]> var_729_cast_fp16 = expand_dims(axes = var_729_axes_0, x = v_3_cast_fp16)[name = string("op_729_cast_fp16")];
            tensor<int32, [5]> var_731_reps_0 = const()[name = string("op_731_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 179, 64]> var_731_cast_fp16 = tile(reps = var_731_reps_0, x = var_729_cast_fp16)[name = string("op_731_cast_fp16")];
            tensor<int32, [4]> concat_6 = const()[name = string("concat_6"), val = tensor<int32, [4]>([1, 8, 179, 64])];
            tensor<fp16, [1, 8, 179, 64]> v_5_cast_fp16 = reshape(shape = concat_6, x = var_731_cast_fp16)[name = string("v_5_cast_fp16")];
            tensor<int32, [4]> transpose_0_perm_0 = const()[name = string("transpose_0_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_734_transpose_x_1 = const()[name = string("op_734_transpose_x_1"), val = bool(false)];
            bool var_734_transpose_y_1 = const()[name = string("op_734_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 150, 64]> transpose_0_cast_fp16 = transpose(perm = transpose_0_perm_0, x = q_1_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [1, 8, 150, 179]> var_734_cast_fp16 = matmul(transpose_x = var_734_transpose_x_1, transpose_y = var_734_transpose_y_1, x = transpose_0_cast_fp16, y = k_5_cast_fp16)[name = string("op_734_cast_fp16")];
            fp16 var_735_to_fp16 = const()[name = string("op_735_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 8, 150, 179]> s_23_cast_fp16 = mul(x = var_734_cast_fp16, y = var_735_to_fp16)[name = string("s_23_cast_fp16")];
            tensor<fp16, [1, 8, 150, 179]> x_31_cast_fp16 = add(x = s_23_cast_fp16, y = mask)[name = string("x_31_cast_fp16")];
            tensor<int32, [1]> var_739_axes_0 = const()[name = string("op_739_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 150, 1]> var_739_cast_fp16 = reduce_max(axes = var_739_axes_0, keep_dims = var_33, x = x_31_cast_fp16)[name = string("op_739_cast_fp16")];
            tensor<fp16, [1, 8, 150, 179]> x_33_cast_fp16 = sub(x = x_31_cast_fp16, y = var_739_cast_fp16)[name = string("x_33_cast_fp16")];
            tensor<fp16, [1, 8, 150, 179]> x_35_cast_fp16 = exp(x = x_33_cast_fp16)[name = string("x_35_cast_fp16")];
            tensor<int32, [1]> var_743_axes_0 = const()[name = string("op_743_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_743_keep_dims_0 = const()[name = string("op_743_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 150, 1]> var_743_cast_fp16 = reduce_sum(axes = var_743_axes_0, keep_dims = var_743_keep_dims_0, x = x_35_cast_fp16)[name = string("op_743_cast_fp16")];
            tensor<fp16, [1, 8, 150, 179]> var_744_cast_fp16 = real_div(x = x_35_cast_fp16, y = var_743_cast_fp16)[name = string("op_744_cast_fp16")];
            bool o_1_transpose_x_0 = const()[name = string("o_1_transpose_x_0"), val = bool(false)];
            bool o_1_transpose_y_0 = const()[name = string("o_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 150, 64]> o_1_cast_fp16 = matmul(transpose_x = o_1_transpose_x_0, transpose_y = o_1_transpose_y_0, x = var_744_cast_fp16, y = v_5_cast_fp16)[name = string("o_1_cast_fp16")];
            tensor<int32, [4]> o_3_perm_0 = const()[name = string("o_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [8, 512]> core_backbone_layers_2_self_attn_gate_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_gate_proj_weight_to_fp16"), val = tensor<fp16, [8, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19232320)))];
            tensor<fp16, [8]> linear_14_bias_0_to_fp16 = const()[name = string("linear_14_bias_0_to_fp16"), val = tensor<fp16, [8]>([0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0])];
            tensor<fp16, [1, 150, 8]> linear_14_cast_fp16 = linear(bias = linear_14_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_gate_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_14_cast_fp16")];
            fp16 var_750_to_fp16 = const()[name = string("op_750_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 150, 8]> var_751_cast_fp16 = mul(x = linear_14_cast_fp16, y = var_750_to_fp16)[name = string("op_751_cast_fp16")];
            tensor<fp16, [1, 150, 8]> var_752_cast_fp16 = tanh(x = var_751_cast_fp16)[name = string("op_752_cast_fp16")];
            fp16 var_753_to_fp16 = const()[name = string("op_753_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 150, 8]> var_754_cast_fp16 = add(x = var_752_cast_fp16, y = var_753_to_fp16)[name = string("op_754_cast_fp16")];
            fp16 var_755_to_fp16 = const()[name = string("op_755_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 150, 8]> var_756_cast_fp16 = mul(x = var_754_cast_fp16, y = var_755_to_fp16)[name = string("op_756_cast_fp16")];
            tensor<int32, [1]> var_757_axes_0 = const()[name = string("op_757_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 8, 1]> var_757_cast_fp16 = expand_dims(axes = var_757_axes_0, x = var_756_cast_fp16)[name = string("op_757_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> o_3_cast_fp16 = transpose(perm = o_3_perm_0, x = o_1_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [1, 150, 8, 64]> o_5_cast_fp16 = mul(x = o_3_cast_fp16, y = var_757_cast_fp16)[name = string("o_5_cast_fp16")];
            tensor<int32, [3]> concat_7 = const()[name = string("concat_7"), val = tensor<int32, [3]>([1, 150, 512])];
            tensor<fp16, [1, 150, 512]> input_35_cast_fp16 = reshape(shape = concat_7, x = o_5_cast_fp16)[name = string("input_35_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19240576)))];
            tensor<fp16, [1, 150, 512]> linear_15_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_o_proj_weight_to_fp16, x = input_35_cast_fp16)[name = string("linear_15_cast_fp16")];
            tensor<int32, [4]> var_764_begin_0 = const()[name = string("op_764_begin_0"), val = tensor<int32, [4]>([0, 0, 150, 0])];
            tensor<int32, [4]> var_764_end_0 = const()[name = string("op_764_end_0"), val = tensor<int32, [4]>([1, 2, 179, 64])];
            tensor<bool, [4]> var_764_end_mask_0 = const()[name = string("op_764_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_764_cast_fp16 = slice_by_index(begin = var_764_begin_0, end = var_764_end_0, end_mask = var_764_end_mask_0, x = x_29_cast_fp16)[name = string("op_764_cast_fp16")];
            tensor<int32, [4]> var_767_begin_0 = const()[name = string("op_767_begin_0"), val = tensor<int32, [4]>([0, 0, 150, 0])];
            tensor<int32, [4]> var_767_end_0 = const()[name = string("op_767_end_0"), val = tensor<int32, [4]>([1, 2, 179, 64])];
            tensor<bool, [4]> var_767_end_mask_0 = const()[name = string("op_767_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_767_cast_fp16 = slice_by_index(begin = var_767_begin_0, end = var_767_end_0, end_mask = var_767_end_mask_0, x = v_3_cast_fp16)[name = string("op_767_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_769_cast_fp16 = abs(x = linear_15_cast_fp16)[name = string("op_769_cast_fp16")];
            tensor<int32, [1]> var_771_axes_0 = const()[name = string("op_771_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_771_cast_fp16 = reduce_max(axes = var_771_axes_0, keep_dims = var_33, x = var_769_cast_fp16)[name = string("op_771_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_25_cast_fp16 = maximum(x = var_771_cast_fp16, y = var_34_to_fp16)[name = string("s_25_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_27_cast_fp16 = real_div(x = linear_15_cast_fp16, y = s_25_cast_fp16)[name = string("y_27_cast_fp16")];
            fp16 var_35_promoted_11_to_fp16 = const()[name = string("op_35_promoted_11_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_774_cast_fp16 = pow(x = y_27_cast_fp16, y = var_35_promoted_11_to_fp16)[name = string("op_774_cast_fp16")];
            tensor<int32, [1]> var_776_axes_0 = const()[name = string("op_776_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_776_keep_dims_0 = const()[name = string("op_776_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_776_cast_fp16 = reduce_mean(axes = var_776_axes_0, keep_dims = var_776_keep_dims_0, x = var_774_cast_fp16)[name = string("op_776_cast_fp16")];
            fp32 var_777_epsilon_0 = const()[name = string("op_777_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_777_cast_fp16 = inverse(epsilon = var_777_epsilon_0, x = s_25_cast_fp16)[name = string("op_777_cast_fp16")];
            fp16 var_778_to_fp16 = const()[name = string("op_778_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_779_cast_fp16 = mul(x = var_777_cast_fp16, y = var_778_to_fp16)[name = string("op_779_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_780_cast_fp16 = real_div(x = var_779_cast_fp16, y = s_25_cast_fp16)[name = string("op_780_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_781_cast_fp16 = add(x = var_776_cast_fp16, y = var_780_cast_fp16)[name = string("op_781_cast_fp16")];
            fp32 var_782_epsilon_0 = const()[name = string("op_782_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_782_cast_fp16 = rsqrt(epsilon = var_782_epsilon_0, x = var_781_cast_fp16)[name = string("op_782_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_783_cast_fp16 = mul(x = y_27_cast_fp16, y = var_782_cast_fp16)[name = string("op_783_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_2_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19764928)))];
            tensor<fp16, [1, 150, 512]> a_5_cast_fp16 = mul(x = core_backbone_layers_2_post_operator_norm_weight_to_fp16, y = var_783_cast_fp16)[name = string("a_5_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_41_cast_fp16 = add(x = x_21_cast_fp16, y = a_5_cast_fp16)[name = string("x_41_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_787_cast_fp16 = abs(x = x_41_cast_fp16)[name = string("op_787_cast_fp16")];
            tensor<int32, [1]> var_789_axes_0 = const()[name = string("op_789_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_789_cast_fp16 = reduce_max(axes = var_789_axes_0, keep_dims = var_33, x = var_787_cast_fp16)[name = string("op_789_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_27_cast_fp16 = maximum(x = var_789_cast_fp16, y = var_34_to_fp16)[name = string("s_27_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_29_cast_fp16 = real_div(x = x_41_cast_fp16, y = s_27_cast_fp16)[name = string("y_29_cast_fp16")];
            fp16 var_35_promoted_12_to_fp16 = const()[name = string("op_35_promoted_12_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_792_cast_fp16 = pow(x = y_29_cast_fp16, y = var_35_promoted_12_to_fp16)[name = string("op_792_cast_fp16")];
            tensor<int32, [1]> var_794_axes_0 = const()[name = string("op_794_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_794_keep_dims_0 = const()[name = string("op_794_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_794_cast_fp16 = reduce_mean(axes = var_794_axes_0, keep_dims = var_794_keep_dims_0, x = var_792_cast_fp16)[name = string("op_794_cast_fp16")];
            fp32 var_795_epsilon_0 = const()[name = string("op_795_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_795_cast_fp16 = inverse(epsilon = var_795_epsilon_0, x = s_27_cast_fp16)[name = string("op_795_cast_fp16")];
            fp16 var_796_to_fp16 = const()[name = string("op_796_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_797_cast_fp16 = mul(x = var_795_cast_fp16, y = var_796_to_fp16)[name = string("op_797_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_798_cast_fp16 = real_div(x = var_797_cast_fp16, y = s_27_cast_fp16)[name = string("op_798_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_799_cast_fp16 = add(x = var_794_cast_fp16, y = var_798_cast_fp16)[name = string("op_799_cast_fp16")];
            fp32 var_800_epsilon_0 = const()[name = string("op_800_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_800_cast_fp16 = rsqrt(epsilon = var_800_epsilon_0, x = var_799_cast_fp16)[name = string("op_800_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_801_cast_fp16 = mul(x = y_29_cast_fp16, y = var_800_cast_fp16)[name = string("op_801_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_2_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19766016)))];
            tensor<fp16, [1, 150, 512]> input_37_cast_fp16 = mul(x = core_backbone_layers_2_ffn_norm_weight_to_fp16, y = var_801_cast_fp16)[name = string("input_37_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_2_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19767104)))];
            tensor<fp16, [1, 150, 2304]> linear_16_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w1_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_16_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_2_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22126464)))];
            tensor<fp16, [1, 150, 2304]> linear_17_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w3_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_17_cast_fp16")];
            fp16 var_810_to_fp16 = const()[name = string("op_810_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 150, 2304]> var_811_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_810_to_fp16)[name = string("op_811_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_814_cast_fp16 = tanh(x = var_811_cast_fp16)[name = string("op_814_cast_fp16")];
            fp16 var_815_to_fp16 = const()[name = string("op_815_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 150, 2304]> var_816_cast_fp16 = add(x = var_814_cast_fp16, y = var_815_to_fp16)[name = string("op_816_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_817_cast_fp16 = mul(x = var_811_cast_fp16, y = var_816_cast_fp16)[name = string("op_817_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> input_39_cast_fp16 = mul(x = var_817_cast_fp16, y = linear_17_cast_fp16)[name = string("input_39_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_2_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24485824)))];
            tensor<fp16, [1, 150, 512]> linear_18_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w2_weight_to_fp16, x = input_39_cast_fp16)[name = string("linear_18_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_822_cast_fp16 = abs(x = linear_18_cast_fp16)[name = string("op_822_cast_fp16")];
            tensor<int32, [1]> var_824_axes_0 = const()[name = string("op_824_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_824_cast_fp16 = reduce_max(axes = var_824_axes_0, keep_dims = var_33, x = var_822_cast_fp16)[name = string("op_824_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_29_cast_fp16 = maximum(x = var_824_cast_fp16, y = var_34_to_fp16)[name = string("s_29_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_31_cast_fp16 = real_div(x = linear_18_cast_fp16, y = s_29_cast_fp16)[name = string("y_31_cast_fp16")];
            fp16 var_35_promoted_13_to_fp16 = const()[name = string("op_35_promoted_13_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_827_cast_fp16 = pow(x = y_31_cast_fp16, y = var_35_promoted_13_to_fp16)[name = string("op_827_cast_fp16")];
            tensor<int32, [1]> var_829_axes_0 = const()[name = string("op_829_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_829_keep_dims_0 = const()[name = string("op_829_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_829_cast_fp16 = reduce_mean(axes = var_829_axes_0, keep_dims = var_829_keep_dims_0, x = var_827_cast_fp16)[name = string("op_829_cast_fp16")];
            fp32 var_830_epsilon_0 = const()[name = string("op_830_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_830_cast_fp16 = inverse(epsilon = var_830_epsilon_0, x = s_29_cast_fp16)[name = string("op_830_cast_fp16")];
            fp16 var_831_to_fp16 = const()[name = string("op_831_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_832_cast_fp16 = mul(x = var_830_cast_fp16, y = var_831_to_fp16)[name = string("op_832_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_833_cast_fp16 = real_div(x = var_832_cast_fp16, y = s_29_cast_fp16)[name = string("op_833_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_834_cast_fp16 = add(x = var_829_cast_fp16, y = var_833_cast_fp16)[name = string("op_834_cast_fp16")];
            fp32 var_835_epsilon_0 = const()[name = string("op_835_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_835_cast_fp16 = rsqrt(epsilon = var_835_epsilon_0, x = var_834_cast_fp16)[name = string("op_835_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_836_cast_fp16 = mul(x = y_31_cast_fp16, y = var_835_cast_fp16)[name = string("op_836_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_2_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26845184)))];
            tensor<fp16, [1, 150, 512]> f_5_cast_fp16 = mul(x = core_backbone_layers_2_post_ffn_norm_weight_to_fp16, y = var_836_cast_fp16)[name = string("f_5_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_47_cast_fp16 = add(x = x_41_cast_fp16, y = f_5_cast_fp16)[name = string("x_47_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_840_cast_fp16 = abs(x = x_47_cast_fp16)[name = string("op_840_cast_fp16")];
            tensor<int32, [1]> var_842_axes_0 = const()[name = string("op_842_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_842_cast_fp16 = reduce_max(axes = var_842_axes_0, keep_dims = var_33, x = var_840_cast_fp16)[name = string("op_842_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_31_cast_fp16 = maximum(x = var_842_cast_fp16, y = var_34_to_fp16)[name = string("s_31_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_33_cast_fp16 = real_div(x = x_47_cast_fp16, y = s_31_cast_fp16)[name = string("y_33_cast_fp16")];
            fp16 var_35_promoted_14_to_fp16 = const()[name = string("op_35_promoted_14_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_845_cast_fp16 = pow(x = y_33_cast_fp16, y = var_35_promoted_14_to_fp16)[name = string("op_845_cast_fp16")];
            tensor<int32, [1]> var_847_axes_0 = const()[name = string("op_847_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_847_keep_dims_0 = const()[name = string("op_847_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_847_cast_fp16 = reduce_mean(axes = var_847_axes_0, keep_dims = var_847_keep_dims_0, x = var_845_cast_fp16)[name = string("op_847_cast_fp16")];
            fp32 var_848_epsilon_0 = const()[name = string("op_848_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_848_cast_fp16 = inverse(epsilon = var_848_epsilon_0, x = s_31_cast_fp16)[name = string("op_848_cast_fp16")];
            fp16 var_849_to_fp16 = const()[name = string("op_849_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_850_cast_fp16 = mul(x = var_848_cast_fp16, y = var_849_to_fp16)[name = string("op_850_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_851_cast_fp16 = real_div(x = var_850_cast_fp16, y = s_31_cast_fp16)[name = string("op_851_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_852_cast_fp16 = add(x = var_847_cast_fp16, y = var_851_cast_fp16)[name = string("op_852_cast_fp16")];
            fp32 var_853_epsilon_0 = const()[name = string("op_853_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_853_cast_fp16 = rsqrt(epsilon = var_853_epsilon_0, x = var_852_cast_fp16)[name = string("op_853_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_854_cast_fp16 = mul(x = y_33_cast_fp16, y = var_853_cast_fp16)[name = string("op_854_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_3_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26846272)))];
            tensor<fp16, [1, 150, 512]> input_41_cast_fp16 = mul(x = core_backbone_layers_3_operator_norm_weight_to_fp16, y = var_854_cast_fp16)[name = string("input_41_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_3_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26847360)))];
            tensor<fp16, [1, 150, 1536]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_3_conv_in_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("linear_19_cast_fp16")];
            tensor<int32, [3]> bcx_5_perm_0 = const()[name = string("bcx_5_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_11_begin_0 = const()[name = string("b_11_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_11_end_0 = const()[name = string("b_11_end_0"), val = tensor<int32, [3]>([1, 512, 150])];
            tensor<bool, [3]> b_11_end_mask_0 = const()[name = string("b_11_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 150]> bcx_5_cast_fp16 = transpose(perm = bcx_5_perm_0, x = linear_19_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [1, 512, 150]> b_11_cast_fp16 = slice_by_index(begin = b_11_begin_0, end = b_11_end_0, end_mask = b_11_end_mask_0, x = bcx_5_cast_fp16)[name = string("b_11_cast_fp16")];
            tensor<int32, [3]> c_5_begin_0 = const()[name = string("c_5_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_5_end_0 = const()[name = string("c_5_end_0"), val = tensor<int32, [3]>([1, 1024, 150])];
            tensor<bool, [3]> c_5_end_mask_0 = const()[name = string("c_5_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 150]> c_5_cast_fp16 = slice_by_index(begin = c_5_begin_0, end = c_5_end_0, end_mask = c_5_end_mask_0, x = bcx_5_cast_fp16)[name = string("c_5_cast_fp16")];
            tensor<int32, [3]> xx_5_begin_0 = const()[name = string("xx_5_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_5_end_0 = const()[name = string("xx_5_end_0"), val = tensor<int32, [3]>([1, 1536, 150])];
            tensor<bool, [3]> xx_5_end_mask_0 = const()[name = string("xx_5_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 150]> xx_5_cast_fp16 = slice_by_index(begin = xx_5_begin_0, end = xx_5_end_0, end_mask = xx_5_end_mask_0, x = bcx_5_cast_fp16)[name = string("xx_5_cast_fp16")];
            tensor<fp16, [1, 512, 150]> new_9_cast_fp16 = mul(x = b_11_cast_fp16, y = xx_5_cast_fp16)[name = string("new_9_cast_fp16")];
            tensor<int32, [4]> input_45_begin_0 = const()[name = string("input_45_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])];
            tensor<int32, [4]> input_45_end_0 = const()[name = string("input_45_end_0"), val = tensor<int32, [4]>([3, 1, 512, 6])];
            tensor<bool, [4]> input_45_end_mask_0 = const()[name = string("input_45_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_45_squeeze_mask_0 = const()[name = string("input_45_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_45_cast_fp16 = slice_by_index(begin = input_45_begin_0, end = input_45_end_0, end_mask = input_45_end_mask_0, squeeze_mask = input_45_squeeze_mask_0, x = conv)[name = string("input_45_cast_fp16")];
            tensor<int32, [6]> var_869_pad_0 = const()[name = string("op_869_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_869_mode_0 = const()[name = string("op_869_mode_0"), val = string("constant")];
            fp16 const_15_to_fp16 = const()[name = string("const_15_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 156]> var_869_cast_fp16 = pad(constant_val = const_15_to_fp16, mode = var_869_mode_0, pad = var_869_pad_0, x = new_9_cast_fp16)[name = string("op_869_cast_fp16")];
            tensor<int32, [6]> var_871_pad_0 = const()[name = string("op_871_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 150])];
            string var_871_mode_0 = const()[name = string("op_871_mode_0"), val = string("constant")];
            fp16 const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 156]> var_871_cast_fp16 = pad(constant_val = const_16_to_fp16, mode = var_871_mode_0, pad = var_871_pad_0, x = input_45_cast_fp16)[name = string("op_871_cast_fp16")];
            tensor<fp16, [1, 512, 156]> win_9_cast_fp16 = add(x = var_869_cast_fp16, y = var_871_cast_fp16)[name = string("win_9_cast_fp16")];
            tensor<int32, [3]> var_873_begin_0 = const()[name = string("op_873_begin_0"), val = tensor<int32, [3]>([0, 0, 150])];
            tensor<int32, [3]> var_873_end_0 = const()[name = string("op_873_end_0"), val = tensor<int32, [3]>([1, 512, 156])];
            tensor<bool, [3]> var_873_end_mask_0 = const()[name = string("op_873_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_873_cast_fp16 = slice_by_index(begin = var_873_begin_0, end = var_873_end_0, end_mask = var_873_end_mask_0, x = win_9_cast_fp16)[name = string("op_873_cast_fp16")];
            string var_879_pad_type_0 = const()[name = string("op_879_pad_type_0"), val = string("valid")];
            int32 var_879_groups_0 = const()[name = string("op_879_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_879_strides_0 = const()[name = string("op_879_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_879_pad_0 = const()[name = string("op_879_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_879_dilations_0 = const()[name = string("op_879_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_3_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28420288)))];
            tensor<fp16, [1, 512, 150]> var_879_cast_fp16 = conv(dilations = var_879_dilations_0, groups = var_879_groups_0, pad = var_879_pad_0, pad_type = var_879_pad_type_0, strides = var_879_strides_0, weight = core_backbone_layers_3_conv_conv_weight_to_fp16, x = win_9_cast_fp16)[name = string("op_879_cast_fp16")];
            tensor<fp16, [1, 512, 150]> var_881_cast_fp16 = mul(x = c_5_cast_fp16, y = var_879_cast_fp16)[name = string("op_881_cast_fp16")];
            tensor<int32, [3]> input_47_perm_0 = const()[name = string("input_47_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_3_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28427520)))];
            tensor<fp16, [1, 150, 512]> input_47_cast_fp16 = transpose(perm = input_47_perm_0, x = var_881_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [1, 150, 512]> linear_20_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_3_conv_out_proj_weight_to_fp16, x = input_47_cast_fp16)[name = string("linear_20_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_886_cast_fp16 = abs(x = linear_20_cast_fp16)[name = string("op_886_cast_fp16")];
            tensor<int32, [1]> var_888_axes_0 = const()[name = string("op_888_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_888_cast_fp16 = reduce_max(axes = var_888_axes_0, keep_dims = var_33, x = var_886_cast_fp16)[name = string("op_888_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_33_cast_fp16 = maximum(x = var_888_cast_fp16, y = var_34_to_fp16)[name = string("s_33_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_37_cast_fp16 = real_div(x = linear_20_cast_fp16, y = s_33_cast_fp16)[name = string("y_37_cast_fp16")];
            fp16 var_35_promoted_15_to_fp16 = const()[name = string("op_35_promoted_15_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_891_cast_fp16 = pow(x = y_37_cast_fp16, y = var_35_promoted_15_to_fp16)[name = string("op_891_cast_fp16")];
            tensor<int32, [1]> var_893_axes_0 = const()[name = string("op_893_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_893_keep_dims_0 = const()[name = string("op_893_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_893_cast_fp16 = reduce_mean(axes = var_893_axes_0, keep_dims = var_893_keep_dims_0, x = var_891_cast_fp16)[name = string("op_893_cast_fp16")];
            fp32 var_894_epsilon_0 = const()[name = string("op_894_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_894_cast_fp16 = inverse(epsilon = var_894_epsilon_0, x = s_33_cast_fp16)[name = string("op_894_cast_fp16")];
            fp16 var_895_to_fp16 = const()[name = string("op_895_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_896_cast_fp16 = mul(x = var_894_cast_fp16, y = var_895_to_fp16)[name = string("op_896_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_897_cast_fp16 = real_div(x = var_896_cast_fp16, y = s_33_cast_fp16)[name = string("op_897_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_898_cast_fp16 = add(x = var_893_cast_fp16, y = var_897_cast_fp16)[name = string("op_898_cast_fp16")];
            fp32 var_899_epsilon_0 = const()[name = string("op_899_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_899_cast_fp16 = rsqrt(epsilon = var_899_epsilon_0, x = var_898_cast_fp16)[name = string("op_899_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_900_cast_fp16 = mul(x = y_37_cast_fp16, y = var_899_cast_fp16)[name = string("op_900_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_3_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28951872)))];
            tensor<fp16, [1, 150, 512]> a_7_cast_fp16 = mul(x = core_backbone_layers_3_post_operator_norm_weight_to_fp16, y = var_900_cast_fp16)[name = string("a_7_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_51_cast_fp16 = add(x = x_47_cast_fp16, y = a_7_cast_fp16)[name = string("x_51_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_904_cast_fp16 = abs(x = x_51_cast_fp16)[name = string("op_904_cast_fp16")];
            tensor<int32, [1]> var_906_axes_0 = const()[name = string("op_906_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_906_cast_fp16 = reduce_max(axes = var_906_axes_0, keep_dims = var_33, x = var_904_cast_fp16)[name = string("op_906_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_35_cast_fp16 = maximum(x = var_906_cast_fp16, y = var_34_to_fp16)[name = string("s_35_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_39_cast_fp16 = real_div(x = x_51_cast_fp16, y = s_35_cast_fp16)[name = string("y_39_cast_fp16")];
            fp16 var_35_promoted_16_to_fp16 = const()[name = string("op_35_promoted_16_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_909_cast_fp16 = pow(x = y_39_cast_fp16, y = var_35_promoted_16_to_fp16)[name = string("op_909_cast_fp16")];
            tensor<int32, [1]> var_911_axes_0 = const()[name = string("op_911_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_911_keep_dims_0 = const()[name = string("op_911_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_911_cast_fp16 = reduce_mean(axes = var_911_axes_0, keep_dims = var_911_keep_dims_0, x = var_909_cast_fp16)[name = string("op_911_cast_fp16")];
            fp32 var_912_epsilon_0 = const()[name = string("op_912_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_912_cast_fp16 = inverse(epsilon = var_912_epsilon_0, x = s_35_cast_fp16)[name = string("op_912_cast_fp16")];
            fp16 var_913_to_fp16 = const()[name = string("op_913_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_914_cast_fp16 = mul(x = var_912_cast_fp16, y = var_913_to_fp16)[name = string("op_914_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_915_cast_fp16 = real_div(x = var_914_cast_fp16, y = s_35_cast_fp16)[name = string("op_915_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_916_cast_fp16 = add(x = var_911_cast_fp16, y = var_915_cast_fp16)[name = string("op_916_cast_fp16")];
            fp32 var_917_epsilon_0 = const()[name = string("op_917_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_917_cast_fp16 = rsqrt(epsilon = var_917_epsilon_0, x = var_916_cast_fp16)[name = string("op_917_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_918_cast_fp16 = mul(x = y_39_cast_fp16, y = var_917_cast_fp16)[name = string("op_918_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_3_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28952960)))];
            tensor<fp16, [1, 150, 512]> input_49_cast_fp16 = mul(x = core_backbone_layers_3_ffn_norm_weight_to_fp16, y = var_918_cast_fp16)[name = string("input_49_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_3_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28954048)))];
            tensor<fp16, [1, 150, 2304]> linear_21_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w1_weight_to_fp16, x = input_49_cast_fp16)[name = string("linear_21_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_3_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31313408)))];
            tensor<fp16, [1, 150, 2304]> linear_22_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w3_weight_to_fp16, x = input_49_cast_fp16)[name = string("linear_22_cast_fp16")];
            fp16 var_927_to_fp16 = const()[name = string("op_927_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 150, 2304]> var_928_cast_fp16 = mul(x = linear_21_cast_fp16, y = var_927_to_fp16)[name = string("op_928_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_931_cast_fp16 = tanh(x = var_928_cast_fp16)[name = string("op_931_cast_fp16")];
            fp16 var_932_to_fp16 = const()[name = string("op_932_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 150, 2304]> var_933_cast_fp16 = add(x = var_931_cast_fp16, y = var_932_to_fp16)[name = string("op_933_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_934_cast_fp16 = mul(x = var_928_cast_fp16, y = var_933_cast_fp16)[name = string("op_934_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> input_51_cast_fp16 = mul(x = var_934_cast_fp16, y = linear_22_cast_fp16)[name = string("input_51_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_3_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33672768)))];
            tensor<fp16, [1, 150, 512]> linear_23_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w2_weight_to_fp16, x = input_51_cast_fp16)[name = string("linear_23_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_939_cast_fp16 = abs(x = linear_23_cast_fp16)[name = string("op_939_cast_fp16")];
            tensor<int32, [1]> var_941_axes_0 = const()[name = string("op_941_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_941_cast_fp16 = reduce_max(axes = var_941_axes_0, keep_dims = var_33, x = var_939_cast_fp16)[name = string("op_941_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_37_cast_fp16 = maximum(x = var_941_cast_fp16, y = var_34_to_fp16)[name = string("s_37_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_41_cast_fp16 = real_div(x = linear_23_cast_fp16, y = s_37_cast_fp16)[name = string("y_41_cast_fp16")];
            fp16 var_35_promoted_17_to_fp16 = const()[name = string("op_35_promoted_17_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_944_cast_fp16 = pow(x = y_41_cast_fp16, y = var_35_promoted_17_to_fp16)[name = string("op_944_cast_fp16")];
            tensor<int32, [1]> var_946_axes_0 = const()[name = string("op_946_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_946_keep_dims_0 = const()[name = string("op_946_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_946_cast_fp16 = reduce_mean(axes = var_946_axes_0, keep_dims = var_946_keep_dims_0, x = var_944_cast_fp16)[name = string("op_946_cast_fp16")];
            fp32 var_947_epsilon_0 = const()[name = string("op_947_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_947_cast_fp16 = inverse(epsilon = var_947_epsilon_0, x = s_37_cast_fp16)[name = string("op_947_cast_fp16")];
            fp16 var_948_to_fp16 = const()[name = string("op_948_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_949_cast_fp16 = mul(x = var_947_cast_fp16, y = var_948_to_fp16)[name = string("op_949_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_950_cast_fp16 = real_div(x = var_949_cast_fp16, y = s_37_cast_fp16)[name = string("op_950_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_951_cast_fp16 = add(x = var_946_cast_fp16, y = var_950_cast_fp16)[name = string("op_951_cast_fp16")];
            fp32 var_952_epsilon_0 = const()[name = string("op_952_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_952_cast_fp16 = rsqrt(epsilon = var_952_epsilon_0, x = var_951_cast_fp16)[name = string("op_952_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_953_cast_fp16 = mul(x = y_41_cast_fp16, y = var_952_cast_fp16)[name = string("op_953_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_3_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36032128)))];
            tensor<fp16, [1, 150, 512]> f_7_cast_fp16 = mul(x = core_backbone_layers_3_post_ffn_norm_weight_to_fp16, y = var_953_cast_fp16)[name = string("f_7_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_57_cast_fp16 = add(x = x_51_cast_fp16, y = f_7_cast_fp16)[name = string("x_57_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_957_cast_fp16 = abs(x = x_57_cast_fp16)[name = string("op_957_cast_fp16")];
            tensor<int32, [1]> var_959_axes_0 = const()[name = string("op_959_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_959_cast_fp16 = reduce_max(axes = var_959_axes_0, keep_dims = var_33, x = var_957_cast_fp16)[name = string("op_959_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_39_cast_fp16 = maximum(x = var_959_cast_fp16, y = var_34_to_fp16)[name = string("s_39_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_43_cast_fp16 = real_div(x = x_57_cast_fp16, y = s_39_cast_fp16)[name = string("y_43_cast_fp16")];
            fp16 var_35_promoted_18_to_fp16 = const()[name = string("op_35_promoted_18_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_962_cast_fp16 = pow(x = y_43_cast_fp16, y = var_35_promoted_18_to_fp16)[name = string("op_962_cast_fp16")];
            tensor<int32, [1]> var_964_axes_0 = const()[name = string("op_964_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_964_keep_dims_0 = const()[name = string("op_964_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_964_cast_fp16 = reduce_mean(axes = var_964_axes_0, keep_dims = var_964_keep_dims_0, x = var_962_cast_fp16)[name = string("op_964_cast_fp16")];
            fp32 var_965_epsilon_0 = const()[name = string("op_965_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_965_cast_fp16 = inverse(epsilon = var_965_epsilon_0, x = s_39_cast_fp16)[name = string("op_965_cast_fp16")];
            fp16 var_966_to_fp16 = const()[name = string("op_966_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_967_cast_fp16 = mul(x = var_965_cast_fp16, y = var_966_to_fp16)[name = string("op_967_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_968_cast_fp16 = real_div(x = var_967_cast_fp16, y = s_39_cast_fp16)[name = string("op_968_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_969_cast_fp16 = add(x = var_964_cast_fp16, y = var_968_cast_fp16)[name = string("op_969_cast_fp16")];
            fp32 var_970_epsilon_0 = const()[name = string("op_970_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_970_cast_fp16 = rsqrt(epsilon = var_970_epsilon_0, x = var_969_cast_fp16)[name = string("op_970_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_971_cast_fp16 = mul(x = y_43_cast_fp16, y = var_970_cast_fp16)[name = string("op_971_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_4_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36033216)))];
            tensor<fp16, [1, 150, 512]> input_53_cast_fp16 = mul(x = core_backbone_layers_4_operator_norm_weight_to_fp16, y = var_971_cast_fp16)[name = string("input_53_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36034304)))];
            tensor<fp16, [1, 150, 512]> linear_24_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_q_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_24_cast_fp16")];
            tensor<int32, [4]> concat_8 = const()[name = string("concat_8"), val = tensor<int32, [4]>([1, 150, 8, 64])];
            tensor<fp16, [1, 150, 8, 64]> x_59_cast_fp16 = reshape(shape = concat_8, x = linear_24_cast_fp16)[name = string("x_59_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> var_978_cast_fp16 = abs(x = x_59_cast_fp16)[name = string("op_978_cast_fp16")];
            tensor<int32, [1]> var_980_axes_0 = const()[name = string("op_980_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 8, 1]> var_980_cast_fp16 = reduce_max(axes = var_980_axes_0, keep_dims = var_33, x = var_978_cast_fp16)[name = string("op_980_cast_fp16")];
            tensor<fp16, [1, 150, 8, 1]> s_41_cast_fp16 = maximum(x = var_980_cast_fp16, y = var_34_to_fp16)[name = string("s_41_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> y_45_cast_fp16 = real_div(x = x_59_cast_fp16, y = s_41_cast_fp16)[name = string("y_45_cast_fp16")];
            fp16 var_35_promoted_19_to_fp16 = const()[name = string("op_35_promoted_19_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 8, 64]> var_983_cast_fp16 = pow(x = y_45_cast_fp16, y = var_35_promoted_19_to_fp16)[name = string("op_983_cast_fp16")];
            tensor<int32, [1]> var_985_axes_0 = const()[name = string("op_985_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_985_keep_dims_0 = const()[name = string("op_985_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 8, 1]> var_985_cast_fp16 = reduce_mean(axes = var_985_axes_0, keep_dims = var_985_keep_dims_0, x = var_983_cast_fp16)[name = string("op_985_cast_fp16")];
            fp32 var_986_epsilon_0 = const()[name = string("op_986_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 8, 1]> var_986_cast_fp16 = inverse(epsilon = var_986_epsilon_0, x = s_41_cast_fp16)[name = string("op_986_cast_fp16")];
            fp16 var_987_to_fp16 = const()[name = string("op_987_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 8, 1]> var_988_cast_fp16 = mul(x = var_986_cast_fp16, y = var_987_to_fp16)[name = string("op_988_cast_fp16")];
            tensor<fp16, [1, 150, 8, 1]> var_989_cast_fp16 = real_div(x = var_988_cast_fp16, y = s_41_cast_fp16)[name = string("op_989_cast_fp16")];
            tensor<fp16, [1, 150, 8, 1]> var_990_cast_fp16 = add(x = var_985_cast_fp16, y = var_989_cast_fp16)[name = string("op_990_cast_fp16")];
            fp32 var_991_epsilon_0 = const()[name = string("op_991_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 8, 1]> var_991_cast_fp16 = rsqrt(epsilon = var_991_epsilon_0, x = var_990_cast_fp16)[name = string("op_991_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> var_992_cast_fp16 = mul(x = y_45_cast_fp16, y = var_991_cast_fp16)[name = string("op_992_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_4_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36558656)))];
            tensor<fp16, [1, 150, 8, 64]> var_993_cast_fp16 = mul(x = core_backbone_layers_4_self_attn_q_norm_weight_to_fp16, y = var_992_cast_fp16)[name = string("op_993_cast_fp16")];
            tensor<fp16, [128, 512]> core_backbone_layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36558848)))];
            tensor<fp16, [1, 150, 128]> linear_25_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_k_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_25_cast_fp16")];
            tensor<int32, [4]> concat_9 = const()[name = string("concat_9"), val = tensor<int32, [4]>([1, 150, 2, 64])];
            tensor<fp16, [1, 150, 2, 64]> x_61_cast_fp16 = reshape(shape = concat_9, x = linear_25_cast_fp16)[name = string("x_61_cast_fp16")];
            tensor<fp16, [1, 150, 2, 64]> var_1000_cast_fp16 = abs(x = x_61_cast_fp16)[name = string("op_1000_cast_fp16")];
            tensor<int32, [1]> var_1002_axes_0 = const()[name = string("op_1002_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 2, 1]> var_1002_cast_fp16 = reduce_max(axes = var_1002_axes_0, keep_dims = var_33, x = var_1000_cast_fp16)[name = string("op_1002_cast_fp16")];
            tensor<fp16, [1, 150, 2, 1]> s_43_cast_fp16 = maximum(x = var_1002_cast_fp16, y = var_34_to_fp16)[name = string("s_43_cast_fp16")];
            tensor<fp16, [1, 150, 2, 64]> y_47_cast_fp16 = real_div(x = x_61_cast_fp16, y = s_43_cast_fp16)[name = string("y_47_cast_fp16")];
            fp16 var_35_promoted_20_to_fp16 = const()[name = string("op_35_promoted_20_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 2, 64]> var_1005_cast_fp16 = pow(x = y_47_cast_fp16, y = var_35_promoted_20_to_fp16)[name = string("op_1005_cast_fp16")];
            tensor<int32, [1]> var_1007_axes_0 = const()[name = string("op_1007_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1007_keep_dims_0 = const()[name = string("op_1007_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 2, 1]> var_1007_cast_fp16 = reduce_mean(axes = var_1007_axes_0, keep_dims = var_1007_keep_dims_0, x = var_1005_cast_fp16)[name = string("op_1007_cast_fp16")];
            fp32 var_1008_epsilon_0 = const()[name = string("op_1008_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 2, 1]> var_1008_cast_fp16 = inverse(epsilon = var_1008_epsilon_0, x = s_43_cast_fp16)[name = string("op_1008_cast_fp16")];
            fp16 var_1009_to_fp16 = const()[name = string("op_1009_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 2, 1]> var_1010_cast_fp16 = mul(x = var_1008_cast_fp16, y = var_1009_to_fp16)[name = string("op_1010_cast_fp16")];
            tensor<fp16, [1, 150, 2, 1]> var_1011_cast_fp16 = real_div(x = var_1010_cast_fp16, y = s_43_cast_fp16)[name = string("op_1011_cast_fp16")];
            tensor<fp16, [1, 150, 2, 1]> var_1012_cast_fp16 = add(x = var_1007_cast_fp16, y = var_1011_cast_fp16)[name = string("op_1012_cast_fp16")];
            fp32 var_1013_epsilon_0 = const()[name = string("op_1013_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 2, 1]> var_1013_cast_fp16 = rsqrt(epsilon = var_1013_epsilon_0, x = var_1012_cast_fp16)[name = string("op_1013_cast_fp16")];
            tensor<fp16, [1, 150, 2, 64]> var_1014_cast_fp16 = mul(x = y_47_cast_fp16, y = var_1013_cast_fp16)[name = string("op_1014_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_4_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36689984)))];
            tensor<fp16, [1, 150, 2, 64]> var_1015_cast_fp16 = mul(x = core_backbone_layers_4_self_attn_k_norm_weight_to_fp16, y = var_1014_cast_fp16)[name = string("op_1015_cast_fp16")];
            tensor<int32, [4]> new_11_perm_0 = const()[name = string("new_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [128, 512]> core_backbone_layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36690176)))];
            tensor<fp16, [1, 150, 128]> linear_26_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_v_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_26_cast_fp16")];
            tensor<int32, [4]> concat_10 = const()[name = string("concat_10"), val = tensor<int32, [4]>([1, 150, 2, 64])];
            tensor<fp16, [1, 150, 2, 64]> var_1020_cast_fp16 = reshape(shape = concat_10, x = linear_26_cast_fp16)[name = string("op_1020_cast_fp16")];
            tensor<int32, [4]> new_13_perm_0 = const()[name = string("new_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_57_begin_0 = const()[name = string("input_57_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
            tensor<int32, [5]> input_57_end_0 = const()[name = string("input_57_end_0"), val = tensor<int32, [5]>([2, 1, 2, 29, 64])];
            tensor<bool, [5]> input_57_end_mask_0 = const()[name = string("input_57_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_57_squeeze_mask_0 = const()[name = string("input_57_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_57_cast_fp16 = slice_by_index(begin = input_57_begin_0, end = input_57_end_0, end_mask = input_57_end_mask_0, squeeze_mask = input_57_squeeze_mask_0, x = k)[name = string("input_57_cast_fp16")];
            tensor<int32, [8]> var_1025_pad_0 = const()[name = string("op_1025_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_1025_mode_0 = const()[name = string("op_1025_mode_0"), val = string("constant")];
            fp16 const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 150, 64]> new_11_cast_fp16 = transpose(perm = new_11_perm_0, x = var_1015_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [1, 2, 179, 64]> var_1025_cast_fp16 = pad(constant_val = const_17_to_fp16, mode = var_1025_mode_0, pad = var_1025_pad_0, x = new_11_cast_fp16)[name = string("op_1025_cast_fp16")];
            tensor<int32, [8]> var_1027_pad_0 = const()[name = string("op_1027_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 150, 0, 0])];
            string var_1027_mode_0 = const()[name = string("op_1027_mode_0"), val = string("constant")];
            fp16 const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 179, 64]> var_1027_cast_fp16 = pad(constant_val = const_18_to_fp16, mode = var_1027_mode_0, pad = var_1027_pad_0, x = input_57_cast_fp16)[name = string("op_1027_cast_fp16")];
            tensor<fp16, [1, 2, 179, 64]> x_65_cast_fp16 = add(x = var_1025_cast_fp16, y = var_1027_cast_fp16)[name = string("x_65_cast_fp16")];
            tensor<int32, [5]> input_61_begin_0 = const()[name = string("input_61_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
            tensor<int32, [5]> input_61_end_0 = const()[name = string("input_61_end_0"), val = tensor<int32, [5]>([2, 1, 2, 29, 64])];
            tensor<bool, [5]> input_61_end_mask_0 = const()[name = string("input_61_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_61_squeeze_mask_0 = const()[name = string("input_61_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_61_cast_fp16 = slice_by_index(begin = input_61_begin_0, end = input_61_end_0, end_mask = input_61_end_mask_0, squeeze_mask = input_61_squeeze_mask_0, x = v)[name = string("input_61_cast_fp16")];
            tensor<int32, [8]> var_1032_pad_0 = const()[name = string("op_1032_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_1032_mode_0 = const()[name = string("op_1032_mode_0"), val = string("constant")];
            fp16 const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 150, 64]> new_13_cast_fp16 = transpose(perm = new_13_perm_0, x = var_1020_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [1, 2, 179, 64]> var_1032_cast_fp16 = pad(constant_val = const_19_to_fp16, mode = var_1032_mode_0, pad = var_1032_pad_0, x = new_13_cast_fp16)[name = string("op_1032_cast_fp16")];
            tensor<int32, [8]> var_1034_pad_0 = const()[name = string("op_1034_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 150, 0, 0])];
            string var_1034_mode_0 = const()[name = string("op_1034_mode_0"), val = string("constant")];
            fp16 const_20_to_fp16 = const()[name = string("const_20_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 179, 64]> var_1034_cast_fp16 = pad(constant_val = const_20_to_fp16, mode = var_1034_mode_0, pad = var_1034_pad_0, x = input_61_cast_fp16)[name = string("op_1034_cast_fp16")];
            tensor<fp16, [1, 2, 179, 64]> v_7_cast_fp16 = add(x = var_1032_cast_fp16, y = var_1034_cast_fp16)[name = string("v_7_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> var_1036_cast_fp16 = mul(x = var_993_cast_fp16, y = const_33_to_fp16)[name = string("op_1036_cast_fp16")];
            tensor<int32, [2]> var_1037_split_sizes_0 = const()[name = string("op_1037_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 const_38 = const()[name = string("const_38"), val = int32(3)];
            tensor<fp16, [1, 150, 8, 32]> var_1037_cast_fp16_0, tensor<fp16, [1, 150, 8, 32]> var_1037_cast_fp16_1 = split(axis = const_38, split_sizes = var_1037_split_sizes_0, x = var_993_cast_fp16)[name = string("op_1037_cast_fp16")];
            fp16 const_21_promoted_to_fp16 = const()[name = string("const_21_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 150, 8, 32]> var_1039_cast_fp16 = mul(x = var_1037_cast_fp16_1, y = const_21_promoted_to_fp16)[name = string("op_1039_cast_fp16")];
            bool var_1041_interleave_0 = const()[name = string("op_1041_interleave_0"), val = bool(false)];
            int32 const_39 = const()[name = string("const_39"), val = int32(3)];
            tensor<fp16, [1, 150, 8, 64]> var_1041_cast_fp16 = concat(axis = const_39, interleave = var_1041_interleave_0, values = (var_1039_cast_fp16, var_1037_cast_fp16_0))[name = string("op_1041_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> var_1042_cast_fp16 = mul(x = var_1041_cast_fp16, y = const_36_to_fp16)[name = string("op_1042_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> q_3_cast_fp16 = add(x = var_1036_cast_fp16, y = var_1042_cast_fp16)[name = string("q_3_cast_fp16")];
            tensor<fp16, [1, 2, 179, 64]> var_1044_cast_fp16 = mul(x = x_65_cast_fp16, y = const_1_to_fp16)[name = string("op_1044_cast_fp16")];
            tensor<int32, [2]> var_1045_split_sizes_0 = const()[name = string("op_1045_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 var_1045_axis_0 = const()[name = string("op_1045_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 2, 179, 32]> var_1045_cast_fp16_0, tensor<fp16, [1, 2, 179, 32]> var_1045_cast_fp16_1 = split(axis = var_1045_axis_0, split_sizes = var_1045_split_sizes_0, x = x_65_cast_fp16)[name = string("op_1045_cast_fp16")];
            fp16 const_22_promoted_to_fp16 = const()[name = string("const_22_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 2, 179, 32]> var_1047_cast_fp16 = mul(x = var_1045_cast_fp16_1, y = const_22_promoted_to_fp16)[name = string("op_1047_cast_fp16")];
            bool var_1049_interleave_0 = const()[name = string("op_1049_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 2, 179, 64]> var_1049_cast_fp16 = concat(axis = var_32, interleave = var_1049_interleave_0, values = (var_1047_cast_fp16, var_1045_cast_fp16_0))[name = string("op_1049_cast_fp16")];
            tensor<fp16, [1, 2, 179, 64]> var_1050_cast_fp16 = mul(x = var_1049_cast_fp16, y = const_2_to_fp16)[name = string("op_1050_cast_fp16")];
            tensor<fp16, [1, 2, 179, 64]> k_7_cast_fp16 = add(x = var_1044_cast_fp16, y = var_1050_cast_fp16)[name = string("k_7_cast_fp16")];
            tensor<int32, [1]> var_1052_axes_0 = const()[name = string("op_1052_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 179, 64]> var_1052_cast_fp16 = expand_dims(axes = var_1052_axes_0, x = k_7_cast_fp16)[name = string("op_1052_cast_fp16")];
            tensor<int32, [5]> var_1054_reps_0 = const()[name = string("op_1054_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 179, 64]> var_1054_cast_fp16 = tile(reps = var_1054_reps_0, x = var_1052_cast_fp16)[name = string("op_1054_cast_fp16")];
            tensor<int32, [4]> concat_12 = const()[name = string("concat_12"), val = tensor<int32, [4]>([1, 8, 179, 64])];
            tensor<fp16, [1, 8, 179, 64]> k_9_cast_fp16 = reshape(shape = concat_12, x = var_1054_cast_fp16)[name = string("k_9_cast_fp16")];
            tensor<int32, [1]> var_1056_axes_0 = const()[name = string("op_1056_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 179, 64]> var_1056_cast_fp16 = expand_dims(axes = var_1056_axes_0, x = v_7_cast_fp16)[name = string("op_1056_cast_fp16")];
            tensor<int32, [5]> var_1058_reps_0 = const()[name = string("op_1058_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 179, 64]> var_1058_cast_fp16 = tile(reps = var_1058_reps_0, x = var_1056_cast_fp16)[name = string("op_1058_cast_fp16")];
            tensor<int32, [4]> concat_14 = const()[name = string("concat_14"), val = tensor<int32, [4]>([1, 8, 179, 64])];
            tensor<fp16, [1, 8, 179, 64]> v_9_cast_fp16 = reshape(shape = concat_14, x = var_1058_cast_fp16)[name = string("v_9_cast_fp16")];
            tensor<int32, [4]> transpose_2_perm_0 = const()[name = string("transpose_2_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1061_transpose_x_1 = const()[name = string("op_1061_transpose_x_1"), val = bool(false)];
            bool var_1061_transpose_y_1 = const()[name = string("op_1061_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 150, 64]> transpose_2_cast_fp16 = transpose(perm = transpose_2_perm_0, x = q_3_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [1, 8, 150, 179]> var_1061_cast_fp16 = matmul(transpose_x = var_1061_transpose_x_1, transpose_y = var_1061_transpose_y_1, x = transpose_2_cast_fp16, y = k_9_cast_fp16)[name = string("op_1061_cast_fp16")];
            fp16 var_1062_to_fp16 = const()[name = string("op_1062_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 8, 150, 179]> s_45_cast_fp16 = mul(x = var_1061_cast_fp16, y = var_1062_to_fp16)[name = string("s_45_cast_fp16")];
            tensor<fp16, [1, 8, 150, 179]> x_67_cast_fp16 = add(x = s_45_cast_fp16, y = mask)[name = string("x_67_cast_fp16")];
            tensor<int32, [1]> var_1066_axes_0 = const()[name = string("op_1066_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 150, 1]> var_1066_cast_fp16 = reduce_max(axes = var_1066_axes_0, keep_dims = var_33, x = x_67_cast_fp16)[name = string("op_1066_cast_fp16")];
            tensor<fp16, [1, 8, 150, 179]> x_69_cast_fp16 = sub(x = x_67_cast_fp16, y = var_1066_cast_fp16)[name = string("x_69_cast_fp16")];
            tensor<fp16, [1, 8, 150, 179]> x_71_cast_fp16 = exp(x = x_69_cast_fp16)[name = string("x_71_cast_fp16")];
            tensor<int32, [1]> var_1070_axes_0 = const()[name = string("op_1070_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1070_keep_dims_0 = const()[name = string("op_1070_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 150, 1]> var_1070_cast_fp16 = reduce_sum(axes = var_1070_axes_0, keep_dims = var_1070_keep_dims_0, x = x_71_cast_fp16)[name = string("op_1070_cast_fp16")];
            tensor<fp16, [1, 8, 150, 179]> var_1071_cast_fp16 = real_div(x = x_71_cast_fp16, y = var_1070_cast_fp16)[name = string("op_1071_cast_fp16")];
            bool o_7_transpose_x_0 = const()[name = string("o_7_transpose_x_0"), val = bool(false)];
            bool o_7_transpose_y_0 = const()[name = string("o_7_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 150, 64]> o_7_cast_fp16 = matmul(transpose_x = o_7_transpose_x_0, transpose_y = o_7_transpose_y_0, x = var_1071_cast_fp16, y = v_9_cast_fp16)[name = string("o_7_cast_fp16")];
            tensor<int32, [4]> o_9_perm_0 = const()[name = string("o_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [8, 512]> core_backbone_layers_4_self_attn_gate_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_gate_proj_weight_to_fp16"), val = tensor<fp16, [8, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36821312)))];
            tensor<fp16, [8]> linear_27_bias_0_to_fp16 = const()[name = string("linear_27_bias_0_to_fp16"), val = tensor<fp16, [8]>([0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0])];
            tensor<fp16, [1, 150, 8]> linear_27_cast_fp16 = linear(bias = linear_27_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_gate_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_27_cast_fp16")];
            fp16 var_1077_to_fp16 = const()[name = string("op_1077_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 150, 8]> var_1078_cast_fp16 = mul(x = linear_27_cast_fp16, y = var_1077_to_fp16)[name = string("op_1078_cast_fp16")];
            tensor<fp16, [1, 150, 8]> var_1079_cast_fp16 = tanh(x = var_1078_cast_fp16)[name = string("op_1079_cast_fp16")];
            fp16 var_1080_to_fp16 = const()[name = string("op_1080_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 150, 8]> var_1081_cast_fp16 = add(x = var_1079_cast_fp16, y = var_1080_to_fp16)[name = string("op_1081_cast_fp16")];
            fp16 var_1082_to_fp16 = const()[name = string("op_1082_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 150, 8]> var_1083_cast_fp16 = mul(x = var_1081_cast_fp16, y = var_1082_to_fp16)[name = string("op_1083_cast_fp16")];
            tensor<int32, [1]> var_1084_axes_0 = const()[name = string("op_1084_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 8, 1]> var_1084_cast_fp16 = expand_dims(axes = var_1084_axes_0, x = var_1083_cast_fp16)[name = string("op_1084_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> o_9_cast_fp16 = transpose(perm = o_9_perm_0, x = o_7_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [1, 150, 8, 64]> o_11_cast_fp16 = mul(x = o_9_cast_fp16, y = var_1084_cast_fp16)[name = string("o_11_cast_fp16")];
            tensor<int32, [3]> concat_15 = const()[name = string("concat_15"), val = tensor<int32, [3]>([1, 150, 512])];
            tensor<fp16, [1, 150, 512]> input_63_cast_fp16 = reshape(shape = concat_15, x = o_11_cast_fp16)[name = string("input_63_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36829568)))];
            tensor<fp16, [1, 150, 512]> linear_28_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_o_proj_weight_to_fp16, x = input_63_cast_fp16)[name = string("linear_28_cast_fp16")];
            tensor<int32, [4]> var_1091_begin_0 = const()[name = string("op_1091_begin_0"), val = tensor<int32, [4]>([0, 0, 150, 0])];
            tensor<int32, [4]> var_1091_end_0 = const()[name = string("op_1091_end_0"), val = tensor<int32, [4]>([1, 2, 179, 64])];
            tensor<bool, [4]> var_1091_end_mask_0 = const()[name = string("op_1091_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_1091_cast_fp16 = slice_by_index(begin = var_1091_begin_0, end = var_1091_end_0, end_mask = var_1091_end_mask_0, x = x_65_cast_fp16)[name = string("op_1091_cast_fp16")];
            tensor<int32, [4]> var_1094_begin_0 = const()[name = string("op_1094_begin_0"), val = tensor<int32, [4]>([0, 0, 150, 0])];
            tensor<int32, [4]> var_1094_end_0 = const()[name = string("op_1094_end_0"), val = tensor<int32, [4]>([1, 2, 179, 64])];
            tensor<bool, [4]> var_1094_end_mask_0 = const()[name = string("op_1094_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_1094_cast_fp16 = slice_by_index(begin = var_1094_begin_0, end = var_1094_end_0, end_mask = var_1094_end_mask_0, x = v_7_cast_fp16)[name = string("op_1094_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1096_cast_fp16 = abs(x = linear_28_cast_fp16)[name = string("op_1096_cast_fp16")];
            tensor<int32, [1]> var_1098_axes_0 = const()[name = string("op_1098_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1098_cast_fp16 = reduce_max(axes = var_1098_axes_0, keep_dims = var_33, x = var_1096_cast_fp16)[name = string("op_1098_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_47_cast_fp16 = maximum(x = var_1098_cast_fp16, y = var_34_to_fp16)[name = string("s_47_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_49_cast_fp16 = real_div(x = linear_28_cast_fp16, y = s_47_cast_fp16)[name = string("y_49_cast_fp16")];
            fp16 var_35_promoted_21_to_fp16 = const()[name = string("op_35_promoted_21_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1101_cast_fp16 = pow(x = y_49_cast_fp16, y = var_35_promoted_21_to_fp16)[name = string("op_1101_cast_fp16")];
            tensor<int32, [1]> var_1103_axes_0 = const()[name = string("op_1103_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1103_keep_dims_0 = const()[name = string("op_1103_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1103_cast_fp16 = reduce_mean(axes = var_1103_axes_0, keep_dims = var_1103_keep_dims_0, x = var_1101_cast_fp16)[name = string("op_1103_cast_fp16")];
            fp32 var_1104_epsilon_0 = const()[name = string("op_1104_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1104_cast_fp16 = inverse(epsilon = var_1104_epsilon_0, x = s_47_cast_fp16)[name = string("op_1104_cast_fp16")];
            fp16 var_1105_to_fp16 = const()[name = string("op_1105_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1106_cast_fp16 = mul(x = var_1104_cast_fp16, y = var_1105_to_fp16)[name = string("op_1106_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1107_cast_fp16 = real_div(x = var_1106_cast_fp16, y = s_47_cast_fp16)[name = string("op_1107_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1108_cast_fp16 = add(x = var_1103_cast_fp16, y = var_1107_cast_fp16)[name = string("op_1108_cast_fp16")];
            fp32 var_1109_epsilon_0 = const()[name = string("op_1109_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1109_cast_fp16 = rsqrt(epsilon = var_1109_epsilon_0, x = var_1108_cast_fp16)[name = string("op_1109_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1110_cast_fp16 = mul(x = y_49_cast_fp16, y = var_1109_cast_fp16)[name = string("op_1110_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_4_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37353920)))];
            tensor<fp16, [1, 150, 512]> a_9_cast_fp16 = mul(x = core_backbone_layers_4_post_operator_norm_weight_to_fp16, y = var_1110_cast_fp16)[name = string("a_9_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_77_cast_fp16 = add(x = x_57_cast_fp16, y = a_9_cast_fp16)[name = string("x_77_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1114_cast_fp16 = abs(x = x_77_cast_fp16)[name = string("op_1114_cast_fp16")];
            tensor<int32, [1]> var_1116_axes_0 = const()[name = string("op_1116_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1116_cast_fp16 = reduce_max(axes = var_1116_axes_0, keep_dims = var_33, x = var_1114_cast_fp16)[name = string("op_1116_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_49_cast_fp16 = maximum(x = var_1116_cast_fp16, y = var_34_to_fp16)[name = string("s_49_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_51_cast_fp16 = real_div(x = x_77_cast_fp16, y = s_49_cast_fp16)[name = string("y_51_cast_fp16")];
            fp16 var_35_promoted_22_to_fp16 = const()[name = string("op_35_promoted_22_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1119_cast_fp16 = pow(x = y_51_cast_fp16, y = var_35_promoted_22_to_fp16)[name = string("op_1119_cast_fp16")];
            tensor<int32, [1]> var_1121_axes_0 = const()[name = string("op_1121_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1121_keep_dims_0 = const()[name = string("op_1121_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1121_cast_fp16 = reduce_mean(axes = var_1121_axes_0, keep_dims = var_1121_keep_dims_0, x = var_1119_cast_fp16)[name = string("op_1121_cast_fp16")];
            fp32 var_1122_epsilon_0 = const()[name = string("op_1122_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1122_cast_fp16 = inverse(epsilon = var_1122_epsilon_0, x = s_49_cast_fp16)[name = string("op_1122_cast_fp16")];
            fp16 var_1123_to_fp16 = const()[name = string("op_1123_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1124_cast_fp16 = mul(x = var_1122_cast_fp16, y = var_1123_to_fp16)[name = string("op_1124_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1125_cast_fp16 = real_div(x = var_1124_cast_fp16, y = s_49_cast_fp16)[name = string("op_1125_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1126_cast_fp16 = add(x = var_1121_cast_fp16, y = var_1125_cast_fp16)[name = string("op_1126_cast_fp16")];
            fp32 var_1127_epsilon_0 = const()[name = string("op_1127_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1127_cast_fp16 = rsqrt(epsilon = var_1127_epsilon_0, x = var_1126_cast_fp16)[name = string("op_1127_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1128_cast_fp16 = mul(x = y_51_cast_fp16, y = var_1127_cast_fp16)[name = string("op_1128_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_4_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37355008)))];
            tensor<fp16, [1, 150, 512]> input_65_cast_fp16 = mul(x = core_backbone_layers_4_ffn_norm_weight_to_fp16, y = var_1128_cast_fp16)[name = string("input_65_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_4_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37356096)))];
            tensor<fp16, [1, 150, 2304]> linear_29_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w1_weight_to_fp16, x = input_65_cast_fp16)[name = string("linear_29_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_4_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39715456)))];
            tensor<fp16, [1, 150, 2304]> linear_30_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w3_weight_to_fp16, x = input_65_cast_fp16)[name = string("linear_30_cast_fp16")];
            fp16 var_1137_to_fp16 = const()[name = string("op_1137_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 150, 2304]> var_1138_cast_fp16 = mul(x = linear_29_cast_fp16, y = var_1137_to_fp16)[name = string("op_1138_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_1141_cast_fp16 = tanh(x = var_1138_cast_fp16)[name = string("op_1141_cast_fp16")];
            fp16 var_1142_to_fp16 = const()[name = string("op_1142_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 150, 2304]> var_1143_cast_fp16 = add(x = var_1141_cast_fp16, y = var_1142_to_fp16)[name = string("op_1143_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_1144_cast_fp16 = mul(x = var_1138_cast_fp16, y = var_1143_cast_fp16)[name = string("op_1144_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> input_67_cast_fp16 = mul(x = var_1144_cast_fp16, y = linear_30_cast_fp16)[name = string("input_67_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_4_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42074816)))];
            tensor<fp16, [1, 150, 512]> linear_31_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w2_weight_to_fp16, x = input_67_cast_fp16)[name = string("linear_31_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1149_cast_fp16 = abs(x = linear_31_cast_fp16)[name = string("op_1149_cast_fp16")];
            tensor<int32, [1]> var_1151_axes_0 = const()[name = string("op_1151_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1151_cast_fp16 = reduce_max(axes = var_1151_axes_0, keep_dims = var_33, x = var_1149_cast_fp16)[name = string("op_1151_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_51_cast_fp16 = maximum(x = var_1151_cast_fp16, y = var_34_to_fp16)[name = string("s_51_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_53_cast_fp16 = real_div(x = linear_31_cast_fp16, y = s_51_cast_fp16)[name = string("y_53_cast_fp16")];
            fp16 var_35_promoted_23_to_fp16 = const()[name = string("op_35_promoted_23_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1154_cast_fp16 = pow(x = y_53_cast_fp16, y = var_35_promoted_23_to_fp16)[name = string("op_1154_cast_fp16")];
            tensor<int32, [1]> var_1156_axes_0 = const()[name = string("op_1156_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1156_keep_dims_0 = const()[name = string("op_1156_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1156_cast_fp16 = reduce_mean(axes = var_1156_axes_0, keep_dims = var_1156_keep_dims_0, x = var_1154_cast_fp16)[name = string("op_1156_cast_fp16")];
            fp32 var_1157_epsilon_0 = const()[name = string("op_1157_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1157_cast_fp16 = inverse(epsilon = var_1157_epsilon_0, x = s_51_cast_fp16)[name = string("op_1157_cast_fp16")];
            fp16 var_1158_to_fp16 = const()[name = string("op_1158_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1159_cast_fp16 = mul(x = var_1157_cast_fp16, y = var_1158_to_fp16)[name = string("op_1159_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1160_cast_fp16 = real_div(x = var_1159_cast_fp16, y = s_51_cast_fp16)[name = string("op_1160_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1161_cast_fp16 = add(x = var_1156_cast_fp16, y = var_1160_cast_fp16)[name = string("op_1161_cast_fp16")];
            fp32 var_1162_epsilon_0 = const()[name = string("op_1162_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1162_cast_fp16 = rsqrt(epsilon = var_1162_epsilon_0, x = var_1161_cast_fp16)[name = string("op_1162_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1163_cast_fp16 = mul(x = y_53_cast_fp16, y = var_1162_cast_fp16)[name = string("op_1163_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_4_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44434176)))];
            tensor<fp16, [1, 150, 512]> f_9_cast_fp16 = mul(x = core_backbone_layers_4_post_ffn_norm_weight_to_fp16, y = var_1163_cast_fp16)[name = string("f_9_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_83_cast_fp16 = add(x = x_77_cast_fp16, y = f_9_cast_fp16)[name = string("x_83_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1167_cast_fp16 = abs(x = x_83_cast_fp16)[name = string("op_1167_cast_fp16")];
            tensor<int32, [1]> var_1169_axes_0 = const()[name = string("op_1169_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1169_cast_fp16 = reduce_max(axes = var_1169_axes_0, keep_dims = var_33, x = var_1167_cast_fp16)[name = string("op_1169_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_53_cast_fp16 = maximum(x = var_1169_cast_fp16, y = var_34_to_fp16)[name = string("s_53_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_55_cast_fp16 = real_div(x = x_83_cast_fp16, y = s_53_cast_fp16)[name = string("y_55_cast_fp16")];
            fp16 var_35_promoted_24_to_fp16 = const()[name = string("op_35_promoted_24_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1172_cast_fp16 = pow(x = y_55_cast_fp16, y = var_35_promoted_24_to_fp16)[name = string("op_1172_cast_fp16")];
            tensor<int32, [1]> var_1174_axes_0 = const()[name = string("op_1174_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1174_keep_dims_0 = const()[name = string("op_1174_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1174_cast_fp16 = reduce_mean(axes = var_1174_axes_0, keep_dims = var_1174_keep_dims_0, x = var_1172_cast_fp16)[name = string("op_1174_cast_fp16")];
            fp32 var_1175_epsilon_0 = const()[name = string("op_1175_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1175_cast_fp16 = inverse(epsilon = var_1175_epsilon_0, x = s_53_cast_fp16)[name = string("op_1175_cast_fp16")];
            fp16 var_1176_to_fp16 = const()[name = string("op_1176_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1177_cast_fp16 = mul(x = var_1175_cast_fp16, y = var_1176_to_fp16)[name = string("op_1177_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1178_cast_fp16 = real_div(x = var_1177_cast_fp16, y = s_53_cast_fp16)[name = string("op_1178_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1179_cast_fp16 = add(x = var_1174_cast_fp16, y = var_1178_cast_fp16)[name = string("op_1179_cast_fp16")];
            fp32 var_1180_epsilon_0 = const()[name = string("op_1180_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1180_cast_fp16 = rsqrt(epsilon = var_1180_epsilon_0, x = var_1179_cast_fp16)[name = string("op_1180_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1181_cast_fp16 = mul(x = y_55_cast_fp16, y = var_1180_cast_fp16)[name = string("op_1181_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_5_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44435264)))];
            tensor<fp16, [1, 150, 512]> input_69_cast_fp16 = mul(x = core_backbone_layers_5_operator_norm_weight_to_fp16, y = var_1181_cast_fp16)[name = string("input_69_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_5_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44436352)))];
            tensor<fp16, [1, 150, 1536]> linear_32_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_5_conv_in_proj_weight_to_fp16, x = input_69_cast_fp16)[name = string("linear_32_cast_fp16")];
            tensor<int32, [3]> bcx_7_perm_0 = const()[name = string("bcx_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_17_begin_0 = const()[name = string("b_17_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_17_end_0 = const()[name = string("b_17_end_0"), val = tensor<int32, [3]>([1, 512, 150])];
            tensor<bool, [3]> b_17_end_mask_0 = const()[name = string("b_17_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 150]> bcx_7_cast_fp16 = transpose(perm = bcx_7_perm_0, x = linear_32_cast_fp16)[name = string("transpose_13")];
            tensor<fp16, [1, 512, 150]> b_17_cast_fp16 = slice_by_index(begin = b_17_begin_0, end = b_17_end_0, end_mask = b_17_end_mask_0, x = bcx_7_cast_fp16)[name = string("b_17_cast_fp16")];
            tensor<int32, [3]> c_7_begin_0 = const()[name = string("c_7_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_7_end_0 = const()[name = string("c_7_end_0"), val = tensor<int32, [3]>([1, 1024, 150])];
            tensor<bool, [3]> c_7_end_mask_0 = const()[name = string("c_7_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 150]> c_7_cast_fp16 = slice_by_index(begin = c_7_begin_0, end = c_7_end_0, end_mask = c_7_end_mask_0, x = bcx_7_cast_fp16)[name = string("c_7_cast_fp16")];
            tensor<int32, [3]> xx_7_begin_0 = const()[name = string("xx_7_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_7_end_0 = const()[name = string("xx_7_end_0"), val = tensor<int32, [3]>([1, 1536, 150])];
            tensor<bool, [3]> xx_7_end_mask_0 = const()[name = string("xx_7_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 150]> xx_7_cast_fp16 = slice_by_index(begin = xx_7_begin_0, end = xx_7_end_0, end_mask = xx_7_end_mask_0, x = bcx_7_cast_fp16)[name = string("xx_7_cast_fp16")];
            tensor<fp16, [1, 512, 150]> new_15_cast_fp16 = mul(x = b_17_cast_fp16, y = xx_7_cast_fp16)[name = string("new_15_cast_fp16")];
            tensor<int32, [4]> input_73_begin_0 = const()[name = string("input_73_begin_0"), val = tensor<int32, [4]>([3, 0, 0, 0])];
            tensor<int32, [4]> input_73_end_0 = const()[name = string("input_73_end_0"), val = tensor<int32, [4]>([4, 1, 512, 6])];
            tensor<bool, [4]> input_73_end_mask_0 = const()[name = string("input_73_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_73_squeeze_mask_0 = const()[name = string("input_73_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_73_cast_fp16 = slice_by_index(begin = input_73_begin_0, end = input_73_end_0, end_mask = input_73_end_mask_0, squeeze_mask = input_73_squeeze_mask_0, x = conv)[name = string("input_73_cast_fp16")];
            tensor<int32, [6]> var_1196_pad_0 = const()[name = string("op_1196_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_1196_mode_0 = const()[name = string("op_1196_mode_0"), val = string("constant")];
            fp16 const_23_to_fp16 = const()[name = string("const_23_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 156]> var_1196_cast_fp16 = pad(constant_val = const_23_to_fp16, mode = var_1196_mode_0, pad = var_1196_pad_0, x = new_15_cast_fp16)[name = string("op_1196_cast_fp16")];
            tensor<int32, [6]> var_1198_pad_0 = const()[name = string("op_1198_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 150])];
            string var_1198_mode_0 = const()[name = string("op_1198_mode_0"), val = string("constant")];
            fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 156]> var_1198_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = var_1198_mode_0, pad = var_1198_pad_0, x = input_73_cast_fp16)[name = string("op_1198_cast_fp16")];
            tensor<fp16, [1, 512, 156]> win_13_cast_fp16 = add(x = var_1196_cast_fp16, y = var_1198_cast_fp16)[name = string("win_13_cast_fp16")];
            tensor<int32, [3]> var_1200_begin_0 = const()[name = string("op_1200_begin_0"), val = tensor<int32, [3]>([0, 0, 150])];
            tensor<int32, [3]> var_1200_end_0 = const()[name = string("op_1200_end_0"), val = tensor<int32, [3]>([1, 512, 156])];
            tensor<bool, [3]> var_1200_end_mask_0 = const()[name = string("op_1200_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_1200_cast_fp16 = slice_by_index(begin = var_1200_begin_0, end = var_1200_end_0, end_mask = var_1200_end_mask_0, x = win_13_cast_fp16)[name = string("op_1200_cast_fp16")];
            string var_1206_pad_type_0 = const()[name = string("op_1206_pad_type_0"), val = string("valid")];
            int32 var_1206_groups_0 = const()[name = string("op_1206_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_1206_strides_0 = const()[name = string("op_1206_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1206_pad_0 = const()[name = string("op_1206_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1206_dilations_0 = const()[name = string("op_1206_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_5_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_5_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46009280)))];
            tensor<fp16, [1, 512, 150]> var_1206_cast_fp16 = conv(dilations = var_1206_dilations_0, groups = var_1206_groups_0, pad = var_1206_pad_0, pad_type = var_1206_pad_type_0, strides = var_1206_strides_0, weight = core_backbone_layers_5_conv_conv_weight_to_fp16, x = win_13_cast_fp16)[name = string("op_1206_cast_fp16")];
            tensor<fp16, [1, 512, 150]> var_1208_cast_fp16 = mul(x = c_7_cast_fp16, y = var_1206_cast_fp16)[name = string("op_1208_cast_fp16")];
            tensor<int32, [3]> input_75_perm_0 = const()[name = string("input_75_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_5_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46016512)))];
            tensor<fp16, [1, 150, 512]> input_75_cast_fp16 = transpose(perm = input_75_perm_0, x = var_1208_cast_fp16)[name = string("transpose_12")];
            tensor<fp16, [1, 150, 512]> linear_33_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_5_conv_out_proj_weight_to_fp16, x = input_75_cast_fp16)[name = string("linear_33_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1213_cast_fp16 = abs(x = linear_33_cast_fp16)[name = string("op_1213_cast_fp16")];
            tensor<int32, [1]> var_1215_axes_0 = const()[name = string("op_1215_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1215_cast_fp16 = reduce_max(axes = var_1215_axes_0, keep_dims = var_33, x = var_1213_cast_fp16)[name = string("op_1215_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_55_cast_fp16 = maximum(x = var_1215_cast_fp16, y = var_34_to_fp16)[name = string("s_55_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_59_cast_fp16 = real_div(x = linear_33_cast_fp16, y = s_55_cast_fp16)[name = string("y_59_cast_fp16")];
            fp16 var_35_promoted_25_to_fp16 = const()[name = string("op_35_promoted_25_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1218_cast_fp16 = pow(x = y_59_cast_fp16, y = var_35_promoted_25_to_fp16)[name = string("op_1218_cast_fp16")];
            tensor<int32, [1]> var_1220_axes_0 = const()[name = string("op_1220_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1220_keep_dims_0 = const()[name = string("op_1220_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1220_cast_fp16 = reduce_mean(axes = var_1220_axes_0, keep_dims = var_1220_keep_dims_0, x = var_1218_cast_fp16)[name = string("op_1220_cast_fp16")];
            fp32 var_1221_epsilon_0 = const()[name = string("op_1221_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1221_cast_fp16 = inverse(epsilon = var_1221_epsilon_0, x = s_55_cast_fp16)[name = string("op_1221_cast_fp16")];
            fp16 var_1222_to_fp16 = const()[name = string("op_1222_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1223_cast_fp16 = mul(x = var_1221_cast_fp16, y = var_1222_to_fp16)[name = string("op_1223_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1224_cast_fp16 = real_div(x = var_1223_cast_fp16, y = s_55_cast_fp16)[name = string("op_1224_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1225_cast_fp16 = add(x = var_1220_cast_fp16, y = var_1224_cast_fp16)[name = string("op_1225_cast_fp16")];
            fp32 var_1226_epsilon_0 = const()[name = string("op_1226_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1226_cast_fp16 = rsqrt(epsilon = var_1226_epsilon_0, x = var_1225_cast_fp16)[name = string("op_1226_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1227_cast_fp16 = mul(x = y_59_cast_fp16, y = var_1226_cast_fp16)[name = string("op_1227_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_5_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46540864)))];
            tensor<fp16, [1, 150, 512]> a_11_cast_fp16 = mul(x = core_backbone_layers_5_post_operator_norm_weight_to_fp16, y = var_1227_cast_fp16)[name = string("a_11_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_87_cast_fp16 = add(x = x_83_cast_fp16, y = a_11_cast_fp16)[name = string("x_87_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1231_cast_fp16 = abs(x = x_87_cast_fp16)[name = string("op_1231_cast_fp16")];
            tensor<int32, [1]> var_1233_axes_0 = const()[name = string("op_1233_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1233_cast_fp16 = reduce_max(axes = var_1233_axes_0, keep_dims = var_33, x = var_1231_cast_fp16)[name = string("op_1233_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_57_cast_fp16 = maximum(x = var_1233_cast_fp16, y = var_34_to_fp16)[name = string("s_57_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_61_cast_fp16 = real_div(x = x_87_cast_fp16, y = s_57_cast_fp16)[name = string("y_61_cast_fp16")];
            fp16 var_35_promoted_26_to_fp16 = const()[name = string("op_35_promoted_26_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1236_cast_fp16 = pow(x = y_61_cast_fp16, y = var_35_promoted_26_to_fp16)[name = string("op_1236_cast_fp16")];
            tensor<int32, [1]> var_1238_axes_0 = const()[name = string("op_1238_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1238_keep_dims_0 = const()[name = string("op_1238_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1238_cast_fp16 = reduce_mean(axes = var_1238_axes_0, keep_dims = var_1238_keep_dims_0, x = var_1236_cast_fp16)[name = string("op_1238_cast_fp16")];
            fp32 var_1239_epsilon_0 = const()[name = string("op_1239_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1239_cast_fp16 = inverse(epsilon = var_1239_epsilon_0, x = s_57_cast_fp16)[name = string("op_1239_cast_fp16")];
            fp16 var_1240_to_fp16 = const()[name = string("op_1240_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1241_cast_fp16 = mul(x = var_1239_cast_fp16, y = var_1240_to_fp16)[name = string("op_1241_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1242_cast_fp16 = real_div(x = var_1241_cast_fp16, y = s_57_cast_fp16)[name = string("op_1242_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1243_cast_fp16 = add(x = var_1238_cast_fp16, y = var_1242_cast_fp16)[name = string("op_1243_cast_fp16")];
            fp32 var_1244_epsilon_0 = const()[name = string("op_1244_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1244_cast_fp16 = rsqrt(epsilon = var_1244_epsilon_0, x = var_1243_cast_fp16)[name = string("op_1244_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1245_cast_fp16 = mul(x = y_61_cast_fp16, y = var_1244_cast_fp16)[name = string("op_1245_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_5_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46541952)))];
            tensor<fp16, [1, 150, 512]> input_77_cast_fp16 = mul(x = core_backbone_layers_5_ffn_norm_weight_to_fp16, y = var_1245_cast_fp16)[name = string("input_77_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_5_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46543040)))];
            tensor<fp16, [1, 150, 2304]> linear_34_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w1_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_34_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_5_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48902400)))];
            tensor<fp16, [1, 150, 2304]> linear_35_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w3_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_35_cast_fp16")];
            fp16 var_1254_to_fp16 = const()[name = string("op_1254_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 150, 2304]> var_1255_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_1254_to_fp16)[name = string("op_1255_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_1258_cast_fp16 = tanh(x = var_1255_cast_fp16)[name = string("op_1258_cast_fp16")];
            fp16 var_1259_to_fp16 = const()[name = string("op_1259_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 150, 2304]> var_1260_cast_fp16 = add(x = var_1258_cast_fp16, y = var_1259_to_fp16)[name = string("op_1260_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_1261_cast_fp16 = mul(x = var_1255_cast_fp16, y = var_1260_cast_fp16)[name = string("op_1261_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> input_79_cast_fp16 = mul(x = var_1261_cast_fp16, y = linear_35_cast_fp16)[name = string("input_79_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_5_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51261760)))];
            tensor<fp16, [1, 150, 512]> linear_36_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w2_weight_to_fp16, x = input_79_cast_fp16)[name = string("linear_36_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1266_cast_fp16 = abs(x = linear_36_cast_fp16)[name = string("op_1266_cast_fp16")];
            tensor<int32, [1]> var_1268_axes_0 = const()[name = string("op_1268_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1268_cast_fp16 = reduce_max(axes = var_1268_axes_0, keep_dims = var_33, x = var_1266_cast_fp16)[name = string("op_1268_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_59_cast_fp16 = maximum(x = var_1268_cast_fp16, y = var_34_to_fp16)[name = string("s_59_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_63_cast_fp16 = real_div(x = linear_36_cast_fp16, y = s_59_cast_fp16)[name = string("y_63_cast_fp16")];
            fp16 var_35_promoted_27_to_fp16 = const()[name = string("op_35_promoted_27_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1271_cast_fp16 = pow(x = y_63_cast_fp16, y = var_35_promoted_27_to_fp16)[name = string("op_1271_cast_fp16")];
            tensor<int32, [1]> var_1273_axes_0 = const()[name = string("op_1273_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1273_keep_dims_0 = const()[name = string("op_1273_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1273_cast_fp16 = reduce_mean(axes = var_1273_axes_0, keep_dims = var_1273_keep_dims_0, x = var_1271_cast_fp16)[name = string("op_1273_cast_fp16")];
            fp32 var_1274_epsilon_0 = const()[name = string("op_1274_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1274_cast_fp16 = inverse(epsilon = var_1274_epsilon_0, x = s_59_cast_fp16)[name = string("op_1274_cast_fp16")];
            fp16 var_1275_to_fp16 = const()[name = string("op_1275_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1276_cast_fp16 = mul(x = var_1274_cast_fp16, y = var_1275_to_fp16)[name = string("op_1276_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1277_cast_fp16 = real_div(x = var_1276_cast_fp16, y = s_59_cast_fp16)[name = string("op_1277_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1278_cast_fp16 = add(x = var_1273_cast_fp16, y = var_1277_cast_fp16)[name = string("op_1278_cast_fp16")];
            fp32 var_1279_epsilon_0 = const()[name = string("op_1279_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1279_cast_fp16 = rsqrt(epsilon = var_1279_epsilon_0, x = var_1278_cast_fp16)[name = string("op_1279_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1280_cast_fp16 = mul(x = y_63_cast_fp16, y = var_1279_cast_fp16)[name = string("op_1280_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_5_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53621120)))];
            tensor<fp16, [1, 150, 512]> f_11_cast_fp16 = mul(x = core_backbone_layers_5_post_ffn_norm_weight_to_fp16, y = var_1280_cast_fp16)[name = string("f_11_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_93_cast_fp16 = add(x = x_87_cast_fp16, y = f_11_cast_fp16)[name = string("x_93_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1284_cast_fp16 = abs(x = x_93_cast_fp16)[name = string("op_1284_cast_fp16")];
            tensor<int32, [1]> var_1286_axes_0 = const()[name = string("op_1286_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1286_cast_fp16 = reduce_max(axes = var_1286_axes_0, keep_dims = var_33, x = var_1284_cast_fp16)[name = string("op_1286_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_61_cast_fp16 = maximum(x = var_1286_cast_fp16, y = var_34_to_fp16)[name = string("s_61_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_65_cast_fp16 = real_div(x = x_93_cast_fp16, y = s_61_cast_fp16)[name = string("y_65_cast_fp16")];
            fp16 var_35_promoted_28_to_fp16 = const()[name = string("op_35_promoted_28_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1289_cast_fp16 = pow(x = y_65_cast_fp16, y = var_35_promoted_28_to_fp16)[name = string("op_1289_cast_fp16")];
            tensor<int32, [1]> var_1291_axes_0 = const()[name = string("op_1291_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1291_keep_dims_0 = const()[name = string("op_1291_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1291_cast_fp16 = reduce_mean(axes = var_1291_axes_0, keep_dims = var_1291_keep_dims_0, x = var_1289_cast_fp16)[name = string("op_1291_cast_fp16")];
            fp32 var_1292_epsilon_0 = const()[name = string("op_1292_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1292_cast_fp16 = inverse(epsilon = var_1292_epsilon_0, x = s_61_cast_fp16)[name = string("op_1292_cast_fp16")];
            fp16 var_1293_to_fp16 = const()[name = string("op_1293_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1294_cast_fp16 = mul(x = var_1292_cast_fp16, y = var_1293_to_fp16)[name = string("op_1294_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1295_cast_fp16 = real_div(x = var_1294_cast_fp16, y = s_61_cast_fp16)[name = string("op_1295_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1296_cast_fp16 = add(x = var_1291_cast_fp16, y = var_1295_cast_fp16)[name = string("op_1296_cast_fp16")];
            fp32 var_1297_epsilon_0 = const()[name = string("op_1297_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1297_cast_fp16 = rsqrt(epsilon = var_1297_epsilon_0, x = var_1296_cast_fp16)[name = string("op_1297_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1298_cast_fp16 = mul(x = y_65_cast_fp16, y = var_1297_cast_fp16)[name = string("op_1298_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_6_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53622208)))];
            tensor<fp16, [1, 150, 512]> input_81_cast_fp16 = mul(x = core_backbone_layers_6_operator_norm_weight_to_fp16, y = var_1298_cast_fp16)[name = string("input_81_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53623296)))];
            tensor<fp16, [1, 150, 512]> linear_37_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_q_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_37_cast_fp16")];
            tensor<int32, [4]> concat_16 = const()[name = string("concat_16"), val = tensor<int32, [4]>([1, 150, 8, 64])];
            tensor<fp16, [1, 150, 8, 64]> x_95_cast_fp16 = reshape(shape = concat_16, x = linear_37_cast_fp16)[name = string("x_95_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> var_1305_cast_fp16 = abs(x = x_95_cast_fp16)[name = string("op_1305_cast_fp16")];
            tensor<int32, [1]> var_1307_axes_0 = const()[name = string("op_1307_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 8, 1]> var_1307_cast_fp16 = reduce_max(axes = var_1307_axes_0, keep_dims = var_33, x = var_1305_cast_fp16)[name = string("op_1307_cast_fp16")];
            tensor<fp16, [1, 150, 8, 1]> s_63_cast_fp16 = maximum(x = var_1307_cast_fp16, y = var_34_to_fp16)[name = string("s_63_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> y_67_cast_fp16 = real_div(x = x_95_cast_fp16, y = s_63_cast_fp16)[name = string("y_67_cast_fp16")];
            fp16 var_35_promoted_29_to_fp16 = const()[name = string("op_35_promoted_29_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 8, 64]> var_1310_cast_fp16 = pow(x = y_67_cast_fp16, y = var_35_promoted_29_to_fp16)[name = string("op_1310_cast_fp16")];
            tensor<int32, [1]> var_1312_axes_0 = const()[name = string("op_1312_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1312_keep_dims_0 = const()[name = string("op_1312_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 8, 1]> var_1312_cast_fp16 = reduce_mean(axes = var_1312_axes_0, keep_dims = var_1312_keep_dims_0, x = var_1310_cast_fp16)[name = string("op_1312_cast_fp16")];
            fp32 var_1313_epsilon_0 = const()[name = string("op_1313_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 8, 1]> var_1313_cast_fp16 = inverse(epsilon = var_1313_epsilon_0, x = s_63_cast_fp16)[name = string("op_1313_cast_fp16")];
            fp16 var_1314_to_fp16 = const()[name = string("op_1314_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 8, 1]> var_1315_cast_fp16 = mul(x = var_1313_cast_fp16, y = var_1314_to_fp16)[name = string("op_1315_cast_fp16")];
            tensor<fp16, [1, 150, 8, 1]> var_1316_cast_fp16 = real_div(x = var_1315_cast_fp16, y = s_63_cast_fp16)[name = string("op_1316_cast_fp16")];
            tensor<fp16, [1, 150, 8, 1]> var_1317_cast_fp16 = add(x = var_1312_cast_fp16, y = var_1316_cast_fp16)[name = string("op_1317_cast_fp16")];
            fp32 var_1318_epsilon_0 = const()[name = string("op_1318_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 8, 1]> var_1318_cast_fp16 = rsqrt(epsilon = var_1318_epsilon_0, x = var_1317_cast_fp16)[name = string("op_1318_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> var_1319_cast_fp16 = mul(x = y_67_cast_fp16, y = var_1318_cast_fp16)[name = string("op_1319_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_6_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54147648)))];
            tensor<fp16, [1, 150, 8, 64]> var_1320_cast_fp16 = mul(x = core_backbone_layers_6_self_attn_q_norm_weight_to_fp16, y = var_1319_cast_fp16)[name = string("op_1320_cast_fp16")];
            tensor<fp16, [128, 512]> core_backbone_layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54147840)))];
            tensor<fp16, [1, 150, 128]> linear_38_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_k_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_38_cast_fp16")];
            tensor<int32, [4]> concat_17 = const()[name = string("concat_17"), val = tensor<int32, [4]>([1, 150, 2, 64])];
            tensor<fp16, [1, 150, 2, 64]> x_97_cast_fp16 = reshape(shape = concat_17, x = linear_38_cast_fp16)[name = string("x_97_cast_fp16")];
            tensor<fp16, [1, 150, 2, 64]> var_1327_cast_fp16 = abs(x = x_97_cast_fp16)[name = string("op_1327_cast_fp16")];
            tensor<int32, [1]> var_1329_axes_0 = const()[name = string("op_1329_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 2, 1]> var_1329_cast_fp16 = reduce_max(axes = var_1329_axes_0, keep_dims = var_33, x = var_1327_cast_fp16)[name = string("op_1329_cast_fp16")];
            tensor<fp16, [1, 150, 2, 1]> s_65_cast_fp16 = maximum(x = var_1329_cast_fp16, y = var_34_to_fp16)[name = string("s_65_cast_fp16")];
            tensor<fp16, [1, 150, 2, 64]> y_69_cast_fp16 = real_div(x = x_97_cast_fp16, y = s_65_cast_fp16)[name = string("y_69_cast_fp16")];
            fp16 var_35_promoted_30_to_fp16 = const()[name = string("op_35_promoted_30_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 2, 64]> var_1332_cast_fp16 = pow(x = y_69_cast_fp16, y = var_35_promoted_30_to_fp16)[name = string("op_1332_cast_fp16")];
            tensor<int32, [1]> var_1334_axes_0 = const()[name = string("op_1334_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1334_keep_dims_0 = const()[name = string("op_1334_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 2, 1]> var_1334_cast_fp16 = reduce_mean(axes = var_1334_axes_0, keep_dims = var_1334_keep_dims_0, x = var_1332_cast_fp16)[name = string("op_1334_cast_fp16")];
            fp32 var_1335_epsilon_0 = const()[name = string("op_1335_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 2, 1]> var_1335_cast_fp16 = inverse(epsilon = var_1335_epsilon_0, x = s_65_cast_fp16)[name = string("op_1335_cast_fp16")];
            fp16 var_1336_to_fp16 = const()[name = string("op_1336_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 2, 1]> var_1337_cast_fp16 = mul(x = var_1335_cast_fp16, y = var_1336_to_fp16)[name = string("op_1337_cast_fp16")];
            tensor<fp16, [1, 150, 2, 1]> var_1338_cast_fp16 = real_div(x = var_1337_cast_fp16, y = s_65_cast_fp16)[name = string("op_1338_cast_fp16")];
            tensor<fp16, [1, 150, 2, 1]> var_1339_cast_fp16 = add(x = var_1334_cast_fp16, y = var_1338_cast_fp16)[name = string("op_1339_cast_fp16")];
            fp32 var_1340_epsilon_0 = const()[name = string("op_1340_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 2, 1]> var_1340_cast_fp16 = rsqrt(epsilon = var_1340_epsilon_0, x = var_1339_cast_fp16)[name = string("op_1340_cast_fp16")];
            tensor<fp16, [1, 150, 2, 64]> var_1341_cast_fp16 = mul(x = y_69_cast_fp16, y = var_1340_cast_fp16)[name = string("op_1341_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_6_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54278976)))];
            tensor<fp16, [1, 150, 2, 64]> var_1342_cast_fp16 = mul(x = core_backbone_layers_6_self_attn_k_norm_weight_to_fp16, y = var_1341_cast_fp16)[name = string("op_1342_cast_fp16")];
            tensor<int32, [4]> new_17_perm_0 = const()[name = string("new_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [128, 512]> core_backbone_layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54279168)))];
            tensor<fp16, [1, 150, 128]> linear_39_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_v_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_39_cast_fp16")];
            tensor<int32, [4]> concat_18 = const()[name = string("concat_18"), val = tensor<int32, [4]>([1, 150, 2, 64])];
            tensor<fp16, [1, 150, 2, 64]> var_1347_cast_fp16 = reshape(shape = concat_18, x = linear_39_cast_fp16)[name = string("op_1347_cast_fp16")];
            tensor<int32, [4]> new_19_perm_0 = const()[name = string("new_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_85_begin_0 = const()[name = string("input_85_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])];
            tensor<int32, [5]> input_85_end_0 = const()[name = string("input_85_end_0"), val = tensor<int32, [5]>([3, 1, 2, 29, 64])];
            tensor<bool, [5]> input_85_end_mask_0 = const()[name = string("input_85_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_85_squeeze_mask_0 = const()[name = string("input_85_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_85_cast_fp16 = slice_by_index(begin = input_85_begin_0, end = input_85_end_0, end_mask = input_85_end_mask_0, squeeze_mask = input_85_squeeze_mask_0, x = k)[name = string("input_85_cast_fp16")];
            tensor<int32, [8]> var_1352_pad_0 = const()[name = string("op_1352_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_1352_mode_0 = const()[name = string("op_1352_mode_0"), val = string("constant")];
            fp16 const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 150, 64]> new_17_cast_fp16 = transpose(perm = new_17_perm_0, x = var_1342_cast_fp16)[name = string("transpose_11")];
            tensor<fp16, [1, 2, 179, 64]> var_1352_cast_fp16 = pad(constant_val = const_25_to_fp16, mode = var_1352_mode_0, pad = var_1352_pad_0, x = new_17_cast_fp16)[name = string("op_1352_cast_fp16")];
            tensor<int32, [8]> var_1354_pad_0 = const()[name = string("op_1354_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 150, 0, 0])];
            string var_1354_mode_0 = const()[name = string("op_1354_mode_0"), val = string("constant")];
            fp16 const_26_to_fp16 = const()[name = string("const_26_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 179, 64]> var_1354_cast_fp16 = pad(constant_val = const_26_to_fp16, mode = var_1354_mode_0, pad = var_1354_pad_0, x = input_85_cast_fp16)[name = string("op_1354_cast_fp16")];
            tensor<fp16, [1, 2, 179, 64]> x_101_cast_fp16 = add(x = var_1352_cast_fp16, y = var_1354_cast_fp16)[name = string("x_101_cast_fp16")];
            tensor<int32, [5]> input_89_begin_0 = const()[name = string("input_89_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])];
            tensor<int32, [5]> input_89_end_0 = const()[name = string("input_89_end_0"), val = tensor<int32, [5]>([3, 1, 2, 29, 64])];
            tensor<bool, [5]> input_89_end_mask_0 = const()[name = string("input_89_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_89_squeeze_mask_0 = const()[name = string("input_89_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_89_cast_fp16 = slice_by_index(begin = input_89_begin_0, end = input_89_end_0, end_mask = input_89_end_mask_0, squeeze_mask = input_89_squeeze_mask_0, x = v)[name = string("input_89_cast_fp16")];
            tensor<int32, [8]> var_1359_pad_0 = const()[name = string("op_1359_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_1359_mode_0 = const()[name = string("op_1359_mode_0"), val = string("constant")];
            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 150, 64]> new_19_cast_fp16 = transpose(perm = new_19_perm_0, x = var_1347_cast_fp16)[name = string("transpose_10")];
            tensor<fp16, [1, 2, 179, 64]> var_1359_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = var_1359_mode_0, pad = var_1359_pad_0, x = new_19_cast_fp16)[name = string("op_1359_cast_fp16")];
            tensor<int32, [8]> var_1361_pad_0 = const()[name = string("op_1361_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 150, 0, 0])];
            string var_1361_mode_0 = const()[name = string("op_1361_mode_0"), val = string("constant")];
            fp16 const_28_to_fp16 = const()[name = string("const_28_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 179, 64]> var_1361_cast_fp16 = pad(constant_val = const_28_to_fp16, mode = var_1361_mode_0, pad = var_1361_pad_0, x = input_89_cast_fp16)[name = string("op_1361_cast_fp16")];
            tensor<fp16, [1, 2, 179, 64]> v_11_cast_fp16 = add(x = var_1359_cast_fp16, y = var_1361_cast_fp16)[name = string("v_11_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> var_1363_cast_fp16 = mul(x = var_1320_cast_fp16, y = const_33_to_fp16)[name = string("op_1363_cast_fp16")];
            tensor<int32, [2]> var_1364_split_sizes_0 = const()[name = string("op_1364_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 const_42 = const()[name = string("const_42"), val = int32(3)];
            tensor<fp16, [1, 150, 8, 32]> var_1364_cast_fp16_0, tensor<fp16, [1, 150, 8, 32]> var_1364_cast_fp16_1 = split(axis = const_42, split_sizes = var_1364_split_sizes_0, x = var_1320_cast_fp16)[name = string("op_1364_cast_fp16")];
            fp16 const_29_promoted_to_fp16 = const()[name = string("const_29_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 150, 8, 32]> var_1366_cast_fp16 = mul(x = var_1364_cast_fp16_1, y = const_29_promoted_to_fp16)[name = string("op_1366_cast_fp16")];
            bool var_1368_interleave_0 = const()[name = string("op_1368_interleave_0"), val = bool(false)];
            int32 const_43 = const()[name = string("const_43"), val = int32(3)];
            tensor<fp16, [1, 150, 8, 64]> var_1368_cast_fp16 = concat(axis = const_43, interleave = var_1368_interleave_0, values = (var_1366_cast_fp16, var_1364_cast_fp16_0))[name = string("op_1368_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> var_1369_cast_fp16 = mul(x = var_1368_cast_fp16, y = const_36_to_fp16)[name = string("op_1369_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> q_cast_fp16 = add(x = var_1363_cast_fp16, y = var_1369_cast_fp16)[name = string("q_cast_fp16")];
            tensor<fp16, [1, 2, 179, 64]> var_1371_cast_fp16 = mul(x = x_101_cast_fp16, y = const_1_to_fp16)[name = string("op_1371_cast_fp16")];
            tensor<int32, [2]> var_1372_split_sizes_0 = const()[name = string("op_1372_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 var_1372_axis_0 = const()[name = string("op_1372_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 2, 179, 32]> var_1372_cast_fp16_0, tensor<fp16, [1, 2, 179, 32]> var_1372_cast_fp16_1 = split(axis = var_1372_axis_0, split_sizes = var_1372_split_sizes_0, x = x_101_cast_fp16)[name = string("op_1372_cast_fp16")];
            fp16 const_30_promoted_to_fp16 = const()[name = string("const_30_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 2, 179, 32]> var_1374_cast_fp16 = mul(x = var_1372_cast_fp16_1, y = const_30_promoted_to_fp16)[name = string("op_1374_cast_fp16")];
            bool var_1376_interleave_0 = const()[name = string("op_1376_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 2, 179, 64]> var_1376_cast_fp16 = concat(axis = var_32, interleave = var_1376_interleave_0, values = (var_1374_cast_fp16, var_1372_cast_fp16_0))[name = string("op_1376_cast_fp16")];
            tensor<fp16, [1, 2, 179, 64]> var_1377_cast_fp16 = mul(x = var_1376_cast_fp16, y = const_2_to_fp16)[name = string("op_1377_cast_fp16")];
            tensor<fp16, [1, 2, 179, 64]> k_11_cast_fp16 = add(x = var_1371_cast_fp16, y = var_1377_cast_fp16)[name = string("k_11_cast_fp16")];
            tensor<int32, [1]> var_1379_axes_0 = const()[name = string("op_1379_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 179, 64]> var_1379_cast_fp16 = expand_dims(axes = var_1379_axes_0, x = k_11_cast_fp16)[name = string("op_1379_cast_fp16")];
            tensor<int32, [5]> var_1381_reps_0 = const()[name = string("op_1381_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 179, 64]> var_1381_cast_fp16 = tile(reps = var_1381_reps_0, x = var_1379_cast_fp16)[name = string("op_1381_cast_fp16")];
            tensor<int32, [4]> concat_20 = const()[name = string("concat_20"), val = tensor<int32, [4]>([1, 8, 179, 64])];
            tensor<fp16, [1, 8, 179, 64]> k_cast_fp16 = reshape(shape = concat_20, x = var_1381_cast_fp16)[name = string("k_cast_fp16")];
            tensor<int32, [1]> var_1383_axes_0 = const()[name = string("op_1383_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 179, 64]> var_1383_cast_fp16 = expand_dims(axes = var_1383_axes_0, x = v_11_cast_fp16)[name = string("op_1383_cast_fp16")];
            tensor<int32, [5]> var_1385_reps_0 = const()[name = string("op_1385_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 179, 64]> var_1385_cast_fp16 = tile(reps = var_1385_reps_0, x = var_1383_cast_fp16)[name = string("op_1385_cast_fp16")];
            tensor<int32, [4]> concat_22 = const()[name = string("concat_22"), val = tensor<int32, [4]>([1, 8, 179, 64])];
            tensor<fp16, [1, 8, 179, 64]> v_cast_fp16 = reshape(shape = concat_22, x = var_1385_cast_fp16)[name = string("v_cast_fp16")];
            tensor<int32, [4]> transpose_4_perm_0 = const()[name = string("transpose_4_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1388_transpose_x_1 = const()[name = string("op_1388_transpose_x_1"), val = bool(false)];
            bool var_1388_transpose_y_1 = const()[name = string("op_1388_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 150, 64]> transpose_4_cast_fp16 = transpose(perm = transpose_4_perm_0, x = q_cast_fp16)[name = string("transpose_9")];
            tensor<fp16, [1, 8, 150, 179]> var_1388_cast_fp16 = matmul(transpose_x = var_1388_transpose_x_1, transpose_y = var_1388_transpose_y_1, x = transpose_4_cast_fp16, y = k_cast_fp16)[name = string("op_1388_cast_fp16")];
            fp16 var_1389_to_fp16 = const()[name = string("op_1389_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 8, 150, 179]> s_67_cast_fp16 = mul(x = var_1388_cast_fp16, y = var_1389_to_fp16)[name = string("s_67_cast_fp16")];
            tensor<fp16, [1, 8, 150, 179]> x_103_cast_fp16 = add(x = s_67_cast_fp16, y = mask)[name = string("x_103_cast_fp16")];
            tensor<int32, [1]> var_1393_axes_0 = const()[name = string("op_1393_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 150, 1]> var_1393_cast_fp16 = reduce_max(axes = var_1393_axes_0, keep_dims = var_33, x = x_103_cast_fp16)[name = string("op_1393_cast_fp16")];
            tensor<fp16, [1, 8, 150, 179]> x_105_cast_fp16 = sub(x = x_103_cast_fp16, y = var_1393_cast_fp16)[name = string("x_105_cast_fp16")];
            tensor<fp16, [1, 8, 150, 179]> x_107_cast_fp16 = exp(x = x_105_cast_fp16)[name = string("x_107_cast_fp16")];
            tensor<int32, [1]> var_1397_axes_0 = const()[name = string("op_1397_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1397_keep_dims_0 = const()[name = string("op_1397_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 150, 1]> var_1397_cast_fp16 = reduce_sum(axes = var_1397_axes_0, keep_dims = var_1397_keep_dims_0, x = x_107_cast_fp16)[name = string("op_1397_cast_fp16")];
            tensor<fp16, [1, 8, 150, 179]> var_1398_cast_fp16 = real_div(x = x_107_cast_fp16, y = var_1397_cast_fp16)[name = string("op_1398_cast_fp16")];
            bool o_13_transpose_x_0 = const()[name = string("o_13_transpose_x_0"), val = bool(false)];
            bool o_13_transpose_y_0 = const()[name = string("o_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 150, 64]> o_13_cast_fp16 = matmul(transpose_x = o_13_transpose_x_0, transpose_y = o_13_transpose_y_0, x = var_1398_cast_fp16, y = v_cast_fp16)[name = string("o_13_cast_fp16")];
            tensor<int32, [4]> o_15_perm_0 = const()[name = string("o_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [8, 512]> core_backbone_layers_6_self_attn_gate_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_gate_proj_weight_to_fp16"), val = tensor<fp16, [8, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54410304)))];
            tensor<fp16, [8]> linear_40_bias_0_to_fp16 = const()[name = string("linear_40_bias_0_to_fp16"), val = tensor<fp16, [8]>([0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0])];
            tensor<fp16, [1, 150, 8]> linear_40_cast_fp16 = linear(bias = linear_40_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_gate_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_40_cast_fp16")];
            fp16 var_1404_to_fp16 = const()[name = string("op_1404_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 150, 8]> var_1405_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_1404_to_fp16)[name = string("op_1405_cast_fp16")];
            tensor<fp16, [1, 150, 8]> var_1406_cast_fp16 = tanh(x = var_1405_cast_fp16)[name = string("op_1406_cast_fp16")];
            fp16 var_1407_to_fp16 = const()[name = string("op_1407_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 150, 8]> var_1408_cast_fp16 = add(x = var_1406_cast_fp16, y = var_1407_to_fp16)[name = string("op_1408_cast_fp16")];
            fp16 var_1409_to_fp16 = const()[name = string("op_1409_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 150, 8]> var_1410_cast_fp16 = mul(x = var_1408_cast_fp16, y = var_1409_to_fp16)[name = string("op_1410_cast_fp16")];
            tensor<int32, [1]> var_1411_axes_0 = const()[name = string("op_1411_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 8, 1]> var_1411_cast_fp16 = expand_dims(axes = var_1411_axes_0, x = var_1410_cast_fp16)[name = string("op_1411_cast_fp16")];
            tensor<fp16, [1, 150, 8, 64]> o_15_cast_fp16 = transpose(perm = o_15_perm_0, x = o_13_cast_fp16)[name = string("transpose_8")];
            tensor<fp16, [1, 150, 8, 64]> o_cast_fp16 = mul(x = o_15_cast_fp16, y = var_1411_cast_fp16)[name = string("o_cast_fp16")];
            tensor<int32, [3]> concat_23 = const()[name = string("concat_23"), val = tensor<int32, [3]>([1, 150, 512])];
            tensor<fp16, [1, 150, 512]> input_91_cast_fp16 = reshape(shape = concat_23, x = o_cast_fp16)[name = string("input_91_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54418560)))];
            tensor<fp16, [1, 150, 512]> linear_41_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_o_proj_weight_to_fp16, x = input_91_cast_fp16)[name = string("linear_41_cast_fp16")];
            tensor<int32, [4]> var_1418_begin_0 = const()[name = string("op_1418_begin_0"), val = tensor<int32, [4]>([0, 0, 150, 0])];
            tensor<int32, [4]> var_1418_end_0 = const()[name = string("op_1418_end_0"), val = tensor<int32, [4]>([1, 2, 179, 64])];
            tensor<bool, [4]> var_1418_end_mask_0 = const()[name = string("op_1418_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_1418_cast_fp16 = slice_by_index(begin = var_1418_begin_0, end = var_1418_end_0, end_mask = var_1418_end_mask_0, x = x_101_cast_fp16)[name = string("op_1418_cast_fp16")];
            tensor<int32, [4]> var_1421_begin_0 = const()[name = string("op_1421_begin_0"), val = tensor<int32, [4]>([0, 0, 150, 0])];
            tensor<int32, [4]> var_1421_end_0 = const()[name = string("op_1421_end_0"), val = tensor<int32, [4]>([1, 2, 179, 64])];
            tensor<bool, [4]> var_1421_end_mask_0 = const()[name = string("op_1421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_1421_cast_fp16 = slice_by_index(begin = var_1421_begin_0, end = var_1421_end_0, end_mask = var_1421_end_mask_0, x = v_11_cast_fp16)[name = string("op_1421_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1423_cast_fp16 = abs(x = linear_41_cast_fp16)[name = string("op_1423_cast_fp16")];
            tensor<int32, [1]> var_1425_axes_0 = const()[name = string("op_1425_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1425_cast_fp16 = reduce_max(axes = var_1425_axes_0, keep_dims = var_33, x = var_1423_cast_fp16)[name = string("op_1425_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_69_cast_fp16 = maximum(x = var_1425_cast_fp16, y = var_34_to_fp16)[name = string("s_69_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_71_cast_fp16 = real_div(x = linear_41_cast_fp16, y = s_69_cast_fp16)[name = string("y_71_cast_fp16")];
            fp16 var_35_promoted_31_to_fp16 = const()[name = string("op_35_promoted_31_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1428_cast_fp16 = pow(x = y_71_cast_fp16, y = var_35_promoted_31_to_fp16)[name = string("op_1428_cast_fp16")];
            tensor<int32, [1]> var_1430_axes_0 = const()[name = string("op_1430_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1430_keep_dims_0 = const()[name = string("op_1430_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1430_cast_fp16 = reduce_mean(axes = var_1430_axes_0, keep_dims = var_1430_keep_dims_0, x = var_1428_cast_fp16)[name = string("op_1430_cast_fp16")];
            fp32 var_1431_epsilon_0 = const()[name = string("op_1431_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1431_cast_fp16 = inverse(epsilon = var_1431_epsilon_0, x = s_69_cast_fp16)[name = string("op_1431_cast_fp16")];
            fp16 var_1432_to_fp16 = const()[name = string("op_1432_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1433_cast_fp16 = mul(x = var_1431_cast_fp16, y = var_1432_to_fp16)[name = string("op_1433_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1434_cast_fp16 = real_div(x = var_1433_cast_fp16, y = s_69_cast_fp16)[name = string("op_1434_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1435_cast_fp16 = add(x = var_1430_cast_fp16, y = var_1434_cast_fp16)[name = string("op_1435_cast_fp16")];
            fp32 var_1436_epsilon_0 = const()[name = string("op_1436_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1436_cast_fp16 = rsqrt(epsilon = var_1436_epsilon_0, x = var_1435_cast_fp16)[name = string("op_1436_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1437_cast_fp16 = mul(x = y_71_cast_fp16, y = var_1436_cast_fp16)[name = string("op_1437_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_6_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54942912)))];
            tensor<fp16, [1, 150, 512]> a_13_cast_fp16 = mul(x = core_backbone_layers_6_post_operator_norm_weight_to_fp16, y = var_1437_cast_fp16)[name = string("a_13_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_113_cast_fp16 = add(x = x_93_cast_fp16, y = a_13_cast_fp16)[name = string("x_113_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1441_cast_fp16 = abs(x = x_113_cast_fp16)[name = string("op_1441_cast_fp16")];
            tensor<int32, [1]> var_1443_axes_0 = const()[name = string("op_1443_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1443_cast_fp16 = reduce_max(axes = var_1443_axes_0, keep_dims = var_33, x = var_1441_cast_fp16)[name = string("op_1443_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_71_cast_fp16 = maximum(x = var_1443_cast_fp16, y = var_34_to_fp16)[name = string("s_71_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_73_cast_fp16 = real_div(x = x_113_cast_fp16, y = s_71_cast_fp16)[name = string("y_73_cast_fp16")];
            fp16 var_35_promoted_32_to_fp16 = const()[name = string("op_35_promoted_32_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1446_cast_fp16 = pow(x = y_73_cast_fp16, y = var_35_promoted_32_to_fp16)[name = string("op_1446_cast_fp16")];
            tensor<int32, [1]> var_1448_axes_0 = const()[name = string("op_1448_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1448_keep_dims_0 = const()[name = string("op_1448_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1448_cast_fp16 = reduce_mean(axes = var_1448_axes_0, keep_dims = var_1448_keep_dims_0, x = var_1446_cast_fp16)[name = string("op_1448_cast_fp16")];
            fp32 var_1449_epsilon_0 = const()[name = string("op_1449_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1449_cast_fp16 = inverse(epsilon = var_1449_epsilon_0, x = s_71_cast_fp16)[name = string("op_1449_cast_fp16")];
            fp16 var_1450_to_fp16 = const()[name = string("op_1450_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1451_cast_fp16 = mul(x = var_1449_cast_fp16, y = var_1450_to_fp16)[name = string("op_1451_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1452_cast_fp16 = real_div(x = var_1451_cast_fp16, y = s_71_cast_fp16)[name = string("op_1452_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1453_cast_fp16 = add(x = var_1448_cast_fp16, y = var_1452_cast_fp16)[name = string("op_1453_cast_fp16")];
            fp32 var_1454_epsilon_0 = const()[name = string("op_1454_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1454_cast_fp16 = rsqrt(epsilon = var_1454_epsilon_0, x = var_1453_cast_fp16)[name = string("op_1454_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1455_cast_fp16 = mul(x = y_73_cast_fp16, y = var_1454_cast_fp16)[name = string("op_1455_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_6_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54944000)))];
            tensor<fp16, [1, 150, 512]> input_93_cast_fp16 = mul(x = core_backbone_layers_6_ffn_norm_weight_to_fp16, y = var_1455_cast_fp16)[name = string("input_93_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_6_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54945088)))];
            tensor<fp16, [1, 150, 2304]> linear_42_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w1_weight_to_fp16, x = input_93_cast_fp16)[name = string("linear_42_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_6_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57304448)))];
            tensor<fp16, [1, 150, 2304]> linear_43_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w3_weight_to_fp16, x = input_93_cast_fp16)[name = string("linear_43_cast_fp16")];
            fp16 var_1464_to_fp16 = const()[name = string("op_1464_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 150, 2304]> var_1465_cast_fp16 = mul(x = linear_42_cast_fp16, y = var_1464_to_fp16)[name = string("op_1465_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_1468_cast_fp16 = tanh(x = var_1465_cast_fp16)[name = string("op_1468_cast_fp16")];
            fp16 var_1469_to_fp16 = const()[name = string("op_1469_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 150, 2304]> var_1470_cast_fp16 = add(x = var_1468_cast_fp16, y = var_1469_to_fp16)[name = string("op_1470_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_1471_cast_fp16 = mul(x = var_1465_cast_fp16, y = var_1470_cast_fp16)[name = string("op_1471_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> input_95_cast_fp16 = mul(x = var_1471_cast_fp16, y = linear_43_cast_fp16)[name = string("input_95_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_6_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59663808)))];
            tensor<fp16, [1, 150, 512]> linear_44_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w2_weight_to_fp16, x = input_95_cast_fp16)[name = string("linear_44_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1476_cast_fp16 = abs(x = linear_44_cast_fp16)[name = string("op_1476_cast_fp16")];
            tensor<int32, [1]> var_1478_axes_0 = const()[name = string("op_1478_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1478_cast_fp16 = reduce_max(axes = var_1478_axes_0, keep_dims = var_33, x = var_1476_cast_fp16)[name = string("op_1478_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_73_cast_fp16 = maximum(x = var_1478_cast_fp16, y = var_34_to_fp16)[name = string("s_73_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_75_cast_fp16 = real_div(x = linear_44_cast_fp16, y = s_73_cast_fp16)[name = string("y_75_cast_fp16")];
            fp16 var_35_promoted_33_to_fp16 = const()[name = string("op_35_promoted_33_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1481_cast_fp16 = pow(x = y_75_cast_fp16, y = var_35_promoted_33_to_fp16)[name = string("op_1481_cast_fp16")];
            tensor<int32, [1]> var_1483_axes_0 = const()[name = string("op_1483_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1483_keep_dims_0 = const()[name = string("op_1483_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1483_cast_fp16 = reduce_mean(axes = var_1483_axes_0, keep_dims = var_1483_keep_dims_0, x = var_1481_cast_fp16)[name = string("op_1483_cast_fp16")];
            fp32 var_1484_epsilon_0 = const()[name = string("op_1484_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1484_cast_fp16 = inverse(epsilon = var_1484_epsilon_0, x = s_73_cast_fp16)[name = string("op_1484_cast_fp16")];
            fp16 var_1485_to_fp16 = const()[name = string("op_1485_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1486_cast_fp16 = mul(x = var_1484_cast_fp16, y = var_1485_to_fp16)[name = string("op_1486_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1487_cast_fp16 = real_div(x = var_1486_cast_fp16, y = s_73_cast_fp16)[name = string("op_1487_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1488_cast_fp16 = add(x = var_1483_cast_fp16, y = var_1487_cast_fp16)[name = string("op_1488_cast_fp16")];
            fp32 var_1489_epsilon_0 = const()[name = string("op_1489_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1489_cast_fp16 = rsqrt(epsilon = var_1489_epsilon_0, x = var_1488_cast_fp16)[name = string("op_1489_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1490_cast_fp16 = mul(x = y_75_cast_fp16, y = var_1489_cast_fp16)[name = string("op_1490_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_6_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62023168)))];
            tensor<fp16, [1, 150, 512]> f_13_cast_fp16 = mul(x = core_backbone_layers_6_post_ffn_norm_weight_to_fp16, y = var_1490_cast_fp16)[name = string("f_13_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_119_cast_fp16 = add(x = x_113_cast_fp16, y = f_13_cast_fp16)[name = string("x_119_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1494_cast_fp16 = abs(x = x_119_cast_fp16)[name = string("op_1494_cast_fp16")];
            tensor<int32, [1]> var_1496_axes_0 = const()[name = string("op_1496_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1496_cast_fp16 = reduce_max(axes = var_1496_axes_0, keep_dims = var_33, x = var_1494_cast_fp16)[name = string("op_1496_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_75_cast_fp16 = maximum(x = var_1496_cast_fp16, y = var_34_to_fp16)[name = string("s_75_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_77_cast_fp16 = real_div(x = x_119_cast_fp16, y = s_75_cast_fp16)[name = string("y_77_cast_fp16")];
            fp16 var_35_promoted_34_to_fp16 = const()[name = string("op_35_promoted_34_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1499_cast_fp16 = pow(x = y_77_cast_fp16, y = var_35_promoted_34_to_fp16)[name = string("op_1499_cast_fp16")];
            tensor<int32, [1]> var_1501_axes_0 = const()[name = string("op_1501_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1501_keep_dims_0 = const()[name = string("op_1501_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1501_cast_fp16 = reduce_mean(axes = var_1501_axes_0, keep_dims = var_1501_keep_dims_0, x = var_1499_cast_fp16)[name = string("op_1501_cast_fp16")];
            fp32 var_1502_epsilon_0 = const()[name = string("op_1502_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1502_cast_fp16 = inverse(epsilon = var_1502_epsilon_0, x = s_75_cast_fp16)[name = string("op_1502_cast_fp16")];
            fp16 var_1503_to_fp16 = const()[name = string("op_1503_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1504_cast_fp16 = mul(x = var_1502_cast_fp16, y = var_1503_to_fp16)[name = string("op_1504_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1505_cast_fp16 = real_div(x = var_1504_cast_fp16, y = s_75_cast_fp16)[name = string("op_1505_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1506_cast_fp16 = add(x = var_1501_cast_fp16, y = var_1505_cast_fp16)[name = string("op_1506_cast_fp16")];
            fp32 var_1507_epsilon_0 = const()[name = string("op_1507_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1507_cast_fp16 = rsqrt(epsilon = var_1507_epsilon_0, x = var_1506_cast_fp16)[name = string("op_1507_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1508_cast_fp16 = mul(x = y_77_cast_fp16, y = var_1507_cast_fp16)[name = string("op_1508_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_7_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62024256)))];
            tensor<fp16, [1, 150, 512]> input_97_cast_fp16 = mul(x = core_backbone_layers_7_operator_norm_weight_to_fp16, y = var_1508_cast_fp16)[name = string("input_97_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_7_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62025344)))];
            tensor<fp16, [1, 150, 1536]> linear_45_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_7_conv_in_proj_weight_to_fp16, x = input_97_cast_fp16)[name = string("linear_45_cast_fp16")];
            tensor<int32, [3]> bcx_perm_0 = const()[name = string("bcx_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_23_begin_0 = const()[name = string("b_23_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_23_end_0 = const()[name = string("b_23_end_0"), val = tensor<int32, [3]>([1, 512, 150])];
            tensor<bool, [3]> b_23_end_mask_0 = const()[name = string("b_23_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 150]> bcx_cast_fp16 = transpose(perm = bcx_perm_0, x = linear_45_cast_fp16)[name = string("transpose_7")];
            tensor<fp16, [1, 512, 150]> b_23_cast_fp16 = slice_by_index(begin = b_23_begin_0, end = b_23_end_0, end_mask = b_23_end_mask_0, x = bcx_cast_fp16)[name = string("b_23_cast_fp16")];
            tensor<int32, [3]> c_begin_0 = const()[name = string("c_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_end_0 = const()[name = string("c_end_0"), val = tensor<int32, [3]>([1, 1024, 150])];
            tensor<bool, [3]> c_end_mask_0 = const()[name = string("c_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 150]> c_cast_fp16 = slice_by_index(begin = c_begin_0, end = c_end_0, end_mask = c_end_mask_0, x = bcx_cast_fp16)[name = string("c_cast_fp16")];
            tensor<int32, [3]> xx_begin_0 = const()[name = string("xx_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_end_0 = const()[name = string("xx_end_0"), val = tensor<int32, [3]>([1, 1536, 150])];
            tensor<bool, [3]> xx_end_mask_0 = const()[name = string("xx_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 150]> xx_cast_fp16 = slice_by_index(begin = xx_begin_0, end = xx_end_0, end_mask = xx_end_mask_0, x = bcx_cast_fp16)[name = string("xx_cast_fp16")];
            tensor<fp16, [1, 512, 150]> new_cast_fp16 = mul(x = b_23_cast_fp16, y = xx_cast_fp16)[name = string("new_cast_fp16")];
            tensor<int32, [4]> input_101_begin_0 = const()[name = string("input_101_begin_0"), val = tensor<int32, [4]>([4, 0, 0, 0])];
            tensor<int32, [4]> input_101_end_0 = const()[name = string("input_101_end_0"), val = tensor<int32, [4]>([5, 1, 512, 6])];
            tensor<bool, [4]> input_101_end_mask_0 = const()[name = string("input_101_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_101_squeeze_mask_0 = const()[name = string("input_101_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_101_cast_fp16 = slice_by_index(begin = input_101_begin_0, end = input_101_end_0, end_mask = input_101_end_mask_0, squeeze_mask = input_101_squeeze_mask_0, x = conv)[name = string("input_101_cast_fp16")];
            tensor<int32, [6]> var_1523_pad_0 = const()[name = string("op_1523_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_1523_mode_0 = const()[name = string("op_1523_mode_0"), val = string("constant")];
            fp16 const_31_to_fp16 = const()[name = string("const_31_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 156]> var_1523_cast_fp16 = pad(constant_val = const_31_to_fp16, mode = var_1523_mode_0, pad = var_1523_pad_0, x = new_cast_fp16)[name = string("op_1523_cast_fp16")];
            tensor<int32, [6]> var_1525_pad_0 = const()[name = string("op_1525_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 150])];
            string var_1525_mode_0 = const()[name = string("op_1525_mode_0"), val = string("constant")];
            fp16 const_32_to_fp16 = const()[name = string("const_32_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 156]> var_1525_cast_fp16 = pad(constant_val = const_32_to_fp16, mode = var_1525_mode_0, pad = var_1525_pad_0, x = input_101_cast_fp16)[name = string("op_1525_cast_fp16")];
            tensor<fp16, [1, 512, 156]> win_17_cast_fp16 = add(x = var_1523_cast_fp16, y = var_1525_cast_fp16)[name = string("win_17_cast_fp16")];
            tensor<int32, [3]> var_1527_begin_0 = const()[name = string("op_1527_begin_0"), val = tensor<int32, [3]>([0, 0, 150])];
            tensor<int32, [3]> var_1527_end_0 = const()[name = string("op_1527_end_0"), val = tensor<int32, [3]>([1, 512, 156])];
            tensor<bool, [3]> var_1527_end_mask_0 = const()[name = string("op_1527_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_1527_cast_fp16 = slice_by_index(begin = var_1527_begin_0, end = var_1527_end_0, end_mask = var_1527_end_mask_0, x = win_17_cast_fp16)[name = string("op_1527_cast_fp16")];
            string var_1533_pad_type_0 = const()[name = string("op_1533_pad_type_0"), val = string("valid")];
            int32 var_1533_groups_0 = const()[name = string("op_1533_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_1533_strides_0 = const()[name = string("op_1533_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1533_pad_0 = const()[name = string("op_1533_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1533_dilations_0 = const()[name = string("op_1533_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_7_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63598272)))];
            tensor<fp16, [1, 512, 150]> var_1533_cast_fp16 = conv(dilations = var_1533_dilations_0, groups = var_1533_groups_0, pad = var_1533_pad_0, pad_type = var_1533_pad_type_0, strides = var_1533_strides_0, weight = core_backbone_layers_7_conv_conv_weight_to_fp16, x = win_17_cast_fp16)[name = string("op_1533_cast_fp16")];
            tensor<fp16, [1, 512, 150]> var_1535_cast_fp16 = mul(x = c_cast_fp16, y = var_1533_cast_fp16)[name = string("op_1535_cast_fp16")];
            tensor<int32, [3]> input_103_perm_0 = const()[name = string("input_103_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_7_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63605504)))];
            tensor<fp16, [1, 150, 512]> input_103_cast_fp16 = transpose(perm = input_103_perm_0, x = var_1535_cast_fp16)[name = string("transpose_6")];
            tensor<fp16, [1, 150, 512]> linear_46_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_7_conv_out_proj_weight_to_fp16, x = input_103_cast_fp16)[name = string("linear_46_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1540_cast_fp16 = abs(x = linear_46_cast_fp16)[name = string("op_1540_cast_fp16")];
            tensor<int32, [1]> var_1542_axes_0 = const()[name = string("op_1542_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1542_cast_fp16 = reduce_max(axes = var_1542_axes_0, keep_dims = var_33, x = var_1540_cast_fp16)[name = string("op_1542_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_77_cast_fp16 = maximum(x = var_1542_cast_fp16, y = var_34_to_fp16)[name = string("s_77_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_81_cast_fp16 = real_div(x = linear_46_cast_fp16, y = s_77_cast_fp16)[name = string("y_81_cast_fp16")];
            fp16 var_35_promoted_35_to_fp16 = const()[name = string("op_35_promoted_35_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1545_cast_fp16 = pow(x = y_81_cast_fp16, y = var_35_promoted_35_to_fp16)[name = string("op_1545_cast_fp16")];
            tensor<int32, [1]> var_1547_axes_0 = const()[name = string("op_1547_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1547_keep_dims_0 = const()[name = string("op_1547_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1547_cast_fp16 = reduce_mean(axes = var_1547_axes_0, keep_dims = var_1547_keep_dims_0, x = var_1545_cast_fp16)[name = string("op_1547_cast_fp16")];
            fp32 var_1548_epsilon_0 = const()[name = string("op_1548_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1548_cast_fp16 = inverse(epsilon = var_1548_epsilon_0, x = s_77_cast_fp16)[name = string("op_1548_cast_fp16")];
            fp16 var_1549_to_fp16 = const()[name = string("op_1549_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1550_cast_fp16 = mul(x = var_1548_cast_fp16, y = var_1549_to_fp16)[name = string("op_1550_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1551_cast_fp16 = real_div(x = var_1550_cast_fp16, y = s_77_cast_fp16)[name = string("op_1551_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1552_cast_fp16 = add(x = var_1547_cast_fp16, y = var_1551_cast_fp16)[name = string("op_1552_cast_fp16")];
            fp32 var_1553_epsilon_0 = const()[name = string("op_1553_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1553_cast_fp16 = rsqrt(epsilon = var_1553_epsilon_0, x = var_1552_cast_fp16)[name = string("op_1553_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1554_cast_fp16 = mul(x = y_81_cast_fp16, y = var_1553_cast_fp16)[name = string("op_1554_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_7_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64129856)))];
            tensor<fp16, [1, 150, 512]> a_cast_fp16 = mul(x = core_backbone_layers_7_post_operator_norm_weight_to_fp16, y = var_1554_cast_fp16)[name = string("a_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_123_cast_fp16 = add(x = x_119_cast_fp16, y = a_cast_fp16)[name = string("x_123_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1558_cast_fp16 = abs(x = x_123_cast_fp16)[name = string("op_1558_cast_fp16")];
            tensor<int32, [1]> var_1560_axes_0 = const()[name = string("op_1560_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1560_cast_fp16 = reduce_max(axes = var_1560_axes_0, keep_dims = var_33, x = var_1558_cast_fp16)[name = string("op_1560_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_79_cast_fp16 = maximum(x = var_1560_cast_fp16, y = var_34_to_fp16)[name = string("s_79_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_83_cast_fp16 = real_div(x = x_123_cast_fp16, y = s_79_cast_fp16)[name = string("y_83_cast_fp16")];
            fp16 var_35_promoted_36_to_fp16 = const()[name = string("op_35_promoted_36_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1563_cast_fp16 = pow(x = y_83_cast_fp16, y = var_35_promoted_36_to_fp16)[name = string("op_1563_cast_fp16")];
            tensor<int32, [1]> var_1565_axes_0 = const()[name = string("op_1565_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1565_keep_dims_0 = const()[name = string("op_1565_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1565_cast_fp16 = reduce_mean(axes = var_1565_axes_0, keep_dims = var_1565_keep_dims_0, x = var_1563_cast_fp16)[name = string("op_1565_cast_fp16")];
            fp32 var_1566_epsilon_0 = const()[name = string("op_1566_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1566_cast_fp16 = inverse(epsilon = var_1566_epsilon_0, x = s_79_cast_fp16)[name = string("op_1566_cast_fp16")];
            fp16 var_1567_to_fp16 = const()[name = string("op_1567_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1568_cast_fp16 = mul(x = var_1566_cast_fp16, y = var_1567_to_fp16)[name = string("op_1568_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1569_cast_fp16 = real_div(x = var_1568_cast_fp16, y = s_79_cast_fp16)[name = string("op_1569_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1570_cast_fp16 = add(x = var_1565_cast_fp16, y = var_1569_cast_fp16)[name = string("op_1570_cast_fp16")];
            fp32 var_1571_epsilon_0 = const()[name = string("op_1571_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1571_cast_fp16 = rsqrt(epsilon = var_1571_epsilon_0, x = var_1570_cast_fp16)[name = string("op_1571_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1572_cast_fp16 = mul(x = y_83_cast_fp16, y = var_1571_cast_fp16)[name = string("op_1572_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_7_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64130944)))];
            tensor<fp16, [1, 150, 512]> input_105_cast_fp16 = mul(x = core_backbone_layers_7_ffn_norm_weight_to_fp16, y = var_1572_cast_fp16)[name = string("input_105_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_7_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64132032)))];
            tensor<fp16, [1, 150, 2304]> linear_47_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w1_weight_to_fp16, x = input_105_cast_fp16)[name = string("linear_47_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_7_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66491392)))];
            tensor<fp16, [1, 150, 2304]> linear_48_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w3_weight_to_fp16, x = input_105_cast_fp16)[name = string("linear_48_cast_fp16")];
            fp16 var_1581_to_fp16 = const()[name = string("op_1581_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 150, 2304]> var_1582_cast_fp16 = mul(x = linear_47_cast_fp16, y = var_1581_to_fp16)[name = string("op_1582_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_1585_cast_fp16 = tanh(x = var_1582_cast_fp16)[name = string("op_1585_cast_fp16")];
            fp16 var_1586_to_fp16 = const()[name = string("op_1586_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 150, 2304]> var_1587_cast_fp16 = add(x = var_1585_cast_fp16, y = var_1586_to_fp16)[name = string("op_1587_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> var_1588_cast_fp16 = mul(x = var_1582_cast_fp16, y = var_1587_cast_fp16)[name = string("op_1588_cast_fp16")];
            tensor<fp16, [1, 150, 2304]> input_cast_fp16 = mul(x = var_1588_cast_fp16, y = linear_48_cast_fp16)[name = string("input_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_7_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68850752)))];
            tensor<fp16, [1, 150, 512]> linear_49_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w2_weight_to_fp16, x = input_cast_fp16)[name = string("linear_49_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1593_cast_fp16 = abs(x = linear_49_cast_fp16)[name = string("op_1593_cast_fp16")];
            tensor<int32, [1]> var_1595_axes_0 = const()[name = string("op_1595_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1595_cast_fp16 = reduce_max(axes = var_1595_axes_0, keep_dims = var_33, x = var_1593_cast_fp16)[name = string("op_1595_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_81_cast_fp16 = maximum(x = var_1595_cast_fp16, y = var_34_to_fp16)[name = string("s_81_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_85_cast_fp16 = real_div(x = linear_49_cast_fp16, y = s_81_cast_fp16)[name = string("y_85_cast_fp16")];
            fp16 var_35_promoted_37_to_fp16 = const()[name = string("op_35_promoted_37_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1598_cast_fp16 = pow(x = y_85_cast_fp16, y = var_35_promoted_37_to_fp16)[name = string("op_1598_cast_fp16")];
            tensor<int32, [1]> var_1600_axes_0 = const()[name = string("op_1600_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1600_keep_dims_0 = const()[name = string("op_1600_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1600_cast_fp16 = reduce_mean(axes = var_1600_axes_0, keep_dims = var_1600_keep_dims_0, x = var_1598_cast_fp16)[name = string("op_1600_cast_fp16")];
            fp32 var_1601_epsilon_0 = const()[name = string("op_1601_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1601_cast_fp16 = inverse(epsilon = var_1601_epsilon_0, x = s_81_cast_fp16)[name = string("op_1601_cast_fp16")];
            fp16 var_1602_to_fp16 = const()[name = string("op_1602_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1603_cast_fp16 = mul(x = var_1601_cast_fp16, y = var_1602_to_fp16)[name = string("op_1603_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1604_cast_fp16 = real_div(x = var_1603_cast_fp16, y = s_81_cast_fp16)[name = string("op_1604_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1605_cast_fp16 = add(x = var_1600_cast_fp16, y = var_1604_cast_fp16)[name = string("op_1605_cast_fp16")];
            fp32 var_1606_epsilon_0 = const()[name = string("op_1606_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1606_cast_fp16 = rsqrt(epsilon = var_1606_epsilon_0, x = var_1605_cast_fp16)[name = string("op_1606_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1607_cast_fp16 = mul(x = y_85_cast_fp16, y = var_1606_cast_fp16)[name = string("op_1607_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_7_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71210112)))];
            tensor<fp16, [1, 150, 512]> f_cast_fp16 = mul(x = core_backbone_layers_7_post_ffn_norm_weight_to_fp16, y = var_1607_cast_fp16)[name = string("f_cast_fp16")];
            tensor<fp16, [1, 150, 512]> x_cast_fp16 = add(x = x_123_cast_fp16, y = f_cast_fp16)[name = string("x_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1611_cast_fp16 = abs(x = x_cast_fp16)[name = string("op_1611_cast_fp16")];
            tensor<int32, [1]> var_1613_axes_0 = const()[name = string("op_1613_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 150, 1]> var_1613_cast_fp16 = reduce_max(axes = var_1613_axes_0, keep_dims = var_33, x = var_1611_cast_fp16)[name = string("op_1613_cast_fp16")];
            tensor<fp16, [1, 150, 1]> s_cast_fp16 = maximum(x = var_1613_cast_fp16, y = var_34_to_fp16)[name = string("s_cast_fp16")];
            tensor<fp16, [1, 150, 512]> y_cast_fp16 = real_div(x = x_cast_fp16, y = s_cast_fp16)[name = string("y_cast_fp16")];
            fp16 var_35_promoted_38_to_fp16 = const()[name = string("op_35_promoted_38_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 150, 512]> var_1616_cast_fp16 = pow(x = y_cast_fp16, y = var_35_promoted_38_to_fp16)[name = string("op_1616_cast_fp16")];
            tensor<int32, [1]> var_1618_axes_0 = const()[name = string("op_1618_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1618_keep_dims_0 = const()[name = string("op_1618_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 150, 1]> var_1618_cast_fp16 = reduce_mean(axes = var_1618_axes_0, keep_dims = var_1618_keep_dims_0, x = var_1616_cast_fp16)[name = string("op_1618_cast_fp16")];
            fp32 var_1619_epsilon_0 = const()[name = string("op_1619_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 150, 1]> var_1619_cast_fp16 = inverse(epsilon = var_1619_epsilon_0, x = s_cast_fp16)[name = string("op_1619_cast_fp16")];
            fp16 var_1620_to_fp16 = const()[name = string("op_1620_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 150, 1]> var_1621_cast_fp16 = mul(x = var_1619_cast_fp16, y = var_1620_to_fp16)[name = string("op_1621_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1622_cast_fp16 = real_div(x = var_1621_cast_fp16, y = s_cast_fp16)[name = string("op_1622_cast_fp16")];
            tensor<fp16, [1, 150, 1]> var_1623_cast_fp16 = add(x = var_1618_cast_fp16, y = var_1622_cast_fp16)[name = string("op_1623_cast_fp16")];
            fp32 var_1624_epsilon_0 = const()[name = string("op_1624_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 150, 1]> var_1624_cast_fp16 = rsqrt(epsilon = var_1624_epsilon_0, x = var_1623_cast_fp16)[name = string("op_1624_cast_fp16")];
            tensor<fp16, [1, 150, 512]> var_1625_cast_fp16 = mul(x = y_cast_fp16, y = var_1624_cast_fp16)[name = string("op_1625_cast_fp16")];
            tensor<fp16, [512]> core_backbone_embedding_norm_weight_to_fp16 = const()[name = string("core_backbone_embedding_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71211200)))];
            tensor<fp16, [1, 150, 512]> hidden = mul(x = core_backbone_embedding_norm_weight_to_fp16, y = var_1625_cast_fp16)[name = string("op_1626_cast_fp16")];
            int32 var_1628_axis_0 = const()[name = string("op_1628_axis_0"), val = int32(0)];
            tensor<fp16, [5, 1, 512, 6]> conv_out = stack(axis = var_1628_axis_0, values = (var_429_cast_fp16, var_546_cast_fp16, var_873_cast_fp16, var_1200_cast_fp16, var_1527_cast_fp16))[name = string("op_1628_cast_fp16")];
            int32 var_1630_axis_0 = const()[name = string("op_1630_axis_0"), val = int32(0)];
            tensor<fp16, [3, 1, 2, 29, 64]> k_out = stack(axis = var_1630_axis_0, values = (var_764_cast_fp16, var_1091_cast_fp16, var_1418_cast_fp16))[name = string("op_1630_cast_fp16")];
            int32 var_1632_axis_0 = const()[name = string("op_1632_axis_0"), val = int32(0)];
            tensor<fp16, [3, 1, 2, 29, 64]> v_out = stack(axis = var_1632_axis_0, values = (var_767_cast_fp16, var_1094_cast_fp16, var_1421_cast_fp16))[name = string("op_1632_cast_fp16")];
        } -> (hidden, conv_out, k_out, v_out);
    func f4<ios18>(tensor<fp16, [5, 1, 512, 6]> conv, tensor<fp16, [3, 1, 2, 29, 64]> k, tensor<fp16, [1, 4, 48]> lat48, tensor<fp16, [1, 1, 24, 53]> mask, tensor<fp16, [3, 1, 2, 29, 64]> v) {
            tensor<fp16, [512, 48]> proj_weight_to_fp16 = const()[name = string("proj_weight_to_fp16"), val = tensor<fp16, [512, 48]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            tensor<fp16, [512]> proj_bias_to_fp16 = const()[name = string("proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49280)))];
            tensor<fp16, [1, 4, 512]> linear_0_cast_fp16 = linear(bias = proj_bias_to_fp16, weight = proj_weight_to_fp16, x = lat48)[name = string("linear_0_cast_fp16")];
            bool x_1_transpose_x_0 = const()[name = string("x_1_transpose_x_0"), val = bool(false)];
            bool x_1_transpose_y_0 = const()[name = string("x_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [24, 4]> const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = tensor<fp16, [24, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71212288)))];
            tensor<fp16, [1, 24, 512]> x_1_cast_fp16 = matmul(transpose_x = x_1_transpose_x_0, transpose_y = x_1_transpose_y_0, x = const_0_to_fp16, y = linear_0_cast_fp16)[name = string("x_1_cast_fp16")];
            int32 var_32 = const()[name = string("op_32"), val = int32(-1)];
            bool var_33 = const()[name = string("op_33"), val = bool(true)];
            tensor<fp16, [1, 24, 512]> var_396_cast_fp16 = abs(x = x_1_cast_fp16)[name = string("op_396_cast_fp16")];
            tensor<int32, [1]> var_398_axes_0 = const()[name = string("op_398_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_398_cast_fp16 = reduce_max(axes = var_398_axes_0, keep_dims = var_33, x = var_396_cast_fp16)[name = string("op_398_cast_fp16")];
            fp16 var_34_to_fp16 = const()[name = string("op_34_to_fp16"), val = fp16(0x1.a38p-14)];
            tensor<fp16, [1, 24, 1]> s_1_cast_fp16 = maximum(x = var_398_cast_fp16, y = var_34_to_fp16)[name = string("s_1_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_1_cast_fp16 = real_div(x = x_1_cast_fp16, y = s_1_cast_fp16)[name = string("y_1_cast_fp16")];
            fp16 var_35_promoted_to_fp16 = const()[name = string("op_35_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_401_cast_fp16 = pow(x = y_1_cast_fp16, y = var_35_promoted_to_fp16)[name = string("op_401_cast_fp16")];
            tensor<int32, [1]> var_403_axes_0 = const()[name = string("op_403_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_403_keep_dims_0 = const()[name = string("op_403_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_403_cast_fp16 = reduce_mean(axes = var_403_axes_0, keep_dims = var_403_keep_dims_0, x = var_401_cast_fp16)[name = string("op_403_cast_fp16")];
            fp32 var_404_epsilon_0 = const()[name = string("op_404_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_404_cast_fp16 = inverse(epsilon = var_404_epsilon_0, x = s_1_cast_fp16)[name = string("op_404_cast_fp16")];
            fp16 var_405_to_fp16 = const()[name = string("op_405_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_406_cast_fp16 = mul(x = var_404_cast_fp16, y = var_405_to_fp16)[name = string("op_406_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_407_cast_fp16 = real_div(x = var_406_cast_fp16, y = s_1_cast_fp16)[name = string("op_407_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_408_cast_fp16 = add(x = var_403_cast_fp16, y = var_407_cast_fp16)[name = string("op_408_cast_fp16")];
            fp32 var_409_epsilon_0 = const()[name = string("op_409_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_409_cast_fp16 = rsqrt(epsilon = var_409_epsilon_0, x = var_408_cast_fp16)[name = string("op_409_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_410_cast_fp16 = mul(x = y_1_cast_fp16, y = var_409_cast_fp16)[name = string("op_410_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_0_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50368)))];
            tensor<fp16, [1, 24, 512]> input_1_cast_fp16 = mul(x = core_backbone_layers_0_operator_norm_weight_to_fp16, y = var_410_cast_fp16)[name = string("input_1_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_0_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51456)))];
            tensor<fp16, [1536]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1624384)))];
            tensor<fp16, [1, 24, 1536]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_0_conv_in_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<int32, [3]> bcx_1_perm_0 = const()[name = string("bcx_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_1_begin_0 = const()[name = string("b_1_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_1_end_0 = const()[name = string("b_1_end_0"), val = tensor<int32, [3]>([1, 512, 24])];
            tensor<bool, [3]> b_1_end_mask_0 = const()[name = string("b_1_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 24]> bcx_1_cast_fp16 = transpose(perm = bcx_1_perm_0, x = linear_1_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [1, 512, 24]> b_1_cast_fp16 = slice_by_index(begin = b_1_begin_0, end = b_1_end_0, end_mask = b_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("b_1_cast_fp16")];
            tensor<int32, [3]> c_1_begin_0 = const()[name = string("c_1_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_1_end_0 = const()[name = string("c_1_end_0"), val = tensor<int32, [3]>([1, 1024, 24])];
            tensor<bool, [3]> c_1_end_mask_0 = const()[name = string("c_1_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 24]> c_1_cast_fp16 = slice_by_index(begin = c_1_begin_0, end = c_1_end_0, end_mask = c_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("c_1_cast_fp16")];
            tensor<int32, [3]> xx_1_begin_0 = const()[name = string("xx_1_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_1_end_0 = const()[name = string("xx_1_end_0"), val = tensor<int32, [3]>([1, 1536, 24])];
            tensor<bool, [3]> xx_1_end_mask_0 = const()[name = string("xx_1_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 24]> xx_1_cast_fp16 = slice_by_index(begin = xx_1_begin_0, end = xx_1_end_0, end_mask = xx_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("xx_1_cast_fp16")];
            tensor<fp16, [1, 512, 24]> new_1_cast_fp16 = mul(x = b_1_cast_fp16, y = xx_1_cast_fp16)[name = string("new_1_cast_fp16")];
            tensor<int32, [4]> input_5_begin_0 = const()[name = string("input_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> input_5_end_0 = const()[name = string("input_5_end_0"), val = tensor<int32, [4]>([1, 1, 512, 6])];
            tensor<bool, [4]> input_5_end_mask_0 = const()[name = string("input_5_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_5_squeeze_mask_0 = const()[name = string("input_5_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_5_cast_fp16 = slice_by_index(begin = input_5_begin_0, end = input_5_end_0, end_mask = input_5_end_mask_0, squeeze_mask = input_5_squeeze_mask_0, x = conv)[name = string("input_5_cast_fp16")];
            tensor<int32, [6]> var_425_pad_0 = const()[name = string("op_425_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_425_mode_0 = const()[name = string("op_425_mode_0"), val = string("constant")];
            fp16 const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 30]> var_425_cast_fp16 = pad(constant_val = const_5_to_fp16, mode = var_425_mode_0, pad = var_425_pad_0, x = new_1_cast_fp16)[name = string("op_425_cast_fp16")];
            tensor<int32, [6]> var_427_pad_0 = const()[name = string("op_427_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 24])];
            string var_427_mode_0 = const()[name = string("op_427_mode_0"), val = string("constant")];
            fp16 const_6_to_fp16 = const()[name = string("const_6_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 30]> var_427_cast_fp16 = pad(constant_val = const_6_to_fp16, mode = var_427_mode_0, pad = var_427_pad_0, x = input_5_cast_fp16)[name = string("op_427_cast_fp16")];
            tensor<fp16, [1, 512, 30]> win_1_cast_fp16 = add(x = var_425_cast_fp16, y = var_427_cast_fp16)[name = string("win_1_cast_fp16")];
            tensor<int32, [3]> var_429_begin_0 = const()[name = string("op_429_begin_0"), val = tensor<int32, [3]>([0, 0, 24])];
            tensor<int32, [3]> var_429_end_0 = const()[name = string("op_429_end_0"), val = tensor<int32, [3]>([1, 512, 30])];
            tensor<bool, [3]> var_429_end_mask_0 = const()[name = string("op_429_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_429_cast_fp16 = slice_by_index(begin = var_429_begin_0, end = var_429_end_0, end_mask = var_429_end_mask_0, x = win_1_cast_fp16)[name = string("op_429_cast_fp16")];
            string var_435_pad_type_0 = const()[name = string("op_435_pad_type_0"), val = string("valid")];
            int32 var_435_groups_0 = const()[name = string("op_435_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_435_strides_0 = const()[name = string("op_435_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_435_pad_0 = const()[name = string("op_435_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_435_dilations_0 = const()[name = string("op_435_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_0_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1627520)))];
            tensor<fp16, [1, 512, 24]> var_435_cast_fp16 = conv(dilations = var_435_dilations_0, groups = var_435_groups_0, pad = var_435_pad_0, pad_type = var_435_pad_type_0, strides = var_435_strides_0, weight = core_backbone_layers_0_conv_conv_weight_to_fp16, x = win_1_cast_fp16)[name = string("op_435_cast_fp16")];
            tensor<fp16, [1, 512, 24]> var_437_cast_fp16 = mul(x = c_1_cast_fp16, y = var_435_cast_fp16)[name = string("op_437_cast_fp16")];
            tensor<int32, [3]> input_7_perm_0 = const()[name = string("input_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_0_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1634752)))];
            tensor<fp16, [512]> linear_2_bias_0_to_fp16 = const()[name = string("linear_2_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2159104)))];
            tensor<fp16, [1, 24, 512]> input_7_cast_fp16 = transpose(perm = input_7_perm_0, x = var_437_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [1, 24, 512]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_0_conv_out_proj_weight_to_fp16, x = input_7_cast_fp16)[name = string("linear_2_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_442_cast_fp16 = abs(x = linear_2_cast_fp16)[name = string("op_442_cast_fp16")];
            tensor<int32, [1]> var_444_axes_0 = const()[name = string("op_444_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_444_cast_fp16 = reduce_max(axes = var_444_axes_0, keep_dims = var_33, x = var_442_cast_fp16)[name = string("op_444_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_3_cast_fp16 = maximum(x = var_444_cast_fp16, y = var_34_to_fp16)[name = string("s_3_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_5_cast_fp16 = real_div(x = linear_2_cast_fp16, y = s_3_cast_fp16)[name = string("y_5_cast_fp16")];
            fp16 var_35_promoted_1_to_fp16 = const()[name = string("op_35_promoted_1_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_447_cast_fp16 = pow(x = y_5_cast_fp16, y = var_35_promoted_1_to_fp16)[name = string("op_447_cast_fp16")];
            tensor<int32, [1]> var_449_axes_0 = const()[name = string("op_449_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_449_keep_dims_0 = const()[name = string("op_449_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_449_cast_fp16 = reduce_mean(axes = var_449_axes_0, keep_dims = var_449_keep_dims_0, x = var_447_cast_fp16)[name = string("op_449_cast_fp16")];
            fp32 var_450_epsilon_0 = const()[name = string("op_450_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_450_cast_fp16 = inverse(epsilon = var_450_epsilon_0, x = s_3_cast_fp16)[name = string("op_450_cast_fp16")];
            fp16 var_451_to_fp16 = const()[name = string("op_451_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_452_cast_fp16 = mul(x = var_450_cast_fp16, y = var_451_to_fp16)[name = string("op_452_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_453_cast_fp16 = real_div(x = var_452_cast_fp16, y = s_3_cast_fp16)[name = string("op_453_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_454_cast_fp16 = add(x = var_449_cast_fp16, y = var_453_cast_fp16)[name = string("op_454_cast_fp16")];
            fp32 var_455_epsilon_0 = const()[name = string("op_455_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_455_cast_fp16 = rsqrt(epsilon = var_455_epsilon_0, x = var_454_cast_fp16)[name = string("op_455_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_456_cast_fp16 = mul(x = y_5_cast_fp16, y = var_455_cast_fp16)[name = string("op_456_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_0_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2160192)))];
            tensor<fp16, [1, 24, 512]> a_1_cast_fp16 = mul(x = core_backbone_layers_0_post_operator_norm_weight_to_fp16, y = var_456_cast_fp16)[name = string("a_1_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_5_cast_fp16 = add(x = x_1_cast_fp16, y = a_1_cast_fp16)[name = string("x_5_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_460_cast_fp16 = abs(x = x_5_cast_fp16)[name = string("op_460_cast_fp16")];
            tensor<int32, [1]> var_462_axes_0 = const()[name = string("op_462_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_462_cast_fp16 = reduce_max(axes = var_462_axes_0, keep_dims = var_33, x = var_460_cast_fp16)[name = string("op_462_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_5_cast_fp16 = maximum(x = var_462_cast_fp16, y = var_34_to_fp16)[name = string("s_5_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_7_cast_fp16 = real_div(x = x_5_cast_fp16, y = s_5_cast_fp16)[name = string("y_7_cast_fp16")];
            fp16 var_35_promoted_2_to_fp16 = const()[name = string("op_35_promoted_2_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_465_cast_fp16 = pow(x = y_7_cast_fp16, y = var_35_promoted_2_to_fp16)[name = string("op_465_cast_fp16")];
            tensor<int32, [1]> var_467_axes_0 = const()[name = string("op_467_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_467_keep_dims_0 = const()[name = string("op_467_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_467_cast_fp16 = reduce_mean(axes = var_467_axes_0, keep_dims = var_467_keep_dims_0, x = var_465_cast_fp16)[name = string("op_467_cast_fp16")];
            fp32 var_468_epsilon_0 = const()[name = string("op_468_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_468_cast_fp16 = inverse(epsilon = var_468_epsilon_0, x = s_5_cast_fp16)[name = string("op_468_cast_fp16")];
            fp16 var_469_to_fp16 = const()[name = string("op_469_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_470_cast_fp16 = mul(x = var_468_cast_fp16, y = var_469_to_fp16)[name = string("op_470_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_471_cast_fp16 = real_div(x = var_470_cast_fp16, y = s_5_cast_fp16)[name = string("op_471_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_472_cast_fp16 = add(x = var_467_cast_fp16, y = var_471_cast_fp16)[name = string("op_472_cast_fp16")];
            fp32 var_473_epsilon_0 = const()[name = string("op_473_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_473_cast_fp16 = rsqrt(epsilon = var_473_epsilon_0, x = var_472_cast_fp16)[name = string("op_473_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_474_cast_fp16 = mul(x = y_7_cast_fp16, y = var_473_cast_fp16)[name = string("op_474_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_0_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2161280)))];
            tensor<fp16, [1, 24, 512]> input_9_cast_fp16 = mul(x = core_backbone_layers_0_ffn_norm_weight_to_fp16, y = var_474_cast_fp16)[name = string("input_9_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_0_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2162368)))];
            tensor<fp16, [2304]> linear_3_bias_0_to_fp16 = const()[name = string("linear_3_bias_0_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4521728)))];
            tensor<fp16, [1, 24, 2304]> linear_3_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w1_weight_to_fp16, x = input_9_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_0_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4526400)))];
            tensor<fp16, [1, 24, 2304]> linear_4_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w3_weight_to_fp16, x = input_9_cast_fp16)[name = string("linear_4_cast_fp16")];
            fp16 var_483_to_fp16 = const()[name = string("op_483_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 24, 2304]> var_484_cast_fp16 = mul(x = linear_3_cast_fp16, y = var_483_to_fp16)[name = string("op_484_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_487_cast_fp16 = tanh(x = var_484_cast_fp16)[name = string("op_487_cast_fp16")];
            fp16 var_488_to_fp16 = const()[name = string("op_488_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 24, 2304]> var_489_cast_fp16 = add(x = var_487_cast_fp16, y = var_488_to_fp16)[name = string("op_489_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_490_cast_fp16 = mul(x = var_484_cast_fp16, y = var_489_cast_fp16)[name = string("op_490_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> input_11_cast_fp16 = mul(x = var_490_cast_fp16, y = linear_4_cast_fp16)[name = string("input_11_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_0_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6885760)))];
            tensor<fp16, [1, 24, 512]> linear_5_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w2_weight_to_fp16, x = input_11_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_495_cast_fp16 = abs(x = linear_5_cast_fp16)[name = string("op_495_cast_fp16")];
            tensor<int32, [1]> var_497_axes_0 = const()[name = string("op_497_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_497_cast_fp16 = reduce_max(axes = var_497_axes_0, keep_dims = var_33, x = var_495_cast_fp16)[name = string("op_497_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_7_cast_fp16 = maximum(x = var_497_cast_fp16, y = var_34_to_fp16)[name = string("s_7_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_9_cast_fp16 = real_div(x = linear_5_cast_fp16, y = s_7_cast_fp16)[name = string("y_9_cast_fp16")];
            fp16 var_35_promoted_3_to_fp16 = const()[name = string("op_35_promoted_3_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_500_cast_fp16 = pow(x = y_9_cast_fp16, y = var_35_promoted_3_to_fp16)[name = string("op_500_cast_fp16")];
            tensor<int32, [1]> var_502_axes_0 = const()[name = string("op_502_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_502_keep_dims_0 = const()[name = string("op_502_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_502_cast_fp16 = reduce_mean(axes = var_502_axes_0, keep_dims = var_502_keep_dims_0, x = var_500_cast_fp16)[name = string("op_502_cast_fp16")];
            fp32 var_503_epsilon_0 = const()[name = string("op_503_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_503_cast_fp16 = inverse(epsilon = var_503_epsilon_0, x = s_7_cast_fp16)[name = string("op_503_cast_fp16")];
            fp16 var_504_to_fp16 = const()[name = string("op_504_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_505_cast_fp16 = mul(x = var_503_cast_fp16, y = var_504_to_fp16)[name = string("op_505_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_506_cast_fp16 = real_div(x = var_505_cast_fp16, y = s_7_cast_fp16)[name = string("op_506_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_507_cast_fp16 = add(x = var_502_cast_fp16, y = var_506_cast_fp16)[name = string("op_507_cast_fp16")];
            fp32 var_508_epsilon_0 = const()[name = string("op_508_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_508_cast_fp16 = rsqrt(epsilon = var_508_epsilon_0, x = var_507_cast_fp16)[name = string("op_508_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_509_cast_fp16 = mul(x = y_9_cast_fp16, y = var_508_cast_fp16)[name = string("op_509_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_0_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9245120)))];
            tensor<fp16, [1, 24, 512]> f_1_cast_fp16 = mul(x = core_backbone_layers_0_post_ffn_norm_weight_to_fp16, y = var_509_cast_fp16)[name = string("f_1_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_11_cast_fp16 = add(x = x_5_cast_fp16, y = f_1_cast_fp16)[name = string("x_11_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_513_cast_fp16 = abs(x = x_11_cast_fp16)[name = string("op_513_cast_fp16")];
            tensor<int32, [1]> var_515_axes_0 = const()[name = string("op_515_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_515_cast_fp16 = reduce_max(axes = var_515_axes_0, keep_dims = var_33, x = var_513_cast_fp16)[name = string("op_515_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_9_cast_fp16 = maximum(x = var_515_cast_fp16, y = var_34_to_fp16)[name = string("s_9_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_11_cast_fp16 = real_div(x = x_11_cast_fp16, y = s_9_cast_fp16)[name = string("y_11_cast_fp16")];
            fp16 var_35_promoted_4_to_fp16 = const()[name = string("op_35_promoted_4_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_518_cast_fp16 = pow(x = y_11_cast_fp16, y = var_35_promoted_4_to_fp16)[name = string("op_518_cast_fp16")];
            tensor<int32, [1]> var_520_axes_0 = const()[name = string("op_520_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_520_keep_dims_0 = const()[name = string("op_520_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_520_cast_fp16 = reduce_mean(axes = var_520_axes_0, keep_dims = var_520_keep_dims_0, x = var_518_cast_fp16)[name = string("op_520_cast_fp16")];
            fp32 var_521_epsilon_0 = const()[name = string("op_521_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_521_cast_fp16 = inverse(epsilon = var_521_epsilon_0, x = s_9_cast_fp16)[name = string("op_521_cast_fp16")];
            fp16 var_522_to_fp16 = const()[name = string("op_522_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_523_cast_fp16 = mul(x = var_521_cast_fp16, y = var_522_to_fp16)[name = string("op_523_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_524_cast_fp16 = real_div(x = var_523_cast_fp16, y = s_9_cast_fp16)[name = string("op_524_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_525_cast_fp16 = add(x = var_520_cast_fp16, y = var_524_cast_fp16)[name = string("op_525_cast_fp16")];
            fp32 var_526_epsilon_0 = const()[name = string("op_526_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_526_cast_fp16 = rsqrt(epsilon = var_526_epsilon_0, x = var_525_cast_fp16)[name = string("op_526_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_527_cast_fp16 = mul(x = y_11_cast_fp16, y = var_526_cast_fp16)[name = string("op_527_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_1_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9246208)))];
            tensor<fp16, [1, 24, 512]> input_13_cast_fp16 = mul(x = core_backbone_layers_1_operator_norm_weight_to_fp16, y = var_527_cast_fp16)[name = string("input_13_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_1_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9247296)))];
            tensor<fp16, [1, 24, 1536]> linear_6_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_1_conv_in_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("linear_6_cast_fp16")];
            tensor<int32, [3]> bcx_3_perm_0 = const()[name = string("bcx_3_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_5_begin_0 = const()[name = string("b_5_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_5_end_0 = const()[name = string("b_5_end_0"), val = tensor<int32, [3]>([1, 512, 24])];
            tensor<bool, [3]> b_5_end_mask_0 = const()[name = string("b_5_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 24]> bcx_3_cast_fp16 = transpose(perm = bcx_3_perm_0, x = linear_6_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [1, 512, 24]> b_5_cast_fp16 = slice_by_index(begin = b_5_begin_0, end = b_5_end_0, end_mask = b_5_end_mask_0, x = bcx_3_cast_fp16)[name = string("b_5_cast_fp16")];
            tensor<int32, [3]> c_3_begin_0 = const()[name = string("c_3_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_3_end_0 = const()[name = string("c_3_end_0"), val = tensor<int32, [3]>([1, 1024, 24])];
            tensor<bool, [3]> c_3_end_mask_0 = const()[name = string("c_3_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 24]> c_3_cast_fp16 = slice_by_index(begin = c_3_begin_0, end = c_3_end_0, end_mask = c_3_end_mask_0, x = bcx_3_cast_fp16)[name = string("c_3_cast_fp16")];
            tensor<int32, [3]> xx_3_begin_0 = const()[name = string("xx_3_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_3_end_0 = const()[name = string("xx_3_end_0"), val = tensor<int32, [3]>([1, 1536, 24])];
            tensor<bool, [3]> xx_3_end_mask_0 = const()[name = string("xx_3_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 24]> xx_3_cast_fp16 = slice_by_index(begin = xx_3_begin_0, end = xx_3_end_0, end_mask = xx_3_end_mask_0, x = bcx_3_cast_fp16)[name = string("xx_3_cast_fp16")];
            tensor<fp16, [1, 512, 24]> new_3_cast_fp16 = mul(x = b_5_cast_fp16, y = xx_3_cast_fp16)[name = string("new_3_cast_fp16")];
            tensor<int32, [4]> input_17_begin_0 = const()[name = string("input_17_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])];
            tensor<int32, [4]> input_17_end_0 = const()[name = string("input_17_end_0"), val = tensor<int32, [4]>([2, 1, 512, 6])];
            tensor<bool, [4]> input_17_end_mask_0 = const()[name = string("input_17_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_17_squeeze_mask_0 = const()[name = string("input_17_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_17_cast_fp16 = slice_by_index(begin = input_17_begin_0, end = input_17_end_0, end_mask = input_17_end_mask_0, squeeze_mask = input_17_squeeze_mask_0, x = conv)[name = string("input_17_cast_fp16")];
            tensor<int32, [6]> var_542_pad_0 = const()[name = string("op_542_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_542_mode_0 = const()[name = string("op_542_mode_0"), val = string("constant")];
            fp16 const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 30]> var_542_cast_fp16 = pad(constant_val = const_7_to_fp16, mode = var_542_mode_0, pad = var_542_pad_0, x = new_3_cast_fp16)[name = string("op_542_cast_fp16")];
            tensor<int32, [6]> var_544_pad_0 = const()[name = string("op_544_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 24])];
            string var_544_mode_0 = const()[name = string("op_544_mode_0"), val = string("constant")];
            fp16 const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 30]> var_544_cast_fp16 = pad(constant_val = const_8_to_fp16, mode = var_544_mode_0, pad = var_544_pad_0, x = input_17_cast_fp16)[name = string("op_544_cast_fp16")];
            tensor<fp16, [1, 512, 30]> win_5_cast_fp16 = add(x = var_542_cast_fp16, y = var_544_cast_fp16)[name = string("win_5_cast_fp16")];
            tensor<int32, [3]> var_546_begin_0 = const()[name = string("op_546_begin_0"), val = tensor<int32, [3]>([0, 0, 24])];
            tensor<int32, [3]> var_546_end_0 = const()[name = string("op_546_end_0"), val = tensor<int32, [3]>([1, 512, 30])];
            tensor<bool, [3]> var_546_end_mask_0 = const()[name = string("op_546_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_546_cast_fp16 = slice_by_index(begin = var_546_begin_0, end = var_546_end_0, end_mask = var_546_end_mask_0, x = win_5_cast_fp16)[name = string("op_546_cast_fp16")];
            string var_552_pad_type_0 = const()[name = string("op_552_pad_type_0"), val = string("valid")];
            int32 var_552_groups_0 = const()[name = string("op_552_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_552_strides_0 = const()[name = string("op_552_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_552_pad_0 = const()[name = string("op_552_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_552_dilations_0 = const()[name = string("op_552_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_1_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10820224)))];
            tensor<fp16, [1, 512, 24]> var_552_cast_fp16 = conv(dilations = var_552_dilations_0, groups = var_552_groups_0, pad = var_552_pad_0, pad_type = var_552_pad_type_0, strides = var_552_strides_0, weight = core_backbone_layers_1_conv_conv_weight_to_fp16, x = win_5_cast_fp16)[name = string("op_552_cast_fp16")];
            tensor<fp16, [1, 512, 24]> var_554_cast_fp16 = mul(x = c_3_cast_fp16, y = var_552_cast_fp16)[name = string("op_554_cast_fp16")];
            tensor<int32, [3]> input_19_perm_0 = const()[name = string("input_19_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_1_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10827456)))];
            tensor<fp16, [1, 24, 512]> input_19_cast_fp16 = transpose(perm = input_19_perm_0, x = var_554_cast_fp16)[name = string("transpose_24")];
            tensor<fp16, [1, 24, 512]> linear_7_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_1_conv_out_proj_weight_to_fp16, x = input_19_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_559_cast_fp16 = abs(x = linear_7_cast_fp16)[name = string("op_559_cast_fp16")];
            tensor<int32, [1]> var_561_axes_0 = const()[name = string("op_561_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_561_cast_fp16 = reduce_max(axes = var_561_axes_0, keep_dims = var_33, x = var_559_cast_fp16)[name = string("op_561_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_11_cast_fp16 = maximum(x = var_561_cast_fp16, y = var_34_to_fp16)[name = string("s_11_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_15_cast_fp16 = real_div(x = linear_7_cast_fp16, y = s_11_cast_fp16)[name = string("y_15_cast_fp16")];
            fp16 var_35_promoted_5_to_fp16 = const()[name = string("op_35_promoted_5_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_564_cast_fp16 = pow(x = y_15_cast_fp16, y = var_35_promoted_5_to_fp16)[name = string("op_564_cast_fp16")];
            tensor<int32, [1]> var_566_axes_0 = const()[name = string("op_566_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_566_keep_dims_0 = const()[name = string("op_566_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_566_cast_fp16 = reduce_mean(axes = var_566_axes_0, keep_dims = var_566_keep_dims_0, x = var_564_cast_fp16)[name = string("op_566_cast_fp16")];
            fp32 var_567_epsilon_0 = const()[name = string("op_567_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_567_cast_fp16 = inverse(epsilon = var_567_epsilon_0, x = s_11_cast_fp16)[name = string("op_567_cast_fp16")];
            fp16 var_568_to_fp16 = const()[name = string("op_568_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_569_cast_fp16 = mul(x = var_567_cast_fp16, y = var_568_to_fp16)[name = string("op_569_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_570_cast_fp16 = real_div(x = var_569_cast_fp16, y = s_11_cast_fp16)[name = string("op_570_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_571_cast_fp16 = add(x = var_566_cast_fp16, y = var_570_cast_fp16)[name = string("op_571_cast_fp16")];
            fp32 var_572_epsilon_0 = const()[name = string("op_572_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_572_cast_fp16 = rsqrt(epsilon = var_572_epsilon_0, x = var_571_cast_fp16)[name = string("op_572_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_573_cast_fp16 = mul(x = y_15_cast_fp16, y = var_572_cast_fp16)[name = string("op_573_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_1_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11351808)))];
            tensor<fp16, [1, 24, 512]> a_3_cast_fp16 = mul(x = core_backbone_layers_1_post_operator_norm_weight_to_fp16, y = var_573_cast_fp16)[name = string("a_3_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_15_cast_fp16 = add(x = x_11_cast_fp16, y = a_3_cast_fp16)[name = string("x_15_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_577_cast_fp16 = abs(x = x_15_cast_fp16)[name = string("op_577_cast_fp16")];
            tensor<int32, [1]> var_579_axes_0 = const()[name = string("op_579_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_579_cast_fp16 = reduce_max(axes = var_579_axes_0, keep_dims = var_33, x = var_577_cast_fp16)[name = string("op_579_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_13_cast_fp16 = maximum(x = var_579_cast_fp16, y = var_34_to_fp16)[name = string("s_13_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_17_cast_fp16 = real_div(x = x_15_cast_fp16, y = s_13_cast_fp16)[name = string("y_17_cast_fp16")];
            fp16 var_35_promoted_6_to_fp16 = const()[name = string("op_35_promoted_6_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_582_cast_fp16 = pow(x = y_17_cast_fp16, y = var_35_promoted_6_to_fp16)[name = string("op_582_cast_fp16")];
            tensor<int32, [1]> var_584_axes_0 = const()[name = string("op_584_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_584_keep_dims_0 = const()[name = string("op_584_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_584_cast_fp16 = reduce_mean(axes = var_584_axes_0, keep_dims = var_584_keep_dims_0, x = var_582_cast_fp16)[name = string("op_584_cast_fp16")];
            fp32 var_585_epsilon_0 = const()[name = string("op_585_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_585_cast_fp16 = inverse(epsilon = var_585_epsilon_0, x = s_13_cast_fp16)[name = string("op_585_cast_fp16")];
            fp16 var_586_to_fp16 = const()[name = string("op_586_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_587_cast_fp16 = mul(x = var_585_cast_fp16, y = var_586_to_fp16)[name = string("op_587_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_588_cast_fp16 = real_div(x = var_587_cast_fp16, y = s_13_cast_fp16)[name = string("op_588_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_589_cast_fp16 = add(x = var_584_cast_fp16, y = var_588_cast_fp16)[name = string("op_589_cast_fp16")];
            fp32 var_590_epsilon_0 = const()[name = string("op_590_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_590_cast_fp16 = rsqrt(epsilon = var_590_epsilon_0, x = var_589_cast_fp16)[name = string("op_590_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_591_cast_fp16 = mul(x = y_17_cast_fp16, y = var_590_cast_fp16)[name = string("op_591_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_1_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11352896)))];
            tensor<fp16, [1, 24, 512]> input_21_cast_fp16 = mul(x = core_backbone_layers_1_ffn_norm_weight_to_fp16, y = var_591_cast_fp16)[name = string("input_21_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_1_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11353984)))];
            tensor<fp16, [1, 24, 2304]> linear_8_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w1_weight_to_fp16, x = input_21_cast_fp16)[name = string("linear_8_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_1_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13713344)))];
            tensor<fp16, [1, 24, 2304]> linear_9_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w3_weight_to_fp16, x = input_21_cast_fp16)[name = string("linear_9_cast_fp16")];
            fp16 var_600_to_fp16 = const()[name = string("op_600_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 24, 2304]> var_601_cast_fp16 = mul(x = linear_8_cast_fp16, y = var_600_to_fp16)[name = string("op_601_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_604_cast_fp16 = tanh(x = var_601_cast_fp16)[name = string("op_604_cast_fp16")];
            fp16 var_605_to_fp16 = const()[name = string("op_605_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 24, 2304]> var_606_cast_fp16 = add(x = var_604_cast_fp16, y = var_605_to_fp16)[name = string("op_606_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_607_cast_fp16 = mul(x = var_601_cast_fp16, y = var_606_cast_fp16)[name = string("op_607_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> input_23_cast_fp16 = mul(x = var_607_cast_fp16, y = linear_9_cast_fp16)[name = string("input_23_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_1_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16072704)))];
            tensor<fp16, [1, 24, 512]> linear_10_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w2_weight_to_fp16, x = input_23_cast_fp16)[name = string("linear_10_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_612_cast_fp16 = abs(x = linear_10_cast_fp16)[name = string("op_612_cast_fp16")];
            tensor<int32, [1]> var_614_axes_0 = const()[name = string("op_614_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_614_cast_fp16 = reduce_max(axes = var_614_axes_0, keep_dims = var_33, x = var_612_cast_fp16)[name = string("op_614_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_15_cast_fp16 = maximum(x = var_614_cast_fp16, y = var_34_to_fp16)[name = string("s_15_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_19_cast_fp16 = real_div(x = linear_10_cast_fp16, y = s_15_cast_fp16)[name = string("y_19_cast_fp16")];
            fp16 var_35_promoted_7_to_fp16 = const()[name = string("op_35_promoted_7_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_617_cast_fp16 = pow(x = y_19_cast_fp16, y = var_35_promoted_7_to_fp16)[name = string("op_617_cast_fp16")];
            tensor<int32, [1]> var_619_axes_0 = const()[name = string("op_619_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_619_keep_dims_0 = const()[name = string("op_619_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_619_cast_fp16 = reduce_mean(axes = var_619_axes_0, keep_dims = var_619_keep_dims_0, x = var_617_cast_fp16)[name = string("op_619_cast_fp16")];
            fp32 var_620_epsilon_0 = const()[name = string("op_620_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_620_cast_fp16 = inverse(epsilon = var_620_epsilon_0, x = s_15_cast_fp16)[name = string("op_620_cast_fp16")];
            fp16 var_621_to_fp16 = const()[name = string("op_621_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_622_cast_fp16 = mul(x = var_620_cast_fp16, y = var_621_to_fp16)[name = string("op_622_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_623_cast_fp16 = real_div(x = var_622_cast_fp16, y = s_15_cast_fp16)[name = string("op_623_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_624_cast_fp16 = add(x = var_619_cast_fp16, y = var_623_cast_fp16)[name = string("op_624_cast_fp16")];
            fp32 var_625_epsilon_0 = const()[name = string("op_625_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_625_cast_fp16 = rsqrt(epsilon = var_625_epsilon_0, x = var_624_cast_fp16)[name = string("op_625_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_626_cast_fp16 = mul(x = y_19_cast_fp16, y = var_625_cast_fp16)[name = string("op_626_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_1_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18432064)))];
            tensor<fp16, [1, 24, 512]> f_3_cast_fp16 = mul(x = core_backbone_layers_1_post_ffn_norm_weight_to_fp16, y = var_626_cast_fp16)[name = string("f_3_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_21_cast_fp16 = add(x = x_15_cast_fp16, y = f_3_cast_fp16)[name = string("x_21_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_630_cast_fp16 = abs(x = x_21_cast_fp16)[name = string("op_630_cast_fp16")];
            tensor<int32, [1]> var_632_axes_0 = const()[name = string("op_632_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_632_cast_fp16 = reduce_max(axes = var_632_axes_0, keep_dims = var_33, x = var_630_cast_fp16)[name = string("op_632_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_17_cast_fp16 = maximum(x = var_632_cast_fp16, y = var_34_to_fp16)[name = string("s_17_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_21_cast_fp16 = real_div(x = x_21_cast_fp16, y = s_17_cast_fp16)[name = string("y_21_cast_fp16")];
            fp16 var_35_promoted_8_to_fp16 = const()[name = string("op_35_promoted_8_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_635_cast_fp16 = pow(x = y_21_cast_fp16, y = var_35_promoted_8_to_fp16)[name = string("op_635_cast_fp16")];
            tensor<int32, [1]> var_637_axes_0 = const()[name = string("op_637_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_637_keep_dims_0 = const()[name = string("op_637_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_637_cast_fp16 = reduce_mean(axes = var_637_axes_0, keep_dims = var_637_keep_dims_0, x = var_635_cast_fp16)[name = string("op_637_cast_fp16")];
            fp32 var_638_epsilon_0 = const()[name = string("op_638_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_638_cast_fp16 = inverse(epsilon = var_638_epsilon_0, x = s_17_cast_fp16)[name = string("op_638_cast_fp16")];
            fp16 var_639_to_fp16 = const()[name = string("op_639_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_640_cast_fp16 = mul(x = var_638_cast_fp16, y = var_639_to_fp16)[name = string("op_640_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_641_cast_fp16 = real_div(x = var_640_cast_fp16, y = s_17_cast_fp16)[name = string("op_641_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_642_cast_fp16 = add(x = var_637_cast_fp16, y = var_641_cast_fp16)[name = string("op_642_cast_fp16")];
            fp32 var_643_epsilon_0 = const()[name = string("op_643_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_643_cast_fp16 = rsqrt(epsilon = var_643_epsilon_0, x = var_642_cast_fp16)[name = string("op_643_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_644_cast_fp16 = mul(x = y_21_cast_fp16, y = var_643_cast_fp16)[name = string("op_644_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_2_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18433152)))];
            tensor<fp16, [1, 24, 512]> input_25_cast_fp16 = mul(x = core_backbone_layers_2_operator_norm_weight_to_fp16, y = var_644_cast_fp16)[name = string("input_25_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18434240)))];
            tensor<fp16, [1, 24, 512]> linear_11_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_q_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_11_cast_fp16")];
            tensor<int32, [4]> concat_0 = const()[name = string("concat_0"), val = tensor<int32, [4]>([1, 24, 8, 64])];
            tensor<fp16, [1, 24, 8, 64]> x_23_cast_fp16 = reshape(shape = concat_0, x = linear_11_cast_fp16)[name = string("x_23_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> var_651_cast_fp16 = abs(x = x_23_cast_fp16)[name = string("op_651_cast_fp16")];
            tensor<int32, [1]> var_653_axes_0 = const()[name = string("op_653_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 8, 1]> var_653_cast_fp16 = reduce_max(axes = var_653_axes_0, keep_dims = var_33, x = var_651_cast_fp16)[name = string("op_653_cast_fp16")];
            tensor<fp16, [1, 24, 8, 1]> s_19_cast_fp16 = maximum(x = var_653_cast_fp16, y = var_34_to_fp16)[name = string("s_19_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> y_23_cast_fp16 = real_div(x = x_23_cast_fp16, y = s_19_cast_fp16)[name = string("y_23_cast_fp16")];
            fp16 var_35_promoted_9_to_fp16 = const()[name = string("op_35_promoted_9_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 8, 64]> var_656_cast_fp16 = pow(x = y_23_cast_fp16, y = var_35_promoted_9_to_fp16)[name = string("op_656_cast_fp16")];
            tensor<int32, [1]> var_658_axes_0 = const()[name = string("op_658_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_658_keep_dims_0 = const()[name = string("op_658_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 8, 1]> var_658_cast_fp16 = reduce_mean(axes = var_658_axes_0, keep_dims = var_658_keep_dims_0, x = var_656_cast_fp16)[name = string("op_658_cast_fp16")];
            fp32 var_659_epsilon_0 = const()[name = string("op_659_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 8, 1]> var_659_cast_fp16 = inverse(epsilon = var_659_epsilon_0, x = s_19_cast_fp16)[name = string("op_659_cast_fp16")];
            fp16 var_660_to_fp16 = const()[name = string("op_660_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 8, 1]> var_661_cast_fp16 = mul(x = var_659_cast_fp16, y = var_660_to_fp16)[name = string("op_661_cast_fp16")];
            tensor<fp16, [1, 24, 8, 1]> var_662_cast_fp16 = real_div(x = var_661_cast_fp16, y = s_19_cast_fp16)[name = string("op_662_cast_fp16")];
            tensor<fp16, [1, 24, 8, 1]> var_663_cast_fp16 = add(x = var_658_cast_fp16, y = var_662_cast_fp16)[name = string("op_663_cast_fp16")];
            fp32 var_664_epsilon_0 = const()[name = string("op_664_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 8, 1]> var_664_cast_fp16 = rsqrt(epsilon = var_664_epsilon_0, x = var_663_cast_fp16)[name = string("op_664_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> var_665_cast_fp16 = mul(x = y_23_cast_fp16, y = var_664_cast_fp16)[name = string("op_665_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_2_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18958592)))];
            tensor<fp16, [1, 24, 8, 64]> var_666_cast_fp16 = mul(x = core_backbone_layers_2_self_attn_q_norm_weight_to_fp16, y = var_665_cast_fp16)[name = string("op_666_cast_fp16")];
            tensor<fp16, [128, 512]> core_backbone_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18958784)))];
            tensor<fp16, [128]> linear_12_bias_0_to_fp16 = const()[name = string("linear_12_bias_0_to_fp16"), val = tensor<fp16, [128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19089920)))];
            tensor<fp16, [1, 24, 128]> linear_12_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_k_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_12_cast_fp16")];
            tensor<int32, [4]> concat_1 = const()[name = string("concat_1"), val = tensor<int32, [4]>([1, 24, 2, 64])];
            tensor<fp16, [1, 24, 2, 64]> x_25_cast_fp16 = reshape(shape = concat_1, x = linear_12_cast_fp16)[name = string("x_25_cast_fp16")];
            tensor<fp16, [1, 24, 2, 64]> var_673_cast_fp16 = abs(x = x_25_cast_fp16)[name = string("op_673_cast_fp16")];
            tensor<int32, [1]> var_675_axes_0 = const()[name = string("op_675_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 2, 1]> var_675_cast_fp16 = reduce_max(axes = var_675_axes_0, keep_dims = var_33, x = var_673_cast_fp16)[name = string("op_675_cast_fp16")];
            tensor<fp16, [1, 24, 2, 1]> s_21_cast_fp16 = maximum(x = var_675_cast_fp16, y = var_34_to_fp16)[name = string("s_21_cast_fp16")];
            tensor<fp16, [1, 24, 2, 64]> y_25_cast_fp16 = real_div(x = x_25_cast_fp16, y = s_21_cast_fp16)[name = string("y_25_cast_fp16")];
            fp16 var_35_promoted_10_to_fp16 = const()[name = string("op_35_promoted_10_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 2, 64]> var_678_cast_fp16 = pow(x = y_25_cast_fp16, y = var_35_promoted_10_to_fp16)[name = string("op_678_cast_fp16")];
            tensor<int32, [1]> var_680_axes_0 = const()[name = string("op_680_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_680_keep_dims_0 = const()[name = string("op_680_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 2, 1]> var_680_cast_fp16 = reduce_mean(axes = var_680_axes_0, keep_dims = var_680_keep_dims_0, x = var_678_cast_fp16)[name = string("op_680_cast_fp16")];
            fp32 var_681_epsilon_0 = const()[name = string("op_681_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 2, 1]> var_681_cast_fp16 = inverse(epsilon = var_681_epsilon_0, x = s_21_cast_fp16)[name = string("op_681_cast_fp16")];
            fp16 var_682_to_fp16 = const()[name = string("op_682_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 2, 1]> var_683_cast_fp16 = mul(x = var_681_cast_fp16, y = var_682_to_fp16)[name = string("op_683_cast_fp16")];
            tensor<fp16, [1, 24, 2, 1]> var_684_cast_fp16 = real_div(x = var_683_cast_fp16, y = s_21_cast_fp16)[name = string("op_684_cast_fp16")];
            tensor<fp16, [1, 24, 2, 1]> var_685_cast_fp16 = add(x = var_680_cast_fp16, y = var_684_cast_fp16)[name = string("op_685_cast_fp16")];
            fp32 var_686_epsilon_0 = const()[name = string("op_686_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 2, 1]> var_686_cast_fp16 = rsqrt(epsilon = var_686_epsilon_0, x = var_685_cast_fp16)[name = string("op_686_cast_fp16")];
            tensor<fp16, [1, 24, 2, 64]> var_687_cast_fp16 = mul(x = y_25_cast_fp16, y = var_686_cast_fp16)[name = string("op_687_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_2_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19090240)))];
            tensor<fp16, [1, 24, 2, 64]> var_688_cast_fp16 = mul(x = core_backbone_layers_2_self_attn_k_norm_weight_to_fp16, y = var_687_cast_fp16)[name = string("op_688_cast_fp16")];
            tensor<int32, [4]> new_5_perm_0 = const()[name = string("new_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [128, 512]> core_backbone_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19090432)))];
            tensor<fp16, [1, 24, 128]> linear_13_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_v_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_13_cast_fp16")];
            tensor<int32, [4]> concat_2 = const()[name = string("concat_2"), val = tensor<int32, [4]>([1, 24, 2, 64])];
            tensor<fp16, [1, 24, 2, 64]> var_693_cast_fp16 = reshape(shape = concat_2, x = linear_13_cast_fp16)[name = string("op_693_cast_fp16")];
            tensor<int32, [4]> new_7_perm_0 = const()[name = string("new_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_29_begin_0 = const()[name = string("input_29_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
            tensor<int32, [5]> input_29_end_0 = const()[name = string("input_29_end_0"), val = tensor<int32, [5]>([1, 1, 2, 29, 64])];
            tensor<bool, [5]> input_29_end_mask_0 = const()[name = string("input_29_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_29_squeeze_mask_0 = const()[name = string("input_29_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_29_cast_fp16 = slice_by_index(begin = input_29_begin_0, end = input_29_end_0, end_mask = input_29_end_mask_0, squeeze_mask = input_29_squeeze_mask_0, x = k)[name = string("input_29_cast_fp16")];
            tensor<int32, [8]> var_698_pad_0 = const()[name = string("op_698_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_698_mode_0 = const()[name = string("op_698_mode_0"), val = string("constant")];
            fp16 const_9_to_fp16 = const()[name = string("const_9_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 24, 64]> new_5_cast_fp16 = transpose(perm = new_5_perm_0, x = var_688_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [1, 2, 53, 64]> var_698_cast_fp16 = pad(constant_val = const_9_to_fp16, mode = var_698_mode_0, pad = var_698_pad_0, x = new_5_cast_fp16)[name = string("op_698_cast_fp16")];
            tensor<int32, [8]> var_700_pad_0 = const()[name = string("op_700_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 24, 0, 0])];
            string var_700_mode_0 = const()[name = string("op_700_mode_0"), val = string("constant")];
            fp16 const_10_to_fp16 = const()[name = string("const_10_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 53, 64]> var_700_cast_fp16 = pad(constant_val = const_10_to_fp16, mode = var_700_mode_0, pad = var_700_pad_0, x = input_29_cast_fp16)[name = string("op_700_cast_fp16")];
            tensor<fp16, [1, 2, 53, 64]> x_29_cast_fp16 = add(x = var_698_cast_fp16, y = var_700_cast_fp16)[name = string("x_29_cast_fp16")];
            tensor<int32, [5]> input_33_begin_0 = const()[name = string("input_33_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
            tensor<int32, [5]> input_33_end_0 = const()[name = string("input_33_end_0"), val = tensor<int32, [5]>([1, 1, 2, 29, 64])];
            tensor<bool, [5]> input_33_end_mask_0 = const()[name = string("input_33_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_33_squeeze_mask_0 = const()[name = string("input_33_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_33_cast_fp16 = slice_by_index(begin = input_33_begin_0, end = input_33_end_0, end_mask = input_33_end_mask_0, squeeze_mask = input_33_squeeze_mask_0, x = v)[name = string("input_33_cast_fp16")];
            tensor<int32, [8]> var_705_pad_0 = const()[name = string("op_705_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_705_mode_0 = const()[name = string("op_705_mode_0"), val = string("constant")];
            fp16 const_11_to_fp16 = const()[name = string("const_11_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 24, 64]> new_7_cast_fp16 = transpose(perm = new_7_perm_0, x = var_693_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [1, 2, 53, 64]> var_705_cast_fp16 = pad(constant_val = const_11_to_fp16, mode = var_705_mode_0, pad = var_705_pad_0, x = new_7_cast_fp16)[name = string("op_705_cast_fp16")];
            tensor<int32, [8]> var_707_pad_0 = const()[name = string("op_707_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 24, 0, 0])];
            string var_707_mode_0 = const()[name = string("op_707_mode_0"), val = string("constant")];
            fp16 const_12_to_fp16 = const()[name = string("const_12_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 53, 64]> var_707_cast_fp16 = pad(constant_val = const_12_to_fp16, mode = var_707_mode_0, pad = var_707_pad_0, x = input_33_cast_fp16)[name = string("op_707_cast_fp16")];
            tensor<fp16, [1, 2, 53, 64]> v_3_cast_fp16 = add(x = var_705_cast_fp16, y = var_707_cast_fp16)[name = string("v_3_cast_fp16")];
            tensor<fp16, [1, 24, 1, 64]> const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = tensor<fp16, [1, 24, 1, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71212544)))];
            tensor<fp16, [1, 24, 8, 64]> var_709_cast_fp16 = mul(x = var_666_cast_fp16, y = const_33_to_fp16)[name = string("op_709_cast_fp16")];
            tensor<int32, [2]> var_710_split_sizes_0 = const()[name = string("op_710_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 const_34 = const()[name = string("const_34"), val = int32(3)];
            tensor<fp16, [1, 24, 8, 32]> var_710_cast_fp16_0, tensor<fp16, [1, 24, 8, 32]> var_710_cast_fp16_1 = split(axis = const_34, split_sizes = var_710_split_sizes_0, x = var_666_cast_fp16)[name = string("op_710_cast_fp16")];
            fp16 const_13_promoted_to_fp16 = const()[name = string("const_13_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 24, 8, 32]> var_712_cast_fp16 = mul(x = var_710_cast_fp16_1, y = const_13_promoted_to_fp16)[name = string("op_712_cast_fp16")];
            bool var_714_interleave_0 = const()[name = string("op_714_interleave_0"), val = bool(false)];
            int32 const_35 = const()[name = string("const_35"), val = int32(3)];
            tensor<fp16, [1, 24, 8, 64]> var_714_cast_fp16 = concat(axis = const_35, interleave = var_714_interleave_0, values = (var_712_cast_fp16, var_710_cast_fp16_0))[name = string("op_714_cast_fp16")];
            tensor<fp16, [1, 24, 1, 64]> const_36_to_fp16 = const()[name = string("const_36_to_fp16"), val = tensor<fp16, [1, 24, 1, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71215680)))];
            tensor<fp16, [1, 24, 8, 64]> var_715_cast_fp16 = mul(x = var_714_cast_fp16, y = const_36_to_fp16)[name = string("op_715_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> q_1_cast_fp16 = add(x = var_709_cast_fp16, y = var_715_cast_fp16)[name = string("q_1_cast_fp16")];
            tensor<fp16, [1, 1, 53, 64]> const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = tensor<fp16, [1, 1, 53, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71218816)))];
            tensor<fp16, [1, 2, 53, 64]> var_717_cast_fp16 = mul(x = x_29_cast_fp16, y = const_1_to_fp16)[name = string("op_717_cast_fp16")];
            tensor<int32, [2]> var_718_split_sizes_0 = const()[name = string("op_718_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 var_718_axis_0 = const()[name = string("op_718_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 2, 53, 32]> var_718_cast_fp16_0, tensor<fp16, [1, 2, 53, 32]> var_718_cast_fp16_1 = split(axis = var_718_axis_0, split_sizes = var_718_split_sizes_0, x = x_29_cast_fp16)[name = string("op_718_cast_fp16")];
            fp16 const_14_promoted_to_fp16 = const()[name = string("const_14_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 2, 53, 32]> var_720_cast_fp16 = mul(x = var_718_cast_fp16_1, y = const_14_promoted_to_fp16)[name = string("op_720_cast_fp16")];
            bool var_722_interleave_0 = const()[name = string("op_722_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 2, 53, 64]> var_722_cast_fp16 = concat(axis = var_32, interleave = var_722_interleave_0, values = (var_720_cast_fp16, var_718_cast_fp16_0))[name = string("op_722_cast_fp16")];
            tensor<fp16, [1, 1, 53, 64]> const_2_to_fp16 = const()[name = string("const_2_to_fp16"), val = tensor<fp16, [1, 1, 53, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71225664)))];
            tensor<fp16, [1, 2, 53, 64]> var_723_cast_fp16 = mul(x = var_722_cast_fp16, y = const_2_to_fp16)[name = string("op_723_cast_fp16")];
            tensor<fp16, [1, 2, 53, 64]> k_3_cast_fp16 = add(x = var_717_cast_fp16, y = var_723_cast_fp16)[name = string("k_3_cast_fp16")];
            tensor<int32, [1]> var_725_axes_0 = const()[name = string("op_725_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 53, 64]> var_725_cast_fp16 = expand_dims(axes = var_725_axes_0, x = k_3_cast_fp16)[name = string("op_725_cast_fp16")];
            tensor<int32, [5]> var_727_reps_0 = const()[name = string("op_727_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 53, 64]> var_727_cast_fp16 = tile(reps = var_727_reps_0, x = var_725_cast_fp16)[name = string("op_727_cast_fp16")];
            tensor<int32, [4]> concat_4 = const()[name = string("concat_4"), val = tensor<int32, [4]>([1, 8, 53, 64])];
            tensor<fp16, [1, 8, 53, 64]> k_5_cast_fp16 = reshape(shape = concat_4, x = var_727_cast_fp16)[name = string("k_5_cast_fp16")];
            tensor<int32, [1]> var_729_axes_0 = const()[name = string("op_729_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 53, 64]> var_729_cast_fp16 = expand_dims(axes = var_729_axes_0, x = v_3_cast_fp16)[name = string("op_729_cast_fp16")];
            tensor<int32, [5]> var_731_reps_0 = const()[name = string("op_731_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 53, 64]> var_731_cast_fp16 = tile(reps = var_731_reps_0, x = var_729_cast_fp16)[name = string("op_731_cast_fp16")];
            tensor<int32, [4]> concat_6 = const()[name = string("concat_6"), val = tensor<int32, [4]>([1, 8, 53, 64])];
            tensor<fp16, [1, 8, 53, 64]> v_5_cast_fp16 = reshape(shape = concat_6, x = var_731_cast_fp16)[name = string("v_5_cast_fp16")];
            tensor<int32, [4]> transpose_0_perm_0 = const()[name = string("transpose_0_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_734_transpose_x_1 = const()[name = string("op_734_transpose_x_1"), val = bool(false)];
            bool var_734_transpose_y_1 = const()[name = string("op_734_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 24, 64]> transpose_0_cast_fp16 = transpose(perm = transpose_0_perm_0, x = q_1_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [1, 8, 24, 53]> var_734_cast_fp16 = matmul(transpose_x = var_734_transpose_x_1, transpose_y = var_734_transpose_y_1, x = transpose_0_cast_fp16, y = k_5_cast_fp16)[name = string("op_734_cast_fp16")];
            fp16 var_735_to_fp16 = const()[name = string("op_735_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 8, 24, 53]> s_23_cast_fp16 = mul(x = var_734_cast_fp16, y = var_735_to_fp16)[name = string("s_23_cast_fp16")];
            tensor<fp16, [1, 8, 24, 53]> x_31_cast_fp16 = add(x = s_23_cast_fp16, y = mask)[name = string("x_31_cast_fp16")];
            tensor<int32, [1]> var_739_axes_0 = const()[name = string("op_739_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 24, 1]> var_739_cast_fp16 = reduce_max(axes = var_739_axes_0, keep_dims = var_33, x = x_31_cast_fp16)[name = string("op_739_cast_fp16")];
            tensor<fp16, [1, 8, 24, 53]> x_33_cast_fp16 = sub(x = x_31_cast_fp16, y = var_739_cast_fp16)[name = string("x_33_cast_fp16")];
            tensor<fp16, [1, 8, 24, 53]> x_35_cast_fp16 = exp(x = x_33_cast_fp16)[name = string("x_35_cast_fp16")];
            tensor<int32, [1]> var_743_axes_0 = const()[name = string("op_743_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_743_keep_dims_0 = const()[name = string("op_743_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 24, 1]> var_743_cast_fp16 = reduce_sum(axes = var_743_axes_0, keep_dims = var_743_keep_dims_0, x = x_35_cast_fp16)[name = string("op_743_cast_fp16")];
            tensor<fp16, [1, 8, 24, 53]> var_744_cast_fp16 = real_div(x = x_35_cast_fp16, y = var_743_cast_fp16)[name = string("op_744_cast_fp16")];
            bool o_1_transpose_x_0 = const()[name = string("o_1_transpose_x_0"), val = bool(false)];
            bool o_1_transpose_y_0 = const()[name = string("o_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 24, 64]> o_1_cast_fp16 = matmul(transpose_x = o_1_transpose_x_0, transpose_y = o_1_transpose_y_0, x = var_744_cast_fp16, y = v_5_cast_fp16)[name = string("o_1_cast_fp16")];
            tensor<int32, [4]> o_3_perm_0 = const()[name = string("o_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [8, 512]> core_backbone_layers_2_self_attn_gate_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_gate_proj_weight_to_fp16"), val = tensor<fp16, [8, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19232320)))];
            tensor<fp16, [8]> linear_14_bias_0_to_fp16 = const()[name = string("linear_14_bias_0_to_fp16"), val = tensor<fp16, [8]>([0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0])];
            tensor<fp16, [1, 24, 8]> linear_14_cast_fp16 = linear(bias = linear_14_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_gate_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_14_cast_fp16")];
            fp16 var_750_to_fp16 = const()[name = string("op_750_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 24, 8]> var_751_cast_fp16 = mul(x = linear_14_cast_fp16, y = var_750_to_fp16)[name = string("op_751_cast_fp16")];
            tensor<fp16, [1, 24, 8]> var_752_cast_fp16 = tanh(x = var_751_cast_fp16)[name = string("op_752_cast_fp16")];
            fp16 var_753_to_fp16 = const()[name = string("op_753_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 24, 8]> var_754_cast_fp16 = add(x = var_752_cast_fp16, y = var_753_to_fp16)[name = string("op_754_cast_fp16")];
            fp16 var_755_to_fp16 = const()[name = string("op_755_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 24, 8]> var_756_cast_fp16 = mul(x = var_754_cast_fp16, y = var_755_to_fp16)[name = string("op_756_cast_fp16")];
            tensor<int32, [1]> var_757_axes_0 = const()[name = string("op_757_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 8, 1]> var_757_cast_fp16 = expand_dims(axes = var_757_axes_0, x = var_756_cast_fp16)[name = string("op_757_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> o_3_cast_fp16 = transpose(perm = o_3_perm_0, x = o_1_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [1, 24, 8, 64]> o_5_cast_fp16 = mul(x = o_3_cast_fp16, y = var_757_cast_fp16)[name = string("o_5_cast_fp16")];
            tensor<int32, [3]> concat_7 = const()[name = string("concat_7"), val = tensor<int32, [3]>([1, 24, 512])];
            tensor<fp16, [1, 24, 512]> input_35_cast_fp16 = reshape(shape = concat_7, x = o_5_cast_fp16)[name = string("input_35_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19240576)))];
            tensor<fp16, [1, 24, 512]> linear_15_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_o_proj_weight_to_fp16, x = input_35_cast_fp16)[name = string("linear_15_cast_fp16")];
            tensor<int32, [4]> var_764_begin_0 = const()[name = string("op_764_begin_0"), val = tensor<int32, [4]>([0, 0, 24, 0])];
            tensor<int32, [4]> var_764_end_0 = const()[name = string("op_764_end_0"), val = tensor<int32, [4]>([1, 2, 53, 64])];
            tensor<bool, [4]> var_764_end_mask_0 = const()[name = string("op_764_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_764_cast_fp16 = slice_by_index(begin = var_764_begin_0, end = var_764_end_0, end_mask = var_764_end_mask_0, x = x_29_cast_fp16)[name = string("op_764_cast_fp16")];
            tensor<int32, [4]> var_767_begin_0 = const()[name = string("op_767_begin_0"), val = tensor<int32, [4]>([0, 0, 24, 0])];
            tensor<int32, [4]> var_767_end_0 = const()[name = string("op_767_end_0"), val = tensor<int32, [4]>([1, 2, 53, 64])];
            tensor<bool, [4]> var_767_end_mask_0 = const()[name = string("op_767_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_767_cast_fp16 = slice_by_index(begin = var_767_begin_0, end = var_767_end_0, end_mask = var_767_end_mask_0, x = v_3_cast_fp16)[name = string("op_767_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_769_cast_fp16 = abs(x = linear_15_cast_fp16)[name = string("op_769_cast_fp16")];
            tensor<int32, [1]> var_771_axes_0 = const()[name = string("op_771_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_771_cast_fp16 = reduce_max(axes = var_771_axes_0, keep_dims = var_33, x = var_769_cast_fp16)[name = string("op_771_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_25_cast_fp16 = maximum(x = var_771_cast_fp16, y = var_34_to_fp16)[name = string("s_25_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_27_cast_fp16 = real_div(x = linear_15_cast_fp16, y = s_25_cast_fp16)[name = string("y_27_cast_fp16")];
            fp16 var_35_promoted_11_to_fp16 = const()[name = string("op_35_promoted_11_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_774_cast_fp16 = pow(x = y_27_cast_fp16, y = var_35_promoted_11_to_fp16)[name = string("op_774_cast_fp16")];
            tensor<int32, [1]> var_776_axes_0 = const()[name = string("op_776_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_776_keep_dims_0 = const()[name = string("op_776_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_776_cast_fp16 = reduce_mean(axes = var_776_axes_0, keep_dims = var_776_keep_dims_0, x = var_774_cast_fp16)[name = string("op_776_cast_fp16")];
            fp32 var_777_epsilon_0 = const()[name = string("op_777_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_777_cast_fp16 = inverse(epsilon = var_777_epsilon_0, x = s_25_cast_fp16)[name = string("op_777_cast_fp16")];
            fp16 var_778_to_fp16 = const()[name = string("op_778_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_779_cast_fp16 = mul(x = var_777_cast_fp16, y = var_778_to_fp16)[name = string("op_779_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_780_cast_fp16 = real_div(x = var_779_cast_fp16, y = s_25_cast_fp16)[name = string("op_780_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_781_cast_fp16 = add(x = var_776_cast_fp16, y = var_780_cast_fp16)[name = string("op_781_cast_fp16")];
            fp32 var_782_epsilon_0 = const()[name = string("op_782_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_782_cast_fp16 = rsqrt(epsilon = var_782_epsilon_0, x = var_781_cast_fp16)[name = string("op_782_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_783_cast_fp16 = mul(x = y_27_cast_fp16, y = var_782_cast_fp16)[name = string("op_783_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_2_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19764928)))];
            tensor<fp16, [1, 24, 512]> a_5_cast_fp16 = mul(x = core_backbone_layers_2_post_operator_norm_weight_to_fp16, y = var_783_cast_fp16)[name = string("a_5_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_41_cast_fp16 = add(x = x_21_cast_fp16, y = a_5_cast_fp16)[name = string("x_41_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_787_cast_fp16 = abs(x = x_41_cast_fp16)[name = string("op_787_cast_fp16")];
            tensor<int32, [1]> var_789_axes_0 = const()[name = string("op_789_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_789_cast_fp16 = reduce_max(axes = var_789_axes_0, keep_dims = var_33, x = var_787_cast_fp16)[name = string("op_789_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_27_cast_fp16 = maximum(x = var_789_cast_fp16, y = var_34_to_fp16)[name = string("s_27_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_29_cast_fp16 = real_div(x = x_41_cast_fp16, y = s_27_cast_fp16)[name = string("y_29_cast_fp16")];
            fp16 var_35_promoted_12_to_fp16 = const()[name = string("op_35_promoted_12_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_792_cast_fp16 = pow(x = y_29_cast_fp16, y = var_35_promoted_12_to_fp16)[name = string("op_792_cast_fp16")];
            tensor<int32, [1]> var_794_axes_0 = const()[name = string("op_794_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_794_keep_dims_0 = const()[name = string("op_794_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_794_cast_fp16 = reduce_mean(axes = var_794_axes_0, keep_dims = var_794_keep_dims_0, x = var_792_cast_fp16)[name = string("op_794_cast_fp16")];
            fp32 var_795_epsilon_0 = const()[name = string("op_795_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_795_cast_fp16 = inverse(epsilon = var_795_epsilon_0, x = s_27_cast_fp16)[name = string("op_795_cast_fp16")];
            fp16 var_796_to_fp16 = const()[name = string("op_796_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_797_cast_fp16 = mul(x = var_795_cast_fp16, y = var_796_to_fp16)[name = string("op_797_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_798_cast_fp16 = real_div(x = var_797_cast_fp16, y = s_27_cast_fp16)[name = string("op_798_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_799_cast_fp16 = add(x = var_794_cast_fp16, y = var_798_cast_fp16)[name = string("op_799_cast_fp16")];
            fp32 var_800_epsilon_0 = const()[name = string("op_800_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_800_cast_fp16 = rsqrt(epsilon = var_800_epsilon_0, x = var_799_cast_fp16)[name = string("op_800_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_801_cast_fp16 = mul(x = y_29_cast_fp16, y = var_800_cast_fp16)[name = string("op_801_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_2_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19766016)))];
            tensor<fp16, [1, 24, 512]> input_37_cast_fp16 = mul(x = core_backbone_layers_2_ffn_norm_weight_to_fp16, y = var_801_cast_fp16)[name = string("input_37_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_2_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19767104)))];
            tensor<fp16, [1, 24, 2304]> linear_16_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w1_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_16_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_2_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22126464)))];
            tensor<fp16, [1, 24, 2304]> linear_17_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w3_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_17_cast_fp16")];
            fp16 var_810_to_fp16 = const()[name = string("op_810_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 24, 2304]> var_811_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_810_to_fp16)[name = string("op_811_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_814_cast_fp16 = tanh(x = var_811_cast_fp16)[name = string("op_814_cast_fp16")];
            fp16 var_815_to_fp16 = const()[name = string("op_815_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 24, 2304]> var_816_cast_fp16 = add(x = var_814_cast_fp16, y = var_815_to_fp16)[name = string("op_816_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_817_cast_fp16 = mul(x = var_811_cast_fp16, y = var_816_cast_fp16)[name = string("op_817_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> input_39_cast_fp16 = mul(x = var_817_cast_fp16, y = linear_17_cast_fp16)[name = string("input_39_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_2_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24485824)))];
            tensor<fp16, [1, 24, 512]> linear_18_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w2_weight_to_fp16, x = input_39_cast_fp16)[name = string("linear_18_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_822_cast_fp16 = abs(x = linear_18_cast_fp16)[name = string("op_822_cast_fp16")];
            tensor<int32, [1]> var_824_axes_0 = const()[name = string("op_824_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_824_cast_fp16 = reduce_max(axes = var_824_axes_0, keep_dims = var_33, x = var_822_cast_fp16)[name = string("op_824_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_29_cast_fp16 = maximum(x = var_824_cast_fp16, y = var_34_to_fp16)[name = string("s_29_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_31_cast_fp16 = real_div(x = linear_18_cast_fp16, y = s_29_cast_fp16)[name = string("y_31_cast_fp16")];
            fp16 var_35_promoted_13_to_fp16 = const()[name = string("op_35_promoted_13_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_827_cast_fp16 = pow(x = y_31_cast_fp16, y = var_35_promoted_13_to_fp16)[name = string("op_827_cast_fp16")];
            tensor<int32, [1]> var_829_axes_0 = const()[name = string("op_829_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_829_keep_dims_0 = const()[name = string("op_829_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_829_cast_fp16 = reduce_mean(axes = var_829_axes_0, keep_dims = var_829_keep_dims_0, x = var_827_cast_fp16)[name = string("op_829_cast_fp16")];
            fp32 var_830_epsilon_0 = const()[name = string("op_830_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_830_cast_fp16 = inverse(epsilon = var_830_epsilon_0, x = s_29_cast_fp16)[name = string("op_830_cast_fp16")];
            fp16 var_831_to_fp16 = const()[name = string("op_831_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_832_cast_fp16 = mul(x = var_830_cast_fp16, y = var_831_to_fp16)[name = string("op_832_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_833_cast_fp16 = real_div(x = var_832_cast_fp16, y = s_29_cast_fp16)[name = string("op_833_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_834_cast_fp16 = add(x = var_829_cast_fp16, y = var_833_cast_fp16)[name = string("op_834_cast_fp16")];
            fp32 var_835_epsilon_0 = const()[name = string("op_835_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_835_cast_fp16 = rsqrt(epsilon = var_835_epsilon_0, x = var_834_cast_fp16)[name = string("op_835_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_836_cast_fp16 = mul(x = y_31_cast_fp16, y = var_835_cast_fp16)[name = string("op_836_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_2_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26845184)))];
            tensor<fp16, [1, 24, 512]> f_5_cast_fp16 = mul(x = core_backbone_layers_2_post_ffn_norm_weight_to_fp16, y = var_836_cast_fp16)[name = string("f_5_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_47_cast_fp16 = add(x = x_41_cast_fp16, y = f_5_cast_fp16)[name = string("x_47_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_840_cast_fp16 = abs(x = x_47_cast_fp16)[name = string("op_840_cast_fp16")];
            tensor<int32, [1]> var_842_axes_0 = const()[name = string("op_842_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_842_cast_fp16 = reduce_max(axes = var_842_axes_0, keep_dims = var_33, x = var_840_cast_fp16)[name = string("op_842_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_31_cast_fp16 = maximum(x = var_842_cast_fp16, y = var_34_to_fp16)[name = string("s_31_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_33_cast_fp16 = real_div(x = x_47_cast_fp16, y = s_31_cast_fp16)[name = string("y_33_cast_fp16")];
            fp16 var_35_promoted_14_to_fp16 = const()[name = string("op_35_promoted_14_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_845_cast_fp16 = pow(x = y_33_cast_fp16, y = var_35_promoted_14_to_fp16)[name = string("op_845_cast_fp16")];
            tensor<int32, [1]> var_847_axes_0 = const()[name = string("op_847_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_847_keep_dims_0 = const()[name = string("op_847_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_847_cast_fp16 = reduce_mean(axes = var_847_axes_0, keep_dims = var_847_keep_dims_0, x = var_845_cast_fp16)[name = string("op_847_cast_fp16")];
            fp32 var_848_epsilon_0 = const()[name = string("op_848_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_848_cast_fp16 = inverse(epsilon = var_848_epsilon_0, x = s_31_cast_fp16)[name = string("op_848_cast_fp16")];
            fp16 var_849_to_fp16 = const()[name = string("op_849_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_850_cast_fp16 = mul(x = var_848_cast_fp16, y = var_849_to_fp16)[name = string("op_850_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_851_cast_fp16 = real_div(x = var_850_cast_fp16, y = s_31_cast_fp16)[name = string("op_851_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_852_cast_fp16 = add(x = var_847_cast_fp16, y = var_851_cast_fp16)[name = string("op_852_cast_fp16")];
            fp32 var_853_epsilon_0 = const()[name = string("op_853_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_853_cast_fp16 = rsqrt(epsilon = var_853_epsilon_0, x = var_852_cast_fp16)[name = string("op_853_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_854_cast_fp16 = mul(x = y_33_cast_fp16, y = var_853_cast_fp16)[name = string("op_854_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_3_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26846272)))];
            tensor<fp16, [1, 24, 512]> input_41_cast_fp16 = mul(x = core_backbone_layers_3_operator_norm_weight_to_fp16, y = var_854_cast_fp16)[name = string("input_41_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_3_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26847360)))];
            tensor<fp16, [1, 24, 1536]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_3_conv_in_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("linear_19_cast_fp16")];
            tensor<int32, [3]> bcx_5_perm_0 = const()[name = string("bcx_5_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_11_begin_0 = const()[name = string("b_11_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_11_end_0 = const()[name = string("b_11_end_0"), val = tensor<int32, [3]>([1, 512, 24])];
            tensor<bool, [3]> b_11_end_mask_0 = const()[name = string("b_11_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 24]> bcx_5_cast_fp16 = transpose(perm = bcx_5_perm_0, x = linear_19_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [1, 512, 24]> b_11_cast_fp16 = slice_by_index(begin = b_11_begin_0, end = b_11_end_0, end_mask = b_11_end_mask_0, x = bcx_5_cast_fp16)[name = string("b_11_cast_fp16")];
            tensor<int32, [3]> c_5_begin_0 = const()[name = string("c_5_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_5_end_0 = const()[name = string("c_5_end_0"), val = tensor<int32, [3]>([1, 1024, 24])];
            tensor<bool, [3]> c_5_end_mask_0 = const()[name = string("c_5_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 24]> c_5_cast_fp16 = slice_by_index(begin = c_5_begin_0, end = c_5_end_0, end_mask = c_5_end_mask_0, x = bcx_5_cast_fp16)[name = string("c_5_cast_fp16")];
            tensor<int32, [3]> xx_5_begin_0 = const()[name = string("xx_5_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_5_end_0 = const()[name = string("xx_5_end_0"), val = tensor<int32, [3]>([1, 1536, 24])];
            tensor<bool, [3]> xx_5_end_mask_0 = const()[name = string("xx_5_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 24]> xx_5_cast_fp16 = slice_by_index(begin = xx_5_begin_0, end = xx_5_end_0, end_mask = xx_5_end_mask_0, x = bcx_5_cast_fp16)[name = string("xx_5_cast_fp16")];
            tensor<fp16, [1, 512, 24]> new_9_cast_fp16 = mul(x = b_11_cast_fp16, y = xx_5_cast_fp16)[name = string("new_9_cast_fp16")];
            tensor<int32, [4]> input_45_begin_0 = const()[name = string("input_45_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])];
            tensor<int32, [4]> input_45_end_0 = const()[name = string("input_45_end_0"), val = tensor<int32, [4]>([3, 1, 512, 6])];
            tensor<bool, [4]> input_45_end_mask_0 = const()[name = string("input_45_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_45_squeeze_mask_0 = const()[name = string("input_45_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_45_cast_fp16 = slice_by_index(begin = input_45_begin_0, end = input_45_end_0, end_mask = input_45_end_mask_0, squeeze_mask = input_45_squeeze_mask_0, x = conv)[name = string("input_45_cast_fp16")];
            tensor<int32, [6]> var_869_pad_0 = const()[name = string("op_869_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_869_mode_0 = const()[name = string("op_869_mode_0"), val = string("constant")];
            fp16 const_15_to_fp16 = const()[name = string("const_15_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 30]> var_869_cast_fp16 = pad(constant_val = const_15_to_fp16, mode = var_869_mode_0, pad = var_869_pad_0, x = new_9_cast_fp16)[name = string("op_869_cast_fp16")];
            tensor<int32, [6]> var_871_pad_0 = const()[name = string("op_871_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 24])];
            string var_871_mode_0 = const()[name = string("op_871_mode_0"), val = string("constant")];
            fp16 const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 30]> var_871_cast_fp16 = pad(constant_val = const_16_to_fp16, mode = var_871_mode_0, pad = var_871_pad_0, x = input_45_cast_fp16)[name = string("op_871_cast_fp16")];
            tensor<fp16, [1, 512, 30]> win_9_cast_fp16 = add(x = var_869_cast_fp16, y = var_871_cast_fp16)[name = string("win_9_cast_fp16")];
            tensor<int32, [3]> var_873_begin_0 = const()[name = string("op_873_begin_0"), val = tensor<int32, [3]>([0, 0, 24])];
            tensor<int32, [3]> var_873_end_0 = const()[name = string("op_873_end_0"), val = tensor<int32, [3]>([1, 512, 30])];
            tensor<bool, [3]> var_873_end_mask_0 = const()[name = string("op_873_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_873_cast_fp16 = slice_by_index(begin = var_873_begin_0, end = var_873_end_0, end_mask = var_873_end_mask_0, x = win_9_cast_fp16)[name = string("op_873_cast_fp16")];
            string var_879_pad_type_0 = const()[name = string("op_879_pad_type_0"), val = string("valid")];
            int32 var_879_groups_0 = const()[name = string("op_879_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_879_strides_0 = const()[name = string("op_879_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_879_pad_0 = const()[name = string("op_879_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_879_dilations_0 = const()[name = string("op_879_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_3_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28420288)))];
            tensor<fp16, [1, 512, 24]> var_879_cast_fp16 = conv(dilations = var_879_dilations_0, groups = var_879_groups_0, pad = var_879_pad_0, pad_type = var_879_pad_type_0, strides = var_879_strides_0, weight = core_backbone_layers_3_conv_conv_weight_to_fp16, x = win_9_cast_fp16)[name = string("op_879_cast_fp16")];
            tensor<fp16, [1, 512, 24]> var_881_cast_fp16 = mul(x = c_5_cast_fp16, y = var_879_cast_fp16)[name = string("op_881_cast_fp16")];
            tensor<int32, [3]> input_47_perm_0 = const()[name = string("input_47_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_3_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28427520)))];
            tensor<fp16, [1, 24, 512]> input_47_cast_fp16 = transpose(perm = input_47_perm_0, x = var_881_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [1, 24, 512]> linear_20_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_3_conv_out_proj_weight_to_fp16, x = input_47_cast_fp16)[name = string("linear_20_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_886_cast_fp16 = abs(x = linear_20_cast_fp16)[name = string("op_886_cast_fp16")];
            tensor<int32, [1]> var_888_axes_0 = const()[name = string("op_888_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_888_cast_fp16 = reduce_max(axes = var_888_axes_0, keep_dims = var_33, x = var_886_cast_fp16)[name = string("op_888_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_33_cast_fp16 = maximum(x = var_888_cast_fp16, y = var_34_to_fp16)[name = string("s_33_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_37_cast_fp16 = real_div(x = linear_20_cast_fp16, y = s_33_cast_fp16)[name = string("y_37_cast_fp16")];
            fp16 var_35_promoted_15_to_fp16 = const()[name = string("op_35_promoted_15_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_891_cast_fp16 = pow(x = y_37_cast_fp16, y = var_35_promoted_15_to_fp16)[name = string("op_891_cast_fp16")];
            tensor<int32, [1]> var_893_axes_0 = const()[name = string("op_893_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_893_keep_dims_0 = const()[name = string("op_893_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_893_cast_fp16 = reduce_mean(axes = var_893_axes_0, keep_dims = var_893_keep_dims_0, x = var_891_cast_fp16)[name = string("op_893_cast_fp16")];
            fp32 var_894_epsilon_0 = const()[name = string("op_894_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_894_cast_fp16 = inverse(epsilon = var_894_epsilon_0, x = s_33_cast_fp16)[name = string("op_894_cast_fp16")];
            fp16 var_895_to_fp16 = const()[name = string("op_895_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_896_cast_fp16 = mul(x = var_894_cast_fp16, y = var_895_to_fp16)[name = string("op_896_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_897_cast_fp16 = real_div(x = var_896_cast_fp16, y = s_33_cast_fp16)[name = string("op_897_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_898_cast_fp16 = add(x = var_893_cast_fp16, y = var_897_cast_fp16)[name = string("op_898_cast_fp16")];
            fp32 var_899_epsilon_0 = const()[name = string("op_899_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_899_cast_fp16 = rsqrt(epsilon = var_899_epsilon_0, x = var_898_cast_fp16)[name = string("op_899_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_900_cast_fp16 = mul(x = y_37_cast_fp16, y = var_899_cast_fp16)[name = string("op_900_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_3_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28951872)))];
            tensor<fp16, [1, 24, 512]> a_7_cast_fp16 = mul(x = core_backbone_layers_3_post_operator_norm_weight_to_fp16, y = var_900_cast_fp16)[name = string("a_7_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_51_cast_fp16 = add(x = x_47_cast_fp16, y = a_7_cast_fp16)[name = string("x_51_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_904_cast_fp16 = abs(x = x_51_cast_fp16)[name = string("op_904_cast_fp16")];
            tensor<int32, [1]> var_906_axes_0 = const()[name = string("op_906_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_906_cast_fp16 = reduce_max(axes = var_906_axes_0, keep_dims = var_33, x = var_904_cast_fp16)[name = string("op_906_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_35_cast_fp16 = maximum(x = var_906_cast_fp16, y = var_34_to_fp16)[name = string("s_35_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_39_cast_fp16 = real_div(x = x_51_cast_fp16, y = s_35_cast_fp16)[name = string("y_39_cast_fp16")];
            fp16 var_35_promoted_16_to_fp16 = const()[name = string("op_35_promoted_16_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_909_cast_fp16 = pow(x = y_39_cast_fp16, y = var_35_promoted_16_to_fp16)[name = string("op_909_cast_fp16")];
            tensor<int32, [1]> var_911_axes_0 = const()[name = string("op_911_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_911_keep_dims_0 = const()[name = string("op_911_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_911_cast_fp16 = reduce_mean(axes = var_911_axes_0, keep_dims = var_911_keep_dims_0, x = var_909_cast_fp16)[name = string("op_911_cast_fp16")];
            fp32 var_912_epsilon_0 = const()[name = string("op_912_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_912_cast_fp16 = inverse(epsilon = var_912_epsilon_0, x = s_35_cast_fp16)[name = string("op_912_cast_fp16")];
            fp16 var_913_to_fp16 = const()[name = string("op_913_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_914_cast_fp16 = mul(x = var_912_cast_fp16, y = var_913_to_fp16)[name = string("op_914_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_915_cast_fp16 = real_div(x = var_914_cast_fp16, y = s_35_cast_fp16)[name = string("op_915_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_916_cast_fp16 = add(x = var_911_cast_fp16, y = var_915_cast_fp16)[name = string("op_916_cast_fp16")];
            fp32 var_917_epsilon_0 = const()[name = string("op_917_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_917_cast_fp16 = rsqrt(epsilon = var_917_epsilon_0, x = var_916_cast_fp16)[name = string("op_917_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_918_cast_fp16 = mul(x = y_39_cast_fp16, y = var_917_cast_fp16)[name = string("op_918_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_3_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28952960)))];
            tensor<fp16, [1, 24, 512]> input_49_cast_fp16 = mul(x = core_backbone_layers_3_ffn_norm_weight_to_fp16, y = var_918_cast_fp16)[name = string("input_49_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_3_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28954048)))];
            tensor<fp16, [1, 24, 2304]> linear_21_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w1_weight_to_fp16, x = input_49_cast_fp16)[name = string("linear_21_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_3_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31313408)))];
            tensor<fp16, [1, 24, 2304]> linear_22_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w3_weight_to_fp16, x = input_49_cast_fp16)[name = string("linear_22_cast_fp16")];
            fp16 var_927_to_fp16 = const()[name = string("op_927_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 24, 2304]> var_928_cast_fp16 = mul(x = linear_21_cast_fp16, y = var_927_to_fp16)[name = string("op_928_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_931_cast_fp16 = tanh(x = var_928_cast_fp16)[name = string("op_931_cast_fp16")];
            fp16 var_932_to_fp16 = const()[name = string("op_932_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 24, 2304]> var_933_cast_fp16 = add(x = var_931_cast_fp16, y = var_932_to_fp16)[name = string("op_933_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_934_cast_fp16 = mul(x = var_928_cast_fp16, y = var_933_cast_fp16)[name = string("op_934_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> input_51_cast_fp16 = mul(x = var_934_cast_fp16, y = linear_22_cast_fp16)[name = string("input_51_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_3_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33672768)))];
            tensor<fp16, [1, 24, 512]> linear_23_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w2_weight_to_fp16, x = input_51_cast_fp16)[name = string("linear_23_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_939_cast_fp16 = abs(x = linear_23_cast_fp16)[name = string("op_939_cast_fp16")];
            tensor<int32, [1]> var_941_axes_0 = const()[name = string("op_941_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_941_cast_fp16 = reduce_max(axes = var_941_axes_0, keep_dims = var_33, x = var_939_cast_fp16)[name = string("op_941_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_37_cast_fp16 = maximum(x = var_941_cast_fp16, y = var_34_to_fp16)[name = string("s_37_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_41_cast_fp16 = real_div(x = linear_23_cast_fp16, y = s_37_cast_fp16)[name = string("y_41_cast_fp16")];
            fp16 var_35_promoted_17_to_fp16 = const()[name = string("op_35_promoted_17_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_944_cast_fp16 = pow(x = y_41_cast_fp16, y = var_35_promoted_17_to_fp16)[name = string("op_944_cast_fp16")];
            tensor<int32, [1]> var_946_axes_0 = const()[name = string("op_946_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_946_keep_dims_0 = const()[name = string("op_946_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_946_cast_fp16 = reduce_mean(axes = var_946_axes_0, keep_dims = var_946_keep_dims_0, x = var_944_cast_fp16)[name = string("op_946_cast_fp16")];
            fp32 var_947_epsilon_0 = const()[name = string("op_947_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_947_cast_fp16 = inverse(epsilon = var_947_epsilon_0, x = s_37_cast_fp16)[name = string("op_947_cast_fp16")];
            fp16 var_948_to_fp16 = const()[name = string("op_948_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_949_cast_fp16 = mul(x = var_947_cast_fp16, y = var_948_to_fp16)[name = string("op_949_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_950_cast_fp16 = real_div(x = var_949_cast_fp16, y = s_37_cast_fp16)[name = string("op_950_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_951_cast_fp16 = add(x = var_946_cast_fp16, y = var_950_cast_fp16)[name = string("op_951_cast_fp16")];
            fp32 var_952_epsilon_0 = const()[name = string("op_952_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_952_cast_fp16 = rsqrt(epsilon = var_952_epsilon_0, x = var_951_cast_fp16)[name = string("op_952_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_953_cast_fp16 = mul(x = y_41_cast_fp16, y = var_952_cast_fp16)[name = string("op_953_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_3_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36032128)))];
            tensor<fp16, [1, 24, 512]> f_7_cast_fp16 = mul(x = core_backbone_layers_3_post_ffn_norm_weight_to_fp16, y = var_953_cast_fp16)[name = string("f_7_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_57_cast_fp16 = add(x = x_51_cast_fp16, y = f_7_cast_fp16)[name = string("x_57_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_957_cast_fp16 = abs(x = x_57_cast_fp16)[name = string("op_957_cast_fp16")];
            tensor<int32, [1]> var_959_axes_0 = const()[name = string("op_959_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_959_cast_fp16 = reduce_max(axes = var_959_axes_0, keep_dims = var_33, x = var_957_cast_fp16)[name = string("op_959_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_39_cast_fp16 = maximum(x = var_959_cast_fp16, y = var_34_to_fp16)[name = string("s_39_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_43_cast_fp16 = real_div(x = x_57_cast_fp16, y = s_39_cast_fp16)[name = string("y_43_cast_fp16")];
            fp16 var_35_promoted_18_to_fp16 = const()[name = string("op_35_promoted_18_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_962_cast_fp16 = pow(x = y_43_cast_fp16, y = var_35_promoted_18_to_fp16)[name = string("op_962_cast_fp16")];
            tensor<int32, [1]> var_964_axes_0 = const()[name = string("op_964_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_964_keep_dims_0 = const()[name = string("op_964_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_964_cast_fp16 = reduce_mean(axes = var_964_axes_0, keep_dims = var_964_keep_dims_0, x = var_962_cast_fp16)[name = string("op_964_cast_fp16")];
            fp32 var_965_epsilon_0 = const()[name = string("op_965_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_965_cast_fp16 = inverse(epsilon = var_965_epsilon_0, x = s_39_cast_fp16)[name = string("op_965_cast_fp16")];
            fp16 var_966_to_fp16 = const()[name = string("op_966_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_967_cast_fp16 = mul(x = var_965_cast_fp16, y = var_966_to_fp16)[name = string("op_967_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_968_cast_fp16 = real_div(x = var_967_cast_fp16, y = s_39_cast_fp16)[name = string("op_968_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_969_cast_fp16 = add(x = var_964_cast_fp16, y = var_968_cast_fp16)[name = string("op_969_cast_fp16")];
            fp32 var_970_epsilon_0 = const()[name = string("op_970_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_970_cast_fp16 = rsqrt(epsilon = var_970_epsilon_0, x = var_969_cast_fp16)[name = string("op_970_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_971_cast_fp16 = mul(x = y_43_cast_fp16, y = var_970_cast_fp16)[name = string("op_971_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_4_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36033216)))];
            tensor<fp16, [1, 24, 512]> input_53_cast_fp16 = mul(x = core_backbone_layers_4_operator_norm_weight_to_fp16, y = var_971_cast_fp16)[name = string("input_53_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36034304)))];
            tensor<fp16, [1, 24, 512]> linear_24_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_q_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_24_cast_fp16")];
            tensor<int32, [4]> concat_8 = const()[name = string("concat_8"), val = tensor<int32, [4]>([1, 24, 8, 64])];
            tensor<fp16, [1, 24, 8, 64]> x_59_cast_fp16 = reshape(shape = concat_8, x = linear_24_cast_fp16)[name = string("x_59_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> var_978_cast_fp16 = abs(x = x_59_cast_fp16)[name = string("op_978_cast_fp16")];
            tensor<int32, [1]> var_980_axes_0 = const()[name = string("op_980_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 8, 1]> var_980_cast_fp16 = reduce_max(axes = var_980_axes_0, keep_dims = var_33, x = var_978_cast_fp16)[name = string("op_980_cast_fp16")];
            tensor<fp16, [1, 24, 8, 1]> s_41_cast_fp16 = maximum(x = var_980_cast_fp16, y = var_34_to_fp16)[name = string("s_41_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> y_45_cast_fp16 = real_div(x = x_59_cast_fp16, y = s_41_cast_fp16)[name = string("y_45_cast_fp16")];
            fp16 var_35_promoted_19_to_fp16 = const()[name = string("op_35_promoted_19_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 8, 64]> var_983_cast_fp16 = pow(x = y_45_cast_fp16, y = var_35_promoted_19_to_fp16)[name = string("op_983_cast_fp16")];
            tensor<int32, [1]> var_985_axes_0 = const()[name = string("op_985_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_985_keep_dims_0 = const()[name = string("op_985_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 8, 1]> var_985_cast_fp16 = reduce_mean(axes = var_985_axes_0, keep_dims = var_985_keep_dims_0, x = var_983_cast_fp16)[name = string("op_985_cast_fp16")];
            fp32 var_986_epsilon_0 = const()[name = string("op_986_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 8, 1]> var_986_cast_fp16 = inverse(epsilon = var_986_epsilon_0, x = s_41_cast_fp16)[name = string("op_986_cast_fp16")];
            fp16 var_987_to_fp16 = const()[name = string("op_987_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 8, 1]> var_988_cast_fp16 = mul(x = var_986_cast_fp16, y = var_987_to_fp16)[name = string("op_988_cast_fp16")];
            tensor<fp16, [1, 24, 8, 1]> var_989_cast_fp16 = real_div(x = var_988_cast_fp16, y = s_41_cast_fp16)[name = string("op_989_cast_fp16")];
            tensor<fp16, [1, 24, 8, 1]> var_990_cast_fp16 = add(x = var_985_cast_fp16, y = var_989_cast_fp16)[name = string("op_990_cast_fp16")];
            fp32 var_991_epsilon_0 = const()[name = string("op_991_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 8, 1]> var_991_cast_fp16 = rsqrt(epsilon = var_991_epsilon_0, x = var_990_cast_fp16)[name = string("op_991_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> var_992_cast_fp16 = mul(x = y_45_cast_fp16, y = var_991_cast_fp16)[name = string("op_992_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_4_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36558656)))];
            tensor<fp16, [1, 24, 8, 64]> var_993_cast_fp16 = mul(x = core_backbone_layers_4_self_attn_q_norm_weight_to_fp16, y = var_992_cast_fp16)[name = string("op_993_cast_fp16")];
            tensor<fp16, [128, 512]> core_backbone_layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36558848)))];
            tensor<fp16, [1, 24, 128]> linear_25_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_k_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_25_cast_fp16")];
            tensor<int32, [4]> concat_9 = const()[name = string("concat_9"), val = tensor<int32, [4]>([1, 24, 2, 64])];
            tensor<fp16, [1, 24, 2, 64]> x_61_cast_fp16 = reshape(shape = concat_9, x = linear_25_cast_fp16)[name = string("x_61_cast_fp16")];
            tensor<fp16, [1, 24, 2, 64]> var_1000_cast_fp16 = abs(x = x_61_cast_fp16)[name = string("op_1000_cast_fp16")];
            tensor<int32, [1]> var_1002_axes_0 = const()[name = string("op_1002_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 2, 1]> var_1002_cast_fp16 = reduce_max(axes = var_1002_axes_0, keep_dims = var_33, x = var_1000_cast_fp16)[name = string("op_1002_cast_fp16")];
            tensor<fp16, [1, 24, 2, 1]> s_43_cast_fp16 = maximum(x = var_1002_cast_fp16, y = var_34_to_fp16)[name = string("s_43_cast_fp16")];
            tensor<fp16, [1, 24, 2, 64]> y_47_cast_fp16 = real_div(x = x_61_cast_fp16, y = s_43_cast_fp16)[name = string("y_47_cast_fp16")];
            fp16 var_35_promoted_20_to_fp16 = const()[name = string("op_35_promoted_20_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 2, 64]> var_1005_cast_fp16 = pow(x = y_47_cast_fp16, y = var_35_promoted_20_to_fp16)[name = string("op_1005_cast_fp16")];
            tensor<int32, [1]> var_1007_axes_0 = const()[name = string("op_1007_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1007_keep_dims_0 = const()[name = string("op_1007_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 2, 1]> var_1007_cast_fp16 = reduce_mean(axes = var_1007_axes_0, keep_dims = var_1007_keep_dims_0, x = var_1005_cast_fp16)[name = string("op_1007_cast_fp16")];
            fp32 var_1008_epsilon_0 = const()[name = string("op_1008_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 2, 1]> var_1008_cast_fp16 = inverse(epsilon = var_1008_epsilon_0, x = s_43_cast_fp16)[name = string("op_1008_cast_fp16")];
            fp16 var_1009_to_fp16 = const()[name = string("op_1009_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 2, 1]> var_1010_cast_fp16 = mul(x = var_1008_cast_fp16, y = var_1009_to_fp16)[name = string("op_1010_cast_fp16")];
            tensor<fp16, [1, 24, 2, 1]> var_1011_cast_fp16 = real_div(x = var_1010_cast_fp16, y = s_43_cast_fp16)[name = string("op_1011_cast_fp16")];
            tensor<fp16, [1, 24, 2, 1]> var_1012_cast_fp16 = add(x = var_1007_cast_fp16, y = var_1011_cast_fp16)[name = string("op_1012_cast_fp16")];
            fp32 var_1013_epsilon_0 = const()[name = string("op_1013_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 2, 1]> var_1013_cast_fp16 = rsqrt(epsilon = var_1013_epsilon_0, x = var_1012_cast_fp16)[name = string("op_1013_cast_fp16")];
            tensor<fp16, [1, 24, 2, 64]> var_1014_cast_fp16 = mul(x = y_47_cast_fp16, y = var_1013_cast_fp16)[name = string("op_1014_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_4_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36689984)))];
            tensor<fp16, [1, 24, 2, 64]> var_1015_cast_fp16 = mul(x = core_backbone_layers_4_self_attn_k_norm_weight_to_fp16, y = var_1014_cast_fp16)[name = string("op_1015_cast_fp16")];
            tensor<int32, [4]> new_11_perm_0 = const()[name = string("new_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [128, 512]> core_backbone_layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36690176)))];
            tensor<fp16, [1, 24, 128]> linear_26_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_v_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_26_cast_fp16")];
            tensor<int32, [4]> concat_10 = const()[name = string("concat_10"), val = tensor<int32, [4]>([1, 24, 2, 64])];
            tensor<fp16, [1, 24, 2, 64]> var_1020_cast_fp16 = reshape(shape = concat_10, x = linear_26_cast_fp16)[name = string("op_1020_cast_fp16")];
            tensor<int32, [4]> new_13_perm_0 = const()[name = string("new_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_57_begin_0 = const()[name = string("input_57_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
            tensor<int32, [5]> input_57_end_0 = const()[name = string("input_57_end_0"), val = tensor<int32, [5]>([2, 1, 2, 29, 64])];
            tensor<bool, [5]> input_57_end_mask_0 = const()[name = string("input_57_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_57_squeeze_mask_0 = const()[name = string("input_57_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_57_cast_fp16 = slice_by_index(begin = input_57_begin_0, end = input_57_end_0, end_mask = input_57_end_mask_0, squeeze_mask = input_57_squeeze_mask_0, x = k)[name = string("input_57_cast_fp16")];
            tensor<int32, [8]> var_1025_pad_0 = const()[name = string("op_1025_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_1025_mode_0 = const()[name = string("op_1025_mode_0"), val = string("constant")];
            fp16 const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 24, 64]> new_11_cast_fp16 = transpose(perm = new_11_perm_0, x = var_1015_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [1, 2, 53, 64]> var_1025_cast_fp16 = pad(constant_val = const_17_to_fp16, mode = var_1025_mode_0, pad = var_1025_pad_0, x = new_11_cast_fp16)[name = string("op_1025_cast_fp16")];
            tensor<int32, [8]> var_1027_pad_0 = const()[name = string("op_1027_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 24, 0, 0])];
            string var_1027_mode_0 = const()[name = string("op_1027_mode_0"), val = string("constant")];
            fp16 const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 53, 64]> var_1027_cast_fp16 = pad(constant_val = const_18_to_fp16, mode = var_1027_mode_0, pad = var_1027_pad_0, x = input_57_cast_fp16)[name = string("op_1027_cast_fp16")];
            tensor<fp16, [1, 2, 53, 64]> x_65_cast_fp16 = add(x = var_1025_cast_fp16, y = var_1027_cast_fp16)[name = string("x_65_cast_fp16")];
            tensor<int32, [5]> input_61_begin_0 = const()[name = string("input_61_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
            tensor<int32, [5]> input_61_end_0 = const()[name = string("input_61_end_0"), val = tensor<int32, [5]>([2, 1, 2, 29, 64])];
            tensor<bool, [5]> input_61_end_mask_0 = const()[name = string("input_61_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_61_squeeze_mask_0 = const()[name = string("input_61_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_61_cast_fp16 = slice_by_index(begin = input_61_begin_0, end = input_61_end_0, end_mask = input_61_end_mask_0, squeeze_mask = input_61_squeeze_mask_0, x = v)[name = string("input_61_cast_fp16")];
            tensor<int32, [8]> var_1032_pad_0 = const()[name = string("op_1032_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_1032_mode_0 = const()[name = string("op_1032_mode_0"), val = string("constant")];
            fp16 const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 24, 64]> new_13_cast_fp16 = transpose(perm = new_13_perm_0, x = var_1020_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [1, 2, 53, 64]> var_1032_cast_fp16 = pad(constant_val = const_19_to_fp16, mode = var_1032_mode_0, pad = var_1032_pad_0, x = new_13_cast_fp16)[name = string("op_1032_cast_fp16")];
            tensor<int32, [8]> var_1034_pad_0 = const()[name = string("op_1034_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 24, 0, 0])];
            string var_1034_mode_0 = const()[name = string("op_1034_mode_0"), val = string("constant")];
            fp16 const_20_to_fp16 = const()[name = string("const_20_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 53, 64]> var_1034_cast_fp16 = pad(constant_val = const_20_to_fp16, mode = var_1034_mode_0, pad = var_1034_pad_0, x = input_61_cast_fp16)[name = string("op_1034_cast_fp16")];
            tensor<fp16, [1, 2, 53, 64]> v_7_cast_fp16 = add(x = var_1032_cast_fp16, y = var_1034_cast_fp16)[name = string("v_7_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> var_1036_cast_fp16 = mul(x = var_993_cast_fp16, y = const_33_to_fp16)[name = string("op_1036_cast_fp16")];
            tensor<int32, [2]> var_1037_split_sizes_0 = const()[name = string("op_1037_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 const_38 = const()[name = string("const_38"), val = int32(3)];
            tensor<fp16, [1, 24, 8, 32]> var_1037_cast_fp16_0, tensor<fp16, [1, 24, 8, 32]> var_1037_cast_fp16_1 = split(axis = const_38, split_sizes = var_1037_split_sizes_0, x = var_993_cast_fp16)[name = string("op_1037_cast_fp16")];
            fp16 const_21_promoted_to_fp16 = const()[name = string("const_21_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 24, 8, 32]> var_1039_cast_fp16 = mul(x = var_1037_cast_fp16_1, y = const_21_promoted_to_fp16)[name = string("op_1039_cast_fp16")];
            bool var_1041_interleave_0 = const()[name = string("op_1041_interleave_0"), val = bool(false)];
            int32 const_39 = const()[name = string("const_39"), val = int32(3)];
            tensor<fp16, [1, 24, 8, 64]> var_1041_cast_fp16 = concat(axis = const_39, interleave = var_1041_interleave_0, values = (var_1039_cast_fp16, var_1037_cast_fp16_0))[name = string("op_1041_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> var_1042_cast_fp16 = mul(x = var_1041_cast_fp16, y = const_36_to_fp16)[name = string("op_1042_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> q_3_cast_fp16 = add(x = var_1036_cast_fp16, y = var_1042_cast_fp16)[name = string("q_3_cast_fp16")];
            tensor<fp16, [1, 2, 53, 64]> var_1044_cast_fp16 = mul(x = x_65_cast_fp16, y = const_1_to_fp16)[name = string("op_1044_cast_fp16")];
            tensor<int32, [2]> var_1045_split_sizes_0 = const()[name = string("op_1045_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 var_1045_axis_0 = const()[name = string("op_1045_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 2, 53, 32]> var_1045_cast_fp16_0, tensor<fp16, [1, 2, 53, 32]> var_1045_cast_fp16_1 = split(axis = var_1045_axis_0, split_sizes = var_1045_split_sizes_0, x = x_65_cast_fp16)[name = string("op_1045_cast_fp16")];
            fp16 const_22_promoted_to_fp16 = const()[name = string("const_22_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 2, 53, 32]> var_1047_cast_fp16 = mul(x = var_1045_cast_fp16_1, y = const_22_promoted_to_fp16)[name = string("op_1047_cast_fp16")];
            bool var_1049_interleave_0 = const()[name = string("op_1049_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 2, 53, 64]> var_1049_cast_fp16 = concat(axis = var_32, interleave = var_1049_interleave_0, values = (var_1047_cast_fp16, var_1045_cast_fp16_0))[name = string("op_1049_cast_fp16")];
            tensor<fp16, [1, 2, 53, 64]> var_1050_cast_fp16 = mul(x = var_1049_cast_fp16, y = const_2_to_fp16)[name = string("op_1050_cast_fp16")];
            tensor<fp16, [1, 2, 53, 64]> k_7_cast_fp16 = add(x = var_1044_cast_fp16, y = var_1050_cast_fp16)[name = string("k_7_cast_fp16")];
            tensor<int32, [1]> var_1052_axes_0 = const()[name = string("op_1052_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 53, 64]> var_1052_cast_fp16 = expand_dims(axes = var_1052_axes_0, x = k_7_cast_fp16)[name = string("op_1052_cast_fp16")];
            tensor<int32, [5]> var_1054_reps_0 = const()[name = string("op_1054_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 53, 64]> var_1054_cast_fp16 = tile(reps = var_1054_reps_0, x = var_1052_cast_fp16)[name = string("op_1054_cast_fp16")];
            tensor<int32, [4]> concat_12 = const()[name = string("concat_12"), val = tensor<int32, [4]>([1, 8, 53, 64])];
            tensor<fp16, [1, 8, 53, 64]> k_9_cast_fp16 = reshape(shape = concat_12, x = var_1054_cast_fp16)[name = string("k_9_cast_fp16")];
            tensor<int32, [1]> var_1056_axes_0 = const()[name = string("op_1056_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 53, 64]> var_1056_cast_fp16 = expand_dims(axes = var_1056_axes_0, x = v_7_cast_fp16)[name = string("op_1056_cast_fp16")];
            tensor<int32, [5]> var_1058_reps_0 = const()[name = string("op_1058_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 53, 64]> var_1058_cast_fp16 = tile(reps = var_1058_reps_0, x = var_1056_cast_fp16)[name = string("op_1058_cast_fp16")];
            tensor<int32, [4]> concat_14 = const()[name = string("concat_14"), val = tensor<int32, [4]>([1, 8, 53, 64])];
            tensor<fp16, [1, 8, 53, 64]> v_9_cast_fp16 = reshape(shape = concat_14, x = var_1058_cast_fp16)[name = string("v_9_cast_fp16")];
            tensor<int32, [4]> transpose_2_perm_0 = const()[name = string("transpose_2_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1061_transpose_x_1 = const()[name = string("op_1061_transpose_x_1"), val = bool(false)];
            bool var_1061_transpose_y_1 = const()[name = string("op_1061_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 24, 64]> transpose_2_cast_fp16 = transpose(perm = transpose_2_perm_0, x = q_3_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [1, 8, 24, 53]> var_1061_cast_fp16 = matmul(transpose_x = var_1061_transpose_x_1, transpose_y = var_1061_transpose_y_1, x = transpose_2_cast_fp16, y = k_9_cast_fp16)[name = string("op_1061_cast_fp16")];
            fp16 var_1062_to_fp16 = const()[name = string("op_1062_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 8, 24, 53]> s_45_cast_fp16 = mul(x = var_1061_cast_fp16, y = var_1062_to_fp16)[name = string("s_45_cast_fp16")];
            tensor<fp16, [1, 8, 24, 53]> x_67_cast_fp16 = add(x = s_45_cast_fp16, y = mask)[name = string("x_67_cast_fp16")];
            tensor<int32, [1]> var_1066_axes_0 = const()[name = string("op_1066_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 24, 1]> var_1066_cast_fp16 = reduce_max(axes = var_1066_axes_0, keep_dims = var_33, x = x_67_cast_fp16)[name = string("op_1066_cast_fp16")];
            tensor<fp16, [1, 8, 24, 53]> x_69_cast_fp16 = sub(x = x_67_cast_fp16, y = var_1066_cast_fp16)[name = string("x_69_cast_fp16")];
            tensor<fp16, [1, 8, 24, 53]> x_71_cast_fp16 = exp(x = x_69_cast_fp16)[name = string("x_71_cast_fp16")];
            tensor<int32, [1]> var_1070_axes_0 = const()[name = string("op_1070_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1070_keep_dims_0 = const()[name = string("op_1070_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 24, 1]> var_1070_cast_fp16 = reduce_sum(axes = var_1070_axes_0, keep_dims = var_1070_keep_dims_0, x = x_71_cast_fp16)[name = string("op_1070_cast_fp16")];
            tensor<fp16, [1, 8, 24, 53]> var_1071_cast_fp16 = real_div(x = x_71_cast_fp16, y = var_1070_cast_fp16)[name = string("op_1071_cast_fp16")];
            bool o_7_transpose_x_0 = const()[name = string("o_7_transpose_x_0"), val = bool(false)];
            bool o_7_transpose_y_0 = const()[name = string("o_7_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 24, 64]> o_7_cast_fp16 = matmul(transpose_x = o_7_transpose_x_0, transpose_y = o_7_transpose_y_0, x = var_1071_cast_fp16, y = v_9_cast_fp16)[name = string("o_7_cast_fp16")];
            tensor<int32, [4]> o_9_perm_0 = const()[name = string("o_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [8, 512]> core_backbone_layers_4_self_attn_gate_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_gate_proj_weight_to_fp16"), val = tensor<fp16, [8, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36821312)))];
            tensor<fp16, [8]> linear_27_bias_0_to_fp16 = const()[name = string("linear_27_bias_0_to_fp16"), val = tensor<fp16, [8]>([0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0])];
            tensor<fp16, [1, 24, 8]> linear_27_cast_fp16 = linear(bias = linear_27_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_gate_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_27_cast_fp16")];
            fp16 var_1077_to_fp16 = const()[name = string("op_1077_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 24, 8]> var_1078_cast_fp16 = mul(x = linear_27_cast_fp16, y = var_1077_to_fp16)[name = string("op_1078_cast_fp16")];
            tensor<fp16, [1, 24, 8]> var_1079_cast_fp16 = tanh(x = var_1078_cast_fp16)[name = string("op_1079_cast_fp16")];
            fp16 var_1080_to_fp16 = const()[name = string("op_1080_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 24, 8]> var_1081_cast_fp16 = add(x = var_1079_cast_fp16, y = var_1080_to_fp16)[name = string("op_1081_cast_fp16")];
            fp16 var_1082_to_fp16 = const()[name = string("op_1082_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 24, 8]> var_1083_cast_fp16 = mul(x = var_1081_cast_fp16, y = var_1082_to_fp16)[name = string("op_1083_cast_fp16")];
            tensor<int32, [1]> var_1084_axes_0 = const()[name = string("op_1084_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 8, 1]> var_1084_cast_fp16 = expand_dims(axes = var_1084_axes_0, x = var_1083_cast_fp16)[name = string("op_1084_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> o_9_cast_fp16 = transpose(perm = o_9_perm_0, x = o_7_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [1, 24, 8, 64]> o_11_cast_fp16 = mul(x = o_9_cast_fp16, y = var_1084_cast_fp16)[name = string("o_11_cast_fp16")];
            tensor<int32, [3]> concat_15 = const()[name = string("concat_15"), val = tensor<int32, [3]>([1, 24, 512])];
            tensor<fp16, [1, 24, 512]> input_63_cast_fp16 = reshape(shape = concat_15, x = o_11_cast_fp16)[name = string("input_63_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36829568)))];
            tensor<fp16, [1, 24, 512]> linear_28_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_self_attn_o_proj_weight_to_fp16, x = input_63_cast_fp16)[name = string("linear_28_cast_fp16")];
            tensor<int32, [4]> var_1091_begin_0 = const()[name = string("op_1091_begin_0"), val = tensor<int32, [4]>([0, 0, 24, 0])];
            tensor<int32, [4]> var_1091_end_0 = const()[name = string("op_1091_end_0"), val = tensor<int32, [4]>([1, 2, 53, 64])];
            tensor<bool, [4]> var_1091_end_mask_0 = const()[name = string("op_1091_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_1091_cast_fp16 = slice_by_index(begin = var_1091_begin_0, end = var_1091_end_0, end_mask = var_1091_end_mask_0, x = x_65_cast_fp16)[name = string("op_1091_cast_fp16")];
            tensor<int32, [4]> var_1094_begin_0 = const()[name = string("op_1094_begin_0"), val = tensor<int32, [4]>([0, 0, 24, 0])];
            tensor<int32, [4]> var_1094_end_0 = const()[name = string("op_1094_end_0"), val = tensor<int32, [4]>([1, 2, 53, 64])];
            tensor<bool, [4]> var_1094_end_mask_0 = const()[name = string("op_1094_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_1094_cast_fp16 = slice_by_index(begin = var_1094_begin_0, end = var_1094_end_0, end_mask = var_1094_end_mask_0, x = v_7_cast_fp16)[name = string("op_1094_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1096_cast_fp16 = abs(x = linear_28_cast_fp16)[name = string("op_1096_cast_fp16")];
            tensor<int32, [1]> var_1098_axes_0 = const()[name = string("op_1098_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1098_cast_fp16 = reduce_max(axes = var_1098_axes_0, keep_dims = var_33, x = var_1096_cast_fp16)[name = string("op_1098_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_47_cast_fp16 = maximum(x = var_1098_cast_fp16, y = var_34_to_fp16)[name = string("s_47_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_49_cast_fp16 = real_div(x = linear_28_cast_fp16, y = s_47_cast_fp16)[name = string("y_49_cast_fp16")];
            fp16 var_35_promoted_21_to_fp16 = const()[name = string("op_35_promoted_21_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1101_cast_fp16 = pow(x = y_49_cast_fp16, y = var_35_promoted_21_to_fp16)[name = string("op_1101_cast_fp16")];
            tensor<int32, [1]> var_1103_axes_0 = const()[name = string("op_1103_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1103_keep_dims_0 = const()[name = string("op_1103_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1103_cast_fp16 = reduce_mean(axes = var_1103_axes_0, keep_dims = var_1103_keep_dims_0, x = var_1101_cast_fp16)[name = string("op_1103_cast_fp16")];
            fp32 var_1104_epsilon_0 = const()[name = string("op_1104_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1104_cast_fp16 = inverse(epsilon = var_1104_epsilon_0, x = s_47_cast_fp16)[name = string("op_1104_cast_fp16")];
            fp16 var_1105_to_fp16 = const()[name = string("op_1105_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1106_cast_fp16 = mul(x = var_1104_cast_fp16, y = var_1105_to_fp16)[name = string("op_1106_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1107_cast_fp16 = real_div(x = var_1106_cast_fp16, y = s_47_cast_fp16)[name = string("op_1107_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1108_cast_fp16 = add(x = var_1103_cast_fp16, y = var_1107_cast_fp16)[name = string("op_1108_cast_fp16")];
            fp32 var_1109_epsilon_0 = const()[name = string("op_1109_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1109_cast_fp16 = rsqrt(epsilon = var_1109_epsilon_0, x = var_1108_cast_fp16)[name = string("op_1109_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1110_cast_fp16 = mul(x = y_49_cast_fp16, y = var_1109_cast_fp16)[name = string("op_1110_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_4_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37353920)))];
            tensor<fp16, [1, 24, 512]> a_9_cast_fp16 = mul(x = core_backbone_layers_4_post_operator_norm_weight_to_fp16, y = var_1110_cast_fp16)[name = string("a_9_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_77_cast_fp16 = add(x = x_57_cast_fp16, y = a_9_cast_fp16)[name = string("x_77_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1114_cast_fp16 = abs(x = x_77_cast_fp16)[name = string("op_1114_cast_fp16")];
            tensor<int32, [1]> var_1116_axes_0 = const()[name = string("op_1116_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1116_cast_fp16 = reduce_max(axes = var_1116_axes_0, keep_dims = var_33, x = var_1114_cast_fp16)[name = string("op_1116_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_49_cast_fp16 = maximum(x = var_1116_cast_fp16, y = var_34_to_fp16)[name = string("s_49_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_51_cast_fp16 = real_div(x = x_77_cast_fp16, y = s_49_cast_fp16)[name = string("y_51_cast_fp16")];
            fp16 var_35_promoted_22_to_fp16 = const()[name = string("op_35_promoted_22_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1119_cast_fp16 = pow(x = y_51_cast_fp16, y = var_35_promoted_22_to_fp16)[name = string("op_1119_cast_fp16")];
            tensor<int32, [1]> var_1121_axes_0 = const()[name = string("op_1121_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1121_keep_dims_0 = const()[name = string("op_1121_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1121_cast_fp16 = reduce_mean(axes = var_1121_axes_0, keep_dims = var_1121_keep_dims_0, x = var_1119_cast_fp16)[name = string("op_1121_cast_fp16")];
            fp32 var_1122_epsilon_0 = const()[name = string("op_1122_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1122_cast_fp16 = inverse(epsilon = var_1122_epsilon_0, x = s_49_cast_fp16)[name = string("op_1122_cast_fp16")];
            fp16 var_1123_to_fp16 = const()[name = string("op_1123_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1124_cast_fp16 = mul(x = var_1122_cast_fp16, y = var_1123_to_fp16)[name = string("op_1124_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1125_cast_fp16 = real_div(x = var_1124_cast_fp16, y = s_49_cast_fp16)[name = string("op_1125_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1126_cast_fp16 = add(x = var_1121_cast_fp16, y = var_1125_cast_fp16)[name = string("op_1126_cast_fp16")];
            fp32 var_1127_epsilon_0 = const()[name = string("op_1127_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1127_cast_fp16 = rsqrt(epsilon = var_1127_epsilon_0, x = var_1126_cast_fp16)[name = string("op_1127_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1128_cast_fp16 = mul(x = y_51_cast_fp16, y = var_1127_cast_fp16)[name = string("op_1128_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_4_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37355008)))];
            tensor<fp16, [1, 24, 512]> input_65_cast_fp16 = mul(x = core_backbone_layers_4_ffn_norm_weight_to_fp16, y = var_1128_cast_fp16)[name = string("input_65_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_4_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37356096)))];
            tensor<fp16, [1, 24, 2304]> linear_29_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w1_weight_to_fp16, x = input_65_cast_fp16)[name = string("linear_29_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_4_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39715456)))];
            tensor<fp16, [1, 24, 2304]> linear_30_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w3_weight_to_fp16, x = input_65_cast_fp16)[name = string("linear_30_cast_fp16")];
            fp16 var_1137_to_fp16 = const()[name = string("op_1137_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 24, 2304]> var_1138_cast_fp16 = mul(x = linear_29_cast_fp16, y = var_1137_to_fp16)[name = string("op_1138_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_1141_cast_fp16 = tanh(x = var_1138_cast_fp16)[name = string("op_1141_cast_fp16")];
            fp16 var_1142_to_fp16 = const()[name = string("op_1142_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 24, 2304]> var_1143_cast_fp16 = add(x = var_1141_cast_fp16, y = var_1142_to_fp16)[name = string("op_1143_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_1144_cast_fp16 = mul(x = var_1138_cast_fp16, y = var_1143_cast_fp16)[name = string("op_1144_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> input_67_cast_fp16 = mul(x = var_1144_cast_fp16, y = linear_30_cast_fp16)[name = string("input_67_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_4_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42074816)))];
            tensor<fp16, [1, 24, 512]> linear_31_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w2_weight_to_fp16, x = input_67_cast_fp16)[name = string("linear_31_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1149_cast_fp16 = abs(x = linear_31_cast_fp16)[name = string("op_1149_cast_fp16")];
            tensor<int32, [1]> var_1151_axes_0 = const()[name = string("op_1151_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1151_cast_fp16 = reduce_max(axes = var_1151_axes_0, keep_dims = var_33, x = var_1149_cast_fp16)[name = string("op_1151_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_51_cast_fp16 = maximum(x = var_1151_cast_fp16, y = var_34_to_fp16)[name = string("s_51_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_53_cast_fp16 = real_div(x = linear_31_cast_fp16, y = s_51_cast_fp16)[name = string("y_53_cast_fp16")];
            fp16 var_35_promoted_23_to_fp16 = const()[name = string("op_35_promoted_23_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1154_cast_fp16 = pow(x = y_53_cast_fp16, y = var_35_promoted_23_to_fp16)[name = string("op_1154_cast_fp16")];
            tensor<int32, [1]> var_1156_axes_0 = const()[name = string("op_1156_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1156_keep_dims_0 = const()[name = string("op_1156_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1156_cast_fp16 = reduce_mean(axes = var_1156_axes_0, keep_dims = var_1156_keep_dims_0, x = var_1154_cast_fp16)[name = string("op_1156_cast_fp16")];
            fp32 var_1157_epsilon_0 = const()[name = string("op_1157_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1157_cast_fp16 = inverse(epsilon = var_1157_epsilon_0, x = s_51_cast_fp16)[name = string("op_1157_cast_fp16")];
            fp16 var_1158_to_fp16 = const()[name = string("op_1158_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1159_cast_fp16 = mul(x = var_1157_cast_fp16, y = var_1158_to_fp16)[name = string("op_1159_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1160_cast_fp16 = real_div(x = var_1159_cast_fp16, y = s_51_cast_fp16)[name = string("op_1160_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1161_cast_fp16 = add(x = var_1156_cast_fp16, y = var_1160_cast_fp16)[name = string("op_1161_cast_fp16")];
            fp32 var_1162_epsilon_0 = const()[name = string("op_1162_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1162_cast_fp16 = rsqrt(epsilon = var_1162_epsilon_0, x = var_1161_cast_fp16)[name = string("op_1162_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1163_cast_fp16 = mul(x = y_53_cast_fp16, y = var_1162_cast_fp16)[name = string("op_1163_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_4_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44434176)))];
            tensor<fp16, [1, 24, 512]> f_9_cast_fp16 = mul(x = core_backbone_layers_4_post_ffn_norm_weight_to_fp16, y = var_1163_cast_fp16)[name = string("f_9_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_83_cast_fp16 = add(x = x_77_cast_fp16, y = f_9_cast_fp16)[name = string("x_83_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1167_cast_fp16 = abs(x = x_83_cast_fp16)[name = string("op_1167_cast_fp16")];
            tensor<int32, [1]> var_1169_axes_0 = const()[name = string("op_1169_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1169_cast_fp16 = reduce_max(axes = var_1169_axes_0, keep_dims = var_33, x = var_1167_cast_fp16)[name = string("op_1169_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_53_cast_fp16 = maximum(x = var_1169_cast_fp16, y = var_34_to_fp16)[name = string("s_53_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_55_cast_fp16 = real_div(x = x_83_cast_fp16, y = s_53_cast_fp16)[name = string("y_55_cast_fp16")];
            fp16 var_35_promoted_24_to_fp16 = const()[name = string("op_35_promoted_24_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1172_cast_fp16 = pow(x = y_55_cast_fp16, y = var_35_promoted_24_to_fp16)[name = string("op_1172_cast_fp16")];
            tensor<int32, [1]> var_1174_axes_0 = const()[name = string("op_1174_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1174_keep_dims_0 = const()[name = string("op_1174_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1174_cast_fp16 = reduce_mean(axes = var_1174_axes_0, keep_dims = var_1174_keep_dims_0, x = var_1172_cast_fp16)[name = string("op_1174_cast_fp16")];
            fp32 var_1175_epsilon_0 = const()[name = string("op_1175_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1175_cast_fp16 = inverse(epsilon = var_1175_epsilon_0, x = s_53_cast_fp16)[name = string("op_1175_cast_fp16")];
            fp16 var_1176_to_fp16 = const()[name = string("op_1176_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1177_cast_fp16 = mul(x = var_1175_cast_fp16, y = var_1176_to_fp16)[name = string("op_1177_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1178_cast_fp16 = real_div(x = var_1177_cast_fp16, y = s_53_cast_fp16)[name = string("op_1178_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1179_cast_fp16 = add(x = var_1174_cast_fp16, y = var_1178_cast_fp16)[name = string("op_1179_cast_fp16")];
            fp32 var_1180_epsilon_0 = const()[name = string("op_1180_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1180_cast_fp16 = rsqrt(epsilon = var_1180_epsilon_0, x = var_1179_cast_fp16)[name = string("op_1180_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1181_cast_fp16 = mul(x = y_55_cast_fp16, y = var_1180_cast_fp16)[name = string("op_1181_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_5_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44435264)))];
            tensor<fp16, [1, 24, 512]> input_69_cast_fp16 = mul(x = core_backbone_layers_5_operator_norm_weight_to_fp16, y = var_1181_cast_fp16)[name = string("input_69_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_5_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44436352)))];
            tensor<fp16, [1, 24, 1536]> linear_32_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_5_conv_in_proj_weight_to_fp16, x = input_69_cast_fp16)[name = string("linear_32_cast_fp16")];
            tensor<int32, [3]> bcx_7_perm_0 = const()[name = string("bcx_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_17_begin_0 = const()[name = string("b_17_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_17_end_0 = const()[name = string("b_17_end_0"), val = tensor<int32, [3]>([1, 512, 24])];
            tensor<bool, [3]> b_17_end_mask_0 = const()[name = string("b_17_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 24]> bcx_7_cast_fp16 = transpose(perm = bcx_7_perm_0, x = linear_32_cast_fp16)[name = string("transpose_13")];
            tensor<fp16, [1, 512, 24]> b_17_cast_fp16 = slice_by_index(begin = b_17_begin_0, end = b_17_end_0, end_mask = b_17_end_mask_0, x = bcx_7_cast_fp16)[name = string("b_17_cast_fp16")];
            tensor<int32, [3]> c_7_begin_0 = const()[name = string("c_7_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_7_end_0 = const()[name = string("c_7_end_0"), val = tensor<int32, [3]>([1, 1024, 24])];
            tensor<bool, [3]> c_7_end_mask_0 = const()[name = string("c_7_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 24]> c_7_cast_fp16 = slice_by_index(begin = c_7_begin_0, end = c_7_end_0, end_mask = c_7_end_mask_0, x = bcx_7_cast_fp16)[name = string("c_7_cast_fp16")];
            tensor<int32, [3]> xx_7_begin_0 = const()[name = string("xx_7_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_7_end_0 = const()[name = string("xx_7_end_0"), val = tensor<int32, [3]>([1, 1536, 24])];
            tensor<bool, [3]> xx_7_end_mask_0 = const()[name = string("xx_7_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 24]> xx_7_cast_fp16 = slice_by_index(begin = xx_7_begin_0, end = xx_7_end_0, end_mask = xx_7_end_mask_0, x = bcx_7_cast_fp16)[name = string("xx_7_cast_fp16")];
            tensor<fp16, [1, 512, 24]> new_15_cast_fp16 = mul(x = b_17_cast_fp16, y = xx_7_cast_fp16)[name = string("new_15_cast_fp16")];
            tensor<int32, [4]> input_73_begin_0 = const()[name = string("input_73_begin_0"), val = tensor<int32, [4]>([3, 0, 0, 0])];
            tensor<int32, [4]> input_73_end_0 = const()[name = string("input_73_end_0"), val = tensor<int32, [4]>([4, 1, 512, 6])];
            tensor<bool, [4]> input_73_end_mask_0 = const()[name = string("input_73_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_73_squeeze_mask_0 = const()[name = string("input_73_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_73_cast_fp16 = slice_by_index(begin = input_73_begin_0, end = input_73_end_0, end_mask = input_73_end_mask_0, squeeze_mask = input_73_squeeze_mask_0, x = conv)[name = string("input_73_cast_fp16")];
            tensor<int32, [6]> var_1196_pad_0 = const()[name = string("op_1196_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_1196_mode_0 = const()[name = string("op_1196_mode_0"), val = string("constant")];
            fp16 const_23_to_fp16 = const()[name = string("const_23_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 30]> var_1196_cast_fp16 = pad(constant_val = const_23_to_fp16, mode = var_1196_mode_0, pad = var_1196_pad_0, x = new_15_cast_fp16)[name = string("op_1196_cast_fp16")];
            tensor<int32, [6]> var_1198_pad_0 = const()[name = string("op_1198_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 24])];
            string var_1198_mode_0 = const()[name = string("op_1198_mode_0"), val = string("constant")];
            fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 30]> var_1198_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = var_1198_mode_0, pad = var_1198_pad_0, x = input_73_cast_fp16)[name = string("op_1198_cast_fp16")];
            tensor<fp16, [1, 512, 30]> win_13_cast_fp16 = add(x = var_1196_cast_fp16, y = var_1198_cast_fp16)[name = string("win_13_cast_fp16")];
            tensor<int32, [3]> var_1200_begin_0 = const()[name = string("op_1200_begin_0"), val = tensor<int32, [3]>([0, 0, 24])];
            tensor<int32, [3]> var_1200_end_0 = const()[name = string("op_1200_end_0"), val = tensor<int32, [3]>([1, 512, 30])];
            tensor<bool, [3]> var_1200_end_mask_0 = const()[name = string("op_1200_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_1200_cast_fp16 = slice_by_index(begin = var_1200_begin_0, end = var_1200_end_0, end_mask = var_1200_end_mask_0, x = win_13_cast_fp16)[name = string("op_1200_cast_fp16")];
            string var_1206_pad_type_0 = const()[name = string("op_1206_pad_type_0"), val = string("valid")];
            int32 var_1206_groups_0 = const()[name = string("op_1206_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_1206_strides_0 = const()[name = string("op_1206_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1206_pad_0 = const()[name = string("op_1206_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1206_dilations_0 = const()[name = string("op_1206_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_5_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_5_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46009280)))];
            tensor<fp16, [1, 512, 24]> var_1206_cast_fp16 = conv(dilations = var_1206_dilations_0, groups = var_1206_groups_0, pad = var_1206_pad_0, pad_type = var_1206_pad_type_0, strides = var_1206_strides_0, weight = core_backbone_layers_5_conv_conv_weight_to_fp16, x = win_13_cast_fp16)[name = string("op_1206_cast_fp16")];
            tensor<fp16, [1, 512, 24]> var_1208_cast_fp16 = mul(x = c_7_cast_fp16, y = var_1206_cast_fp16)[name = string("op_1208_cast_fp16")];
            tensor<int32, [3]> input_75_perm_0 = const()[name = string("input_75_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_5_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46016512)))];
            tensor<fp16, [1, 24, 512]> input_75_cast_fp16 = transpose(perm = input_75_perm_0, x = var_1208_cast_fp16)[name = string("transpose_12")];
            tensor<fp16, [1, 24, 512]> linear_33_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_5_conv_out_proj_weight_to_fp16, x = input_75_cast_fp16)[name = string("linear_33_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1213_cast_fp16 = abs(x = linear_33_cast_fp16)[name = string("op_1213_cast_fp16")];
            tensor<int32, [1]> var_1215_axes_0 = const()[name = string("op_1215_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1215_cast_fp16 = reduce_max(axes = var_1215_axes_0, keep_dims = var_33, x = var_1213_cast_fp16)[name = string("op_1215_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_55_cast_fp16 = maximum(x = var_1215_cast_fp16, y = var_34_to_fp16)[name = string("s_55_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_59_cast_fp16 = real_div(x = linear_33_cast_fp16, y = s_55_cast_fp16)[name = string("y_59_cast_fp16")];
            fp16 var_35_promoted_25_to_fp16 = const()[name = string("op_35_promoted_25_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1218_cast_fp16 = pow(x = y_59_cast_fp16, y = var_35_promoted_25_to_fp16)[name = string("op_1218_cast_fp16")];
            tensor<int32, [1]> var_1220_axes_0 = const()[name = string("op_1220_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1220_keep_dims_0 = const()[name = string("op_1220_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1220_cast_fp16 = reduce_mean(axes = var_1220_axes_0, keep_dims = var_1220_keep_dims_0, x = var_1218_cast_fp16)[name = string("op_1220_cast_fp16")];
            fp32 var_1221_epsilon_0 = const()[name = string("op_1221_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1221_cast_fp16 = inverse(epsilon = var_1221_epsilon_0, x = s_55_cast_fp16)[name = string("op_1221_cast_fp16")];
            fp16 var_1222_to_fp16 = const()[name = string("op_1222_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1223_cast_fp16 = mul(x = var_1221_cast_fp16, y = var_1222_to_fp16)[name = string("op_1223_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1224_cast_fp16 = real_div(x = var_1223_cast_fp16, y = s_55_cast_fp16)[name = string("op_1224_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1225_cast_fp16 = add(x = var_1220_cast_fp16, y = var_1224_cast_fp16)[name = string("op_1225_cast_fp16")];
            fp32 var_1226_epsilon_0 = const()[name = string("op_1226_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1226_cast_fp16 = rsqrt(epsilon = var_1226_epsilon_0, x = var_1225_cast_fp16)[name = string("op_1226_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1227_cast_fp16 = mul(x = y_59_cast_fp16, y = var_1226_cast_fp16)[name = string("op_1227_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_5_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46540864)))];
            tensor<fp16, [1, 24, 512]> a_11_cast_fp16 = mul(x = core_backbone_layers_5_post_operator_norm_weight_to_fp16, y = var_1227_cast_fp16)[name = string("a_11_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_87_cast_fp16 = add(x = x_83_cast_fp16, y = a_11_cast_fp16)[name = string("x_87_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1231_cast_fp16 = abs(x = x_87_cast_fp16)[name = string("op_1231_cast_fp16")];
            tensor<int32, [1]> var_1233_axes_0 = const()[name = string("op_1233_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1233_cast_fp16 = reduce_max(axes = var_1233_axes_0, keep_dims = var_33, x = var_1231_cast_fp16)[name = string("op_1233_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_57_cast_fp16 = maximum(x = var_1233_cast_fp16, y = var_34_to_fp16)[name = string("s_57_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_61_cast_fp16 = real_div(x = x_87_cast_fp16, y = s_57_cast_fp16)[name = string("y_61_cast_fp16")];
            fp16 var_35_promoted_26_to_fp16 = const()[name = string("op_35_promoted_26_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1236_cast_fp16 = pow(x = y_61_cast_fp16, y = var_35_promoted_26_to_fp16)[name = string("op_1236_cast_fp16")];
            tensor<int32, [1]> var_1238_axes_0 = const()[name = string("op_1238_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1238_keep_dims_0 = const()[name = string("op_1238_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1238_cast_fp16 = reduce_mean(axes = var_1238_axes_0, keep_dims = var_1238_keep_dims_0, x = var_1236_cast_fp16)[name = string("op_1238_cast_fp16")];
            fp32 var_1239_epsilon_0 = const()[name = string("op_1239_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1239_cast_fp16 = inverse(epsilon = var_1239_epsilon_0, x = s_57_cast_fp16)[name = string("op_1239_cast_fp16")];
            fp16 var_1240_to_fp16 = const()[name = string("op_1240_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1241_cast_fp16 = mul(x = var_1239_cast_fp16, y = var_1240_to_fp16)[name = string("op_1241_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1242_cast_fp16 = real_div(x = var_1241_cast_fp16, y = s_57_cast_fp16)[name = string("op_1242_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1243_cast_fp16 = add(x = var_1238_cast_fp16, y = var_1242_cast_fp16)[name = string("op_1243_cast_fp16")];
            fp32 var_1244_epsilon_0 = const()[name = string("op_1244_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1244_cast_fp16 = rsqrt(epsilon = var_1244_epsilon_0, x = var_1243_cast_fp16)[name = string("op_1244_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1245_cast_fp16 = mul(x = y_61_cast_fp16, y = var_1244_cast_fp16)[name = string("op_1245_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_5_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46541952)))];
            tensor<fp16, [1, 24, 512]> input_77_cast_fp16 = mul(x = core_backbone_layers_5_ffn_norm_weight_to_fp16, y = var_1245_cast_fp16)[name = string("input_77_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_5_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46543040)))];
            tensor<fp16, [1, 24, 2304]> linear_34_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w1_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_34_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_5_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48902400)))];
            tensor<fp16, [1, 24, 2304]> linear_35_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w3_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_35_cast_fp16")];
            fp16 var_1254_to_fp16 = const()[name = string("op_1254_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 24, 2304]> var_1255_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_1254_to_fp16)[name = string("op_1255_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_1258_cast_fp16 = tanh(x = var_1255_cast_fp16)[name = string("op_1258_cast_fp16")];
            fp16 var_1259_to_fp16 = const()[name = string("op_1259_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 24, 2304]> var_1260_cast_fp16 = add(x = var_1258_cast_fp16, y = var_1259_to_fp16)[name = string("op_1260_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_1261_cast_fp16 = mul(x = var_1255_cast_fp16, y = var_1260_cast_fp16)[name = string("op_1261_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> input_79_cast_fp16 = mul(x = var_1261_cast_fp16, y = linear_35_cast_fp16)[name = string("input_79_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_5_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51261760)))];
            tensor<fp16, [1, 24, 512]> linear_36_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w2_weight_to_fp16, x = input_79_cast_fp16)[name = string("linear_36_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1266_cast_fp16 = abs(x = linear_36_cast_fp16)[name = string("op_1266_cast_fp16")];
            tensor<int32, [1]> var_1268_axes_0 = const()[name = string("op_1268_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1268_cast_fp16 = reduce_max(axes = var_1268_axes_0, keep_dims = var_33, x = var_1266_cast_fp16)[name = string("op_1268_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_59_cast_fp16 = maximum(x = var_1268_cast_fp16, y = var_34_to_fp16)[name = string("s_59_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_63_cast_fp16 = real_div(x = linear_36_cast_fp16, y = s_59_cast_fp16)[name = string("y_63_cast_fp16")];
            fp16 var_35_promoted_27_to_fp16 = const()[name = string("op_35_promoted_27_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1271_cast_fp16 = pow(x = y_63_cast_fp16, y = var_35_promoted_27_to_fp16)[name = string("op_1271_cast_fp16")];
            tensor<int32, [1]> var_1273_axes_0 = const()[name = string("op_1273_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1273_keep_dims_0 = const()[name = string("op_1273_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1273_cast_fp16 = reduce_mean(axes = var_1273_axes_0, keep_dims = var_1273_keep_dims_0, x = var_1271_cast_fp16)[name = string("op_1273_cast_fp16")];
            fp32 var_1274_epsilon_0 = const()[name = string("op_1274_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1274_cast_fp16 = inverse(epsilon = var_1274_epsilon_0, x = s_59_cast_fp16)[name = string("op_1274_cast_fp16")];
            fp16 var_1275_to_fp16 = const()[name = string("op_1275_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1276_cast_fp16 = mul(x = var_1274_cast_fp16, y = var_1275_to_fp16)[name = string("op_1276_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1277_cast_fp16 = real_div(x = var_1276_cast_fp16, y = s_59_cast_fp16)[name = string("op_1277_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1278_cast_fp16 = add(x = var_1273_cast_fp16, y = var_1277_cast_fp16)[name = string("op_1278_cast_fp16")];
            fp32 var_1279_epsilon_0 = const()[name = string("op_1279_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1279_cast_fp16 = rsqrt(epsilon = var_1279_epsilon_0, x = var_1278_cast_fp16)[name = string("op_1279_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1280_cast_fp16 = mul(x = y_63_cast_fp16, y = var_1279_cast_fp16)[name = string("op_1280_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_5_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53621120)))];
            tensor<fp16, [1, 24, 512]> f_11_cast_fp16 = mul(x = core_backbone_layers_5_post_ffn_norm_weight_to_fp16, y = var_1280_cast_fp16)[name = string("f_11_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_93_cast_fp16 = add(x = x_87_cast_fp16, y = f_11_cast_fp16)[name = string("x_93_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1284_cast_fp16 = abs(x = x_93_cast_fp16)[name = string("op_1284_cast_fp16")];
            tensor<int32, [1]> var_1286_axes_0 = const()[name = string("op_1286_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1286_cast_fp16 = reduce_max(axes = var_1286_axes_0, keep_dims = var_33, x = var_1284_cast_fp16)[name = string("op_1286_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_61_cast_fp16 = maximum(x = var_1286_cast_fp16, y = var_34_to_fp16)[name = string("s_61_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_65_cast_fp16 = real_div(x = x_93_cast_fp16, y = s_61_cast_fp16)[name = string("y_65_cast_fp16")];
            fp16 var_35_promoted_28_to_fp16 = const()[name = string("op_35_promoted_28_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1289_cast_fp16 = pow(x = y_65_cast_fp16, y = var_35_promoted_28_to_fp16)[name = string("op_1289_cast_fp16")];
            tensor<int32, [1]> var_1291_axes_0 = const()[name = string("op_1291_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1291_keep_dims_0 = const()[name = string("op_1291_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1291_cast_fp16 = reduce_mean(axes = var_1291_axes_0, keep_dims = var_1291_keep_dims_0, x = var_1289_cast_fp16)[name = string("op_1291_cast_fp16")];
            fp32 var_1292_epsilon_0 = const()[name = string("op_1292_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1292_cast_fp16 = inverse(epsilon = var_1292_epsilon_0, x = s_61_cast_fp16)[name = string("op_1292_cast_fp16")];
            fp16 var_1293_to_fp16 = const()[name = string("op_1293_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1294_cast_fp16 = mul(x = var_1292_cast_fp16, y = var_1293_to_fp16)[name = string("op_1294_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1295_cast_fp16 = real_div(x = var_1294_cast_fp16, y = s_61_cast_fp16)[name = string("op_1295_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1296_cast_fp16 = add(x = var_1291_cast_fp16, y = var_1295_cast_fp16)[name = string("op_1296_cast_fp16")];
            fp32 var_1297_epsilon_0 = const()[name = string("op_1297_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1297_cast_fp16 = rsqrt(epsilon = var_1297_epsilon_0, x = var_1296_cast_fp16)[name = string("op_1297_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1298_cast_fp16 = mul(x = y_65_cast_fp16, y = var_1297_cast_fp16)[name = string("op_1298_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_6_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53622208)))];
            tensor<fp16, [1, 24, 512]> input_81_cast_fp16 = mul(x = core_backbone_layers_6_operator_norm_weight_to_fp16, y = var_1298_cast_fp16)[name = string("input_81_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53623296)))];
            tensor<fp16, [1, 24, 512]> linear_37_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_q_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_37_cast_fp16")];
            tensor<int32, [4]> concat_16 = const()[name = string("concat_16"), val = tensor<int32, [4]>([1, 24, 8, 64])];
            tensor<fp16, [1, 24, 8, 64]> x_95_cast_fp16 = reshape(shape = concat_16, x = linear_37_cast_fp16)[name = string("x_95_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> var_1305_cast_fp16 = abs(x = x_95_cast_fp16)[name = string("op_1305_cast_fp16")];
            tensor<int32, [1]> var_1307_axes_0 = const()[name = string("op_1307_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 8, 1]> var_1307_cast_fp16 = reduce_max(axes = var_1307_axes_0, keep_dims = var_33, x = var_1305_cast_fp16)[name = string("op_1307_cast_fp16")];
            tensor<fp16, [1, 24, 8, 1]> s_63_cast_fp16 = maximum(x = var_1307_cast_fp16, y = var_34_to_fp16)[name = string("s_63_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> y_67_cast_fp16 = real_div(x = x_95_cast_fp16, y = s_63_cast_fp16)[name = string("y_67_cast_fp16")];
            fp16 var_35_promoted_29_to_fp16 = const()[name = string("op_35_promoted_29_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 8, 64]> var_1310_cast_fp16 = pow(x = y_67_cast_fp16, y = var_35_promoted_29_to_fp16)[name = string("op_1310_cast_fp16")];
            tensor<int32, [1]> var_1312_axes_0 = const()[name = string("op_1312_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1312_keep_dims_0 = const()[name = string("op_1312_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 8, 1]> var_1312_cast_fp16 = reduce_mean(axes = var_1312_axes_0, keep_dims = var_1312_keep_dims_0, x = var_1310_cast_fp16)[name = string("op_1312_cast_fp16")];
            fp32 var_1313_epsilon_0 = const()[name = string("op_1313_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 8, 1]> var_1313_cast_fp16 = inverse(epsilon = var_1313_epsilon_0, x = s_63_cast_fp16)[name = string("op_1313_cast_fp16")];
            fp16 var_1314_to_fp16 = const()[name = string("op_1314_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 8, 1]> var_1315_cast_fp16 = mul(x = var_1313_cast_fp16, y = var_1314_to_fp16)[name = string("op_1315_cast_fp16")];
            tensor<fp16, [1, 24, 8, 1]> var_1316_cast_fp16 = real_div(x = var_1315_cast_fp16, y = s_63_cast_fp16)[name = string("op_1316_cast_fp16")];
            tensor<fp16, [1, 24, 8, 1]> var_1317_cast_fp16 = add(x = var_1312_cast_fp16, y = var_1316_cast_fp16)[name = string("op_1317_cast_fp16")];
            fp32 var_1318_epsilon_0 = const()[name = string("op_1318_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 8, 1]> var_1318_cast_fp16 = rsqrt(epsilon = var_1318_epsilon_0, x = var_1317_cast_fp16)[name = string("op_1318_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> var_1319_cast_fp16 = mul(x = y_67_cast_fp16, y = var_1318_cast_fp16)[name = string("op_1319_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_6_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54147648)))];
            tensor<fp16, [1, 24, 8, 64]> var_1320_cast_fp16 = mul(x = core_backbone_layers_6_self_attn_q_norm_weight_to_fp16, y = var_1319_cast_fp16)[name = string("op_1320_cast_fp16")];
            tensor<fp16, [128, 512]> core_backbone_layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54147840)))];
            tensor<fp16, [1, 24, 128]> linear_38_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_k_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_38_cast_fp16")];
            tensor<int32, [4]> concat_17 = const()[name = string("concat_17"), val = tensor<int32, [4]>([1, 24, 2, 64])];
            tensor<fp16, [1, 24, 2, 64]> x_97_cast_fp16 = reshape(shape = concat_17, x = linear_38_cast_fp16)[name = string("x_97_cast_fp16")];
            tensor<fp16, [1, 24, 2, 64]> var_1327_cast_fp16 = abs(x = x_97_cast_fp16)[name = string("op_1327_cast_fp16")];
            tensor<int32, [1]> var_1329_axes_0 = const()[name = string("op_1329_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 2, 1]> var_1329_cast_fp16 = reduce_max(axes = var_1329_axes_0, keep_dims = var_33, x = var_1327_cast_fp16)[name = string("op_1329_cast_fp16")];
            tensor<fp16, [1, 24, 2, 1]> s_65_cast_fp16 = maximum(x = var_1329_cast_fp16, y = var_34_to_fp16)[name = string("s_65_cast_fp16")];
            tensor<fp16, [1, 24, 2, 64]> y_69_cast_fp16 = real_div(x = x_97_cast_fp16, y = s_65_cast_fp16)[name = string("y_69_cast_fp16")];
            fp16 var_35_promoted_30_to_fp16 = const()[name = string("op_35_promoted_30_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 2, 64]> var_1332_cast_fp16 = pow(x = y_69_cast_fp16, y = var_35_promoted_30_to_fp16)[name = string("op_1332_cast_fp16")];
            tensor<int32, [1]> var_1334_axes_0 = const()[name = string("op_1334_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1334_keep_dims_0 = const()[name = string("op_1334_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 2, 1]> var_1334_cast_fp16 = reduce_mean(axes = var_1334_axes_0, keep_dims = var_1334_keep_dims_0, x = var_1332_cast_fp16)[name = string("op_1334_cast_fp16")];
            fp32 var_1335_epsilon_0 = const()[name = string("op_1335_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 2, 1]> var_1335_cast_fp16 = inverse(epsilon = var_1335_epsilon_0, x = s_65_cast_fp16)[name = string("op_1335_cast_fp16")];
            fp16 var_1336_to_fp16 = const()[name = string("op_1336_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 2, 1]> var_1337_cast_fp16 = mul(x = var_1335_cast_fp16, y = var_1336_to_fp16)[name = string("op_1337_cast_fp16")];
            tensor<fp16, [1, 24, 2, 1]> var_1338_cast_fp16 = real_div(x = var_1337_cast_fp16, y = s_65_cast_fp16)[name = string("op_1338_cast_fp16")];
            tensor<fp16, [1, 24, 2, 1]> var_1339_cast_fp16 = add(x = var_1334_cast_fp16, y = var_1338_cast_fp16)[name = string("op_1339_cast_fp16")];
            fp32 var_1340_epsilon_0 = const()[name = string("op_1340_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 2, 1]> var_1340_cast_fp16 = rsqrt(epsilon = var_1340_epsilon_0, x = var_1339_cast_fp16)[name = string("op_1340_cast_fp16")];
            tensor<fp16, [1, 24, 2, 64]> var_1341_cast_fp16 = mul(x = y_69_cast_fp16, y = var_1340_cast_fp16)[name = string("op_1341_cast_fp16")];
            tensor<fp16, [64]> core_backbone_layers_6_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54278976)))];
            tensor<fp16, [1, 24, 2, 64]> var_1342_cast_fp16 = mul(x = core_backbone_layers_6_self_attn_k_norm_weight_to_fp16, y = var_1341_cast_fp16)[name = string("op_1342_cast_fp16")];
            tensor<int32, [4]> new_17_perm_0 = const()[name = string("new_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [128, 512]> core_backbone_layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [128, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54279168)))];
            tensor<fp16, [1, 24, 128]> linear_39_cast_fp16 = linear(bias = linear_12_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_v_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_39_cast_fp16")];
            tensor<int32, [4]> concat_18 = const()[name = string("concat_18"), val = tensor<int32, [4]>([1, 24, 2, 64])];
            tensor<fp16, [1, 24, 2, 64]> var_1347_cast_fp16 = reshape(shape = concat_18, x = linear_39_cast_fp16)[name = string("op_1347_cast_fp16")];
            tensor<int32, [4]> new_19_perm_0 = const()[name = string("new_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_85_begin_0 = const()[name = string("input_85_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])];
            tensor<int32, [5]> input_85_end_0 = const()[name = string("input_85_end_0"), val = tensor<int32, [5]>([3, 1, 2, 29, 64])];
            tensor<bool, [5]> input_85_end_mask_0 = const()[name = string("input_85_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_85_squeeze_mask_0 = const()[name = string("input_85_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_85_cast_fp16 = slice_by_index(begin = input_85_begin_0, end = input_85_end_0, end_mask = input_85_end_mask_0, squeeze_mask = input_85_squeeze_mask_0, x = k)[name = string("input_85_cast_fp16")];
            tensor<int32, [8]> var_1352_pad_0 = const()[name = string("op_1352_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_1352_mode_0 = const()[name = string("op_1352_mode_0"), val = string("constant")];
            fp16 const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 24, 64]> new_17_cast_fp16 = transpose(perm = new_17_perm_0, x = var_1342_cast_fp16)[name = string("transpose_11")];
            tensor<fp16, [1, 2, 53, 64]> var_1352_cast_fp16 = pad(constant_val = const_25_to_fp16, mode = var_1352_mode_0, pad = var_1352_pad_0, x = new_17_cast_fp16)[name = string("op_1352_cast_fp16")];
            tensor<int32, [8]> var_1354_pad_0 = const()[name = string("op_1354_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 24, 0, 0])];
            string var_1354_mode_0 = const()[name = string("op_1354_mode_0"), val = string("constant")];
            fp16 const_26_to_fp16 = const()[name = string("const_26_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 53, 64]> var_1354_cast_fp16 = pad(constant_val = const_26_to_fp16, mode = var_1354_mode_0, pad = var_1354_pad_0, x = input_85_cast_fp16)[name = string("op_1354_cast_fp16")];
            tensor<fp16, [1, 2, 53, 64]> x_101_cast_fp16 = add(x = var_1352_cast_fp16, y = var_1354_cast_fp16)[name = string("x_101_cast_fp16")];
            tensor<int32, [5]> input_89_begin_0 = const()[name = string("input_89_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])];
            tensor<int32, [5]> input_89_end_0 = const()[name = string("input_89_end_0"), val = tensor<int32, [5]>([3, 1, 2, 29, 64])];
            tensor<bool, [5]> input_89_end_mask_0 = const()[name = string("input_89_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_89_squeeze_mask_0 = const()[name = string("input_89_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 2, 29, 64]> input_89_cast_fp16 = slice_by_index(begin = input_89_begin_0, end = input_89_end_0, end_mask = input_89_end_mask_0, squeeze_mask = input_89_squeeze_mask_0, x = v)[name = string("input_89_cast_fp16")];
            tensor<int32, [8]> var_1359_pad_0 = const()[name = string("op_1359_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 29, 0, 0, 0])];
            string var_1359_mode_0 = const()[name = string("op_1359_mode_0"), val = string("constant")];
            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 24, 64]> new_19_cast_fp16 = transpose(perm = new_19_perm_0, x = var_1347_cast_fp16)[name = string("transpose_10")];
            tensor<fp16, [1, 2, 53, 64]> var_1359_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = var_1359_mode_0, pad = var_1359_pad_0, x = new_19_cast_fp16)[name = string("op_1359_cast_fp16")];
            tensor<int32, [8]> var_1361_pad_0 = const()[name = string("op_1361_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 24, 0, 0])];
            string var_1361_mode_0 = const()[name = string("op_1361_mode_0"), val = string("constant")];
            fp16 const_28_to_fp16 = const()[name = string("const_28_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 2, 53, 64]> var_1361_cast_fp16 = pad(constant_val = const_28_to_fp16, mode = var_1361_mode_0, pad = var_1361_pad_0, x = input_89_cast_fp16)[name = string("op_1361_cast_fp16")];
            tensor<fp16, [1, 2, 53, 64]> v_11_cast_fp16 = add(x = var_1359_cast_fp16, y = var_1361_cast_fp16)[name = string("v_11_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> var_1363_cast_fp16 = mul(x = var_1320_cast_fp16, y = const_33_to_fp16)[name = string("op_1363_cast_fp16")];
            tensor<int32, [2]> var_1364_split_sizes_0 = const()[name = string("op_1364_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 const_42 = const()[name = string("const_42"), val = int32(3)];
            tensor<fp16, [1, 24, 8, 32]> var_1364_cast_fp16_0, tensor<fp16, [1, 24, 8, 32]> var_1364_cast_fp16_1 = split(axis = const_42, split_sizes = var_1364_split_sizes_0, x = var_1320_cast_fp16)[name = string("op_1364_cast_fp16")];
            fp16 const_29_promoted_to_fp16 = const()[name = string("const_29_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 24, 8, 32]> var_1366_cast_fp16 = mul(x = var_1364_cast_fp16_1, y = const_29_promoted_to_fp16)[name = string("op_1366_cast_fp16")];
            bool var_1368_interleave_0 = const()[name = string("op_1368_interleave_0"), val = bool(false)];
            int32 const_43 = const()[name = string("const_43"), val = int32(3)];
            tensor<fp16, [1, 24, 8, 64]> var_1368_cast_fp16 = concat(axis = const_43, interleave = var_1368_interleave_0, values = (var_1366_cast_fp16, var_1364_cast_fp16_0))[name = string("op_1368_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> var_1369_cast_fp16 = mul(x = var_1368_cast_fp16, y = const_36_to_fp16)[name = string("op_1369_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> q_cast_fp16 = add(x = var_1363_cast_fp16, y = var_1369_cast_fp16)[name = string("q_cast_fp16")];
            tensor<fp16, [1, 2, 53, 64]> var_1371_cast_fp16 = mul(x = x_101_cast_fp16, y = const_1_to_fp16)[name = string("op_1371_cast_fp16")];
            tensor<int32, [2]> var_1372_split_sizes_0 = const()[name = string("op_1372_split_sizes_0"), val = tensor<int32, [2]>([32, 32])];
            int32 var_1372_axis_0 = const()[name = string("op_1372_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 2, 53, 32]> var_1372_cast_fp16_0, tensor<fp16, [1, 2, 53, 32]> var_1372_cast_fp16_1 = split(axis = var_1372_axis_0, split_sizes = var_1372_split_sizes_0, x = x_101_cast_fp16)[name = string("op_1372_cast_fp16")];
            fp16 const_30_promoted_to_fp16 = const()[name = string("const_30_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 2, 53, 32]> var_1374_cast_fp16 = mul(x = var_1372_cast_fp16_1, y = const_30_promoted_to_fp16)[name = string("op_1374_cast_fp16")];
            bool var_1376_interleave_0 = const()[name = string("op_1376_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 2, 53, 64]> var_1376_cast_fp16 = concat(axis = var_32, interleave = var_1376_interleave_0, values = (var_1374_cast_fp16, var_1372_cast_fp16_0))[name = string("op_1376_cast_fp16")];
            tensor<fp16, [1, 2, 53, 64]> var_1377_cast_fp16 = mul(x = var_1376_cast_fp16, y = const_2_to_fp16)[name = string("op_1377_cast_fp16")];
            tensor<fp16, [1, 2, 53, 64]> k_11_cast_fp16 = add(x = var_1371_cast_fp16, y = var_1377_cast_fp16)[name = string("k_11_cast_fp16")];
            tensor<int32, [1]> var_1379_axes_0 = const()[name = string("op_1379_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 53, 64]> var_1379_cast_fp16 = expand_dims(axes = var_1379_axes_0, x = k_11_cast_fp16)[name = string("op_1379_cast_fp16")];
            tensor<int32, [5]> var_1381_reps_0 = const()[name = string("op_1381_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 53, 64]> var_1381_cast_fp16 = tile(reps = var_1381_reps_0, x = var_1379_cast_fp16)[name = string("op_1381_cast_fp16")];
            tensor<int32, [4]> concat_20 = const()[name = string("concat_20"), val = tensor<int32, [4]>([1, 8, 53, 64])];
            tensor<fp16, [1, 8, 53, 64]> k_cast_fp16 = reshape(shape = concat_20, x = var_1381_cast_fp16)[name = string("k_cast_fp16")];
            tensor<int32, [1]> var_1383_axes_0 = const()[name = string("op_1383_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2, 1, 53, 64]> var_1383_cast_fp16 = expand_dims(axes = var_1383_axes_0, x = v_11_cast_fp16)[name = string("op_1383_cast_fp16")];
            tensor<int32, [5]> var_1385_reps_0 = const()[name = string("op_1385_reps_0"), val = tensor<int32, [5]>([1, 1, 4, 1, 1])];
            tensor<fp16, [1, 2, 4, 53, 64]> var_1385_cast_fp16 = tile(reps = var_1385_reps_0, x = var_1383_cast_fp16)[name = string("op_1385_cast_fp16")];
            tensor<int32, [4]> concat_22 = const()[name = string("concat_22"), val = tensor<int32, [4]>([1, 8, 53, 64])];
            tensor<fp16, [1, 8, 53, 64]> v_cast_fp16 = reshape(shape = concat_22, x = var_1385_cast_fp16)[name = string("v_cast_fp16")];
            tensor<int32, [4]> transpose_4_perm_0 = const()[name = string("transpose_4_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1388_transpose_x_1 = const()[name = string("op_1388_transpose_x_1"), val = bool(false)];
            bool var_1388_transpose_y_1 = const()[name = string("op_1388_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 24, 64]> transpose_4_cast_fp16 = transpose(perm = transpose_4_perm_0, x = q_cast_fp16)[name = string("transpose_9")];
            tensor<fp16, [1, 8, 24, 53]> var_1388_cast_fp16 = matmul(transpose_x = var_1388_transpose_x_1, transpose_y = var_1388_transpose_y_1, x = transpose_4_cast_fp16, y = k_cast_fp16)[name = string("op_1388_cast_fp16")];
            fp16 var_1389_to_fp16 = const()[name = string("op_1389_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 8, 24, 53]> s_67_cast_fp16 = mul(x = var_1388_cast_fp16, y = var_1389_to_fp16)[name = string("s_67_cast_fp16")];
            tensor<fp16, [1, 8, 24, 53]> x_103_cast_fp16 = add(x = s_67_cast_fp16, y = mask)[name = string("x_103_cast_fp16")];
            tensor<int32, [1]> var_1393_axes_0 = const()[name = string("op_1393_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 24, 1]> var_1393_cast_fp16 = reduce_max(axes = var_1393_axes_0, keep_dims = var_33, x = x_103_cast_fp16)[name = string("op_1393_cast_fp16")];
            tensor<fp16, [1, 8, 24, 53]> x_105_cast_fp16 = sub(x = x_103_cast_fp16, y = var_1393_cast_fp16)[name = string("x_105_cast_fp16")];
            tensor<fp16, [1, 8, 24, 53]> x_107_cast_fp16 = exp(x = x_105_cast_fp16)[name = string("x_107_cast_fp16")];
            tensor<int32, [1]> var_1397_axes_0 = const()[name = string("op_1397_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1397_keep_dims_0 = const()[name = string("op_1397_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 24, 1]> var_1397_cast_fp16 = reduce_sum(axes = var_1397_axes_0, keep_dims = var_1397_keep_dims_0, x = x_107_cast_fp16)[name = string("op_1397_cast_fp16")];
            tensor<fp16, [1, 8, 24, 53]> var_1398_cast_fp16 = real_div(x = x_107_cast_fp16, y = var_1397_cast_fp16)[name = string("op_1398_cast_fp16")];
            bool o_13_transpose_x_0 = const()[name = string("o_13_transpose_x_0"), val = bool(false)];
            bool o_13_transpose_y_0 = const()[name = string("o_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 24, 64]> o_13_cast_fp16 = matmul(transpose_x = o_13_transpose_x_0, transpose_y = o_13_transpose_y_0, x = var_1398_cast_fp16, y = v_cast_fp16)[name = string("o_13_cast_fp16")];
            tensor<int32, [4]> o_15_perm_0 = const()[name = string("o_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [8, 512]> core_backbone_layers_6_self_attn_gate_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_gate_proj_weight_to_fp16"), val = tensor<fp16, [8, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54410304)))];
            tensor<fp16, [8]> linear_40_bias_0_to_fp16 = const()[name = string("linear_40_bias_0_to_fp16"), val = tensor<fp16, [8]>([0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0])];
            tensor<fp16, [1, 24, 8]> linear_40_cast_fp16 = linear(bias = linear_40_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_gate_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_40_cast_fp16")];
            fp16 var_1404_to_fp16 = const()[name = string("op_1404_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 24, 8]> var_1405_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_1404_to_fp16)[name = string("op_1405_cast_fp16")];
            tensor<fp16, [1, 24, 8]> var_1406_cast_fp16 = tanh(x = var_1405_cast_fp16)[name = string("op_1406_cast_fp16")];
            fp16 var_1407_to_fp16 = const()[name = string("op_1407_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 24, 8]> var_1408_cast_fp16 = add(x = var_1406_cast_fp16, y = var_1407_to_fp16)[name = string("op_1408_cast_fp16")];
            fp16 var_1409_to_fp16 = const()[name = string("op_1409_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 24, 8]> var_1410_cast_fp16 = mul(x = var_1408_cast_fp16, y = var_1409_to_fp16)[name = string("op_1410_cast_fp16")];
            tensor<int32, [1]> var_1411_axes_0 = const()[name = string("op_1411_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 8, 1]> var_1411_cast_fp16 = expand_dims(axes = var_1411_axes_0, x = var_1410_cast_fp16)[name = string("op_1411_cast_fp16")];
            tensor<fp16, [1, 24, 8, 64]> o_15_cast_fp16 = transpose(perm = o_15_perm_0, x = o_13_cast_fp16)[name = string("transpose_8")];
            tensor<fp16, [1, 24, 8, 64]> o_cast_fp16 = mul(x = o_15_cast_fp16, y = var_1411_cast_fp16)[name = string("o_cast_fp16")];
            tensor<int32, [3]> concat_23 = const()[name = string("concat_23"), val = tensor<int32, [3]>([1, 24, 512])];
            tensor<fp16, [1, 24, 512]> input_91_cast_fp16 = reshape(shape = concat_23, x = o_cast_fp16)[name = string("input_91_cast_fp16")];
            tensor<fp16, [512, 512]> core_backbone_layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54418560)))];
            tensor<fp16, [1, 24, 512]> linear_41_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_self_attn_o_proj_weight_to_fp16, x = input_91_cast_fp16)[name = string("linear_41_cast_fp16")];
            tensor<int32, [4]> var_1418_begin_0 = const()[name = string("op_1418_begin_0"), val = tensor<int32, [4]>([0, 0, 24, 0])];
            tensor<int32, [4]> var_1418_end_0 = const()[name = string("op_1418_end_0"), val = tensor<int32, [4]>([1, 2, 53, 64])];
            tensor<bool, [4]> var_1418_end_mask_0 = const()[name = string("op_1418_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_1418_cast_fp16 = slice_by_index(begin = var_1418_begin_0, end = var_1418_end_0, end_mask = var_1418_end_mask_0, x = x_101_cast_fp16)[name = string("op_1418_cast_fp16")];
            tensor<int32, [4]> var_1421_begin_0 = const()[name = string("op_1421_begin_0"), val = tensor<int32, [4]>([0, 0, 24, 0])];
            tensor<int32, [4]> var_1421_end_0 = const()[name = string("op_1421_end_0"), val = tensor<int32, [4]>([1, 2, 53, 64])];
            tensor<bool, [4]> var_1421_end_mask_0 = const()[name = string("op_1421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 2, 29, 64]> var_1421_cast_fp16 = slice_by_index(begin = var_1421_begin_0, end = var_1421_end_0, end_mask = var_1421_end_mask_0, x = v_11_cast_fp16)[name = string("op_1421_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1423_cast_fp16 = abs(x = linear_41_cast_fp16)[name = string("op_1423_cast_fp16")];
            tensor<int32, [1]> var_1425_axes_0 = const()[name = string("op_1425_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1425_cast_fp16 = reduce_max(axes = var_1425_axes_0, keep_dims = var_33, x = var_1423_cast_fp16)[name = string("op_1425_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_69_cast_fp16 = maximum(x = var_1425_cast_fp16, y = var_34_to_fp16)[name = string("s_69_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_71_cast_fp16 = real_div(x = linear_41_cast_fp16, y = s_69_cast_fp16)[name = string("y_71_cast_fp16")];
            fp16 var_35_promoted_31_to_fp16 = const()[name = string("op_35_promoted_31_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1428_cast_fp16 = pow(x = y_71_cast_fp16, y = var_35_promoted_31_to_fp16)[name = string("op_1428_cast_fp16")];
            tensor<int32, [1]> var_1430_axes_0 = const()[name = string("op_1430_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1430_keep_dims_0 = const()[name = string("op_1430_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1430_cast_fp16 = reduce_mean(axes = var_1430_axes_0, keep_dims = var_1430_keep_dims_0, x = var_1428_cast_fp16)[name = string("op_1430_cast_fp16")];
            fp32 var_1431_epsilon_0 = const()[name = string("op_1431_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1431_cast_fp16 = inverse(epsilon = var_1431_epsilon_0, x = s_69_cast_fp16)[name = string("op_1431_cast_fp16")];
            fp16 var_1432_to_fp16 = const()[name = string("op_1432_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1433_cast_fp16 = mul(x = var_1431_cast_fp16, y = var_1432_to_fp16)[name = string("op_1433_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1434_cast_fp16 = real_div(x = var_1433_cast_fp16, y = s_69_cast_fp16)[name = string("op_1434_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1435_cast_fp16 = add(x = var_1430_cast_fp16, y = var_1434_cast_fp16)[name = string("op_1435_cast_fp16")];
            fp32 var_1436_epsilon_0 = const()[name = string("op_1436_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1436_cast_fp16 = rsqrt(epsilon = var_1436_epsilon_0, x = var_1435_cast_fp16)[name = string("op_1436_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1437_cast_fp16 = mul(x = y_71_cast_fp16, y = var_1436_cast_fp16)[name = string("op_1437_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_6_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54942912)))];
            tensor<fp16, [1, 24, 512]> a_13_cast_fp16 = mul(x = core_backbone_layers_6_post_operator_norm_weight_to_fp16, y = var_1437_cast_fp16)[name = string("a_13_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_113_cast_fp16 = add(x = x_93_cast_fp16, y = a_13_cast_fp16)[name = string("x_113_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1441_cast_fp16 = abs(x = x_113_cast_fp16)[name = string("op_1441_cast_fp16")];
            tensor<int32, [1]> var_1443_axes_0 = const()[name = string("op_1443_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1443_cast_fp16 = reduce_max(axes = var_1443_axes_0, keep_dims = var_33, x = var_1441_cast_fp16)[name = string("op_1443_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_71_cast_fp16 = maximum(x = var_1443_cast_fp16, y = var_34_to_fp16)[name = string("s_71_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_73_cast_fp16 = real_div(x = x_113_cast_fp16, y = s_71_cast_fp16)[name = string("y_73_cast_fp16")];
            fp16 var_35_promoted_32_to_fp16 = const()[name = string("op_35_promoted_32_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1446_cast_fp16 = pow(x = y_73_cast_fp16, y = var_35_promoted_32_to_fp16)[name = string("op_1446_cast_fp16")];
            tensor<int32, [1]> var_1448_axes_0 = const()[name = string("op_1448_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1448_keep_dims_0 = const()[name = string("op_1448_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1448_cast_fp16 = reduce_mean(axes = var_1448_axes_0, keep_dims = var_1448_keep_dims_0, x = var_1446_cast_fp16)[name = string("op_1448_cast_fp16")];
            fp32 var_1449_epsilon_0 = const()[name = string("op_1449_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1449_cast_fp16 = inverse(epsilon = var_1449_epsilon_0, x = s_71_cast_fp16)[name = string("op_1449_cast_fp16")];
            fp16 var_1450_to_fp16 = const()[name = string("op_1450_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1451_cast_fp16 = mul(x = var_1449_cast_fp16, y = var_1450_to_fp16)[name = string("op_1451_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1452_cast_fp16 = real_div(x = var_1451_cast_fp16, y = s_71_cast_fp16)[name = string("op_1452_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1453_cast_fp16 = add(x = var_1448_cast_fp16, y = var_1452_cast_fp16)[name = string("op_1453_cast_fp16")];
            fp32 var_1454_epsilon_0 = const()[name = string("op_1454_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1454_cast_fp16 = rsqrt(epsilon = var_1454_epsilon_0, x = var_1453_cast_fp16)[name = string("op_1454_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1455_cast_fp16 = mul(x = y_73_cast_fp16, y = var_1454_cast_fp16)[name = string("op_1455_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_6_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54944000)))];
            tensor<fp16, [1, 24, 512]> input_93_cast_fp16 = mul(x = core_backbone_layers_6_ffn_norm_weight_to_fp16, y = var_1455_cast_fp16)[name = string("input_93_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_6_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54945088)))];
            tensor<fp16, [1, 24, 2304]> linear_42_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w1_weight_to_fp16, x = input_93_cast_fp16)[name = string("linear_42_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_6_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57304448)))];
            tensor<fp16, [1, 24, 2304]> linear_43_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w3_weight_to_fp16, x = input_93_cast_fp16)[name = string("linear_43_cast_fp16")];
            fp16 var_1464_to_fp16 = const()[name = string("op_1464_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 24, 2304]> var_1465_cast_fp16 = mul(x = linear_42_cast_fp16, y = var_1464_to_fp16)[name = string("op_1465_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_1468_cast_fp16 = tanh(x = var_1465_cast_fp16)[name = string("op_1468_cast_fp16")];
            fp16 var_1469_to_fp16 = const()[name = string("op_1469_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 24, 2304]> var_1470_cast_fp16 = add(x = var_1468_cast_fp16, y = var_1469_to_fp16)[name = string("op_1470_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_1471_cast_fp16 = mul(x = var_1465_cast_fp16, y = var_1470_cast_fp16)[name = string("op_1471_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> input_95_cast_fp16 = mul(x = var_1471_cast_fp16, y = linear_43_cast_fp16)[name = string("input_95_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_6_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59663808)))];
            tensor<fp16, [1, 24, 512]> linear_44_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w2_weight_to_fp16, x = input_95_cast_fp16)[name = string("linear_44_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1476_cast_fp16 = abs(x = linear_44_cast_fp16)[name = string("op_1476_cast_fp16")];
            tensor<int32, [1]> var_1478_axes_0 = const()[name = string("op_1478_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1478_cast_fp16 = reduce_max(axes = var_1478_axes_0, keep_dims = var_33, x = var_1476_cast_fp16)[name = string("op_1478_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_73_cast_fp16 = maximum(x = var_1478_cast_fp16, y = var_34_to_fp16)[name = string("s_73_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_75_cast_fp16 = real_div(x = linear_44_cast_fp16, y = s_73_cast_fp16)[name = string("y_75_cast_fp16")];
            fp16 var_35_promoted_33_to_fp16 = const()[name = string("op_35_promoted_33_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1481_cast_fp16 = pow(x = y_75_cast_fp16, y = var_35_promoted_33_to_fp16)[name = string("op_1481_cast_fp16")];
            tensor<int32, [1]> var_1483_axes_0 = const()[name = string("op_1483_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1483_keep_dims_0 = const()[name = string("op_1483_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1483_cast_fp16 = reduce_mean(axes = var_1483_axes_0, keep_dims = var_1483_keep_dims_0, x = var_1481_cast_fp16)[name = string("op_1483_cast_fp16")];
            fp32 var_1484_epsilon_0 = const()[name = string("op_1484_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1484_cast_fp16 = inverse(epsilon = var_1484_epsilon_0, x = s_73_cast_fp16)[name = string("op_1484_cast_fp16")];
            fp16 var_1485_to_fp16 = const()[name = string("op_1485_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1486_cast_fp16 = mul(x = var_1484_cast_fp16, y = var_1485_to_fp16)[name = string("op_1486_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1487_cast_fp16 = real_div(x = var_1486_cast_fp16, y = s_73_cast_fp16)[name = string("op_1487_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1488_cast_fp16 = add(x = var_1483_cast_fp16, y = var_1487_cast_fp16)[name = string("op_1488_cast_fp16")];
            fp32 var_1489_epsilon_0 = const()[name = string("op_1489_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1489_cast_fp16 = rsqrt(epsilon = var_1489_epsilon_0, x = var_1488_cast_fp16)[name = string("op_1489_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1490_cast_fp16 = mul(x = y_75_cast_fp16, y = var_1489_cast_fp16)[name = string("op_1490_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_6_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62023168)))];
            tensor<fp16, [1, 24, 512]> f_13_cast_fp16 = mul(x = core_backbone_layers_6_post_ffn_norm_weight_to_fp16, y = var_1490_cast_fp16)[name = string("f_13_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_119_cast_fp16 = add(x = x_113_cast_fp16, y = f_13_cast_fp16)[name = string("x_119_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1494_cast_fp16 = abs(x = x_119_cast_fp16)[name = string("op_1494_cast_fp16")];
            tensor<int32, [1]> var_1496_axes_0 = const()[name = string("op_1496_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1496_cast_fp16 = reduce_max(axes = var_1496_axes_0, keep_dims = var_33, x = var_1494_cast_fp16)[name = string("op_1496_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_75_cast_fp16 = maximum(x = var_1496_cast_fp16, y = var_34_to_fp16)[name = string("s_75_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_77_cast_fp16 = real_div(x = x_119_cast_fp16, y = s_75_cast_fp16)[name = string("y_77_cast_fp16")];
            fp16 var_35_promoted_34_to_fp16 = const()[name = string("op_35_promoted_34_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1499_cast_fp16 = pow(x = y_77_cast_fp16, y = var_35_promoted_34_to_fp16)[name = string("op_1499_cast_fp16")];
            tensor<int32, [1]> var_1501_axes_0 = const()[name = string("op_1501_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1501_keep_dims_0 = const()[name = string("op_1501_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1501_cast_fp16 = reduce_mean(axes = var_1501_axes_0, keep_dims = var_1501_keep_dims_0, x = var_1499_cast_fp16)[name = string("op_1501_cast_fp16")];
            fp32 var_1502_epsilon_0 = const()[name = string("op_1502_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1502_cast_fp16 = inverse(epsilon = var_1502_epsilon_0, x = s_75_cast_fp16)[name = string("op_1502_cast_fp16")];
            fp16 var_1503_to_fp16 = const()[name = string("op_1503_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1504_cast_fp16 = mul(x = var_1502_cast_fp16, y = var_1503_to_fp16)[name = string("op_1504_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1505_cast_fp16 = real_div(x = var_1504_cast_fp16, y = s_75_cast_fp16)[name = string("op_1505_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1506_cast_fp16 = add(x = var_1501_cast_fp16, y = var_1505_cast_fp16)[name = string("op_1506_cast_fp16")];
            fp32 var_1507_epsilon_0 = const()[name = string("op_1507_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1507_cast_fp16 = rsqrt(epsilon = var_1507_epsilon_0, x = var_1506_cast_fp16)[name = string("op_1507_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1508_cast_fp16 = mul(x = y_77_cast_fp16, y = var_1507_cast_fp16)[name = string("op_1508_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_7_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62024256)))];
            tensor<fp16, [1, 24, 512]> input_97_cast_fp16 = mul(x = core_backbone_layers_7_operator_norm_weight_to_fp16, y = var_1508_cast_fp16)[name = string("input_97_cast_fp16")];
            tensor<fp16, [1536, 512]> core_backbone_layers_7_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62025344)))];
            tensor<fp16, [1, 24, 1536]> linear_45_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_7_conv_in_proj_weight_to_fp16, x = input_97_cast_fp16)[name = string("linear_45_cast_fp16")];
            tensor<int32, [3]> bcx_perm_0 = const()[name = string("bcx_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_23_begin_0 = const()[name = string("b_23_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_23_end_0 = const()[name = string("b_23_end_0"), val = tensor<int32, [3]>([1, 512, 24])];
            tensor<bool, [3]> b_23_end_mask_0 = const()[name = string("b_23_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1536, 24]> bcx_cast_fp16 = transpose(perm = bcx_perm_0, x = linear_45_cast_fp16)[name = string("transpose_7")];
            tensor<fp16, [1, 512, 24]> b_23_cast_fp16 = slice_by_index(begin = b_23_begin_0, end = b_23_end_0, end_mask = b_23_end_mask_0, x = bcx_cast_fp16)[name = string("b_23_cast_fp16")];
            tensor<int32, [3]> c_begin_0 = const()[name = string("c_begin_0"), val = tensor<int32, [3]>([0, 512, 0])];
            tensor<int32, [3]> c_end_0 = const()[name = string("c_end_0"), val = tensor<int32, [3]>([1, 1024, 24])];
            tensor<bool, [3]> c_end_mask_0 = const()[name = string("c_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 512, 24]> c_cast_fp16 = slice_by_index(begin = c_begin_0, end = c_end_0, end_mask = c_end_mask_0, x = bcx_cast_fp16)[name = string("c_cast_fp16")];
            tensor<int32, [3]> xx_begin_0 = const()[name = string("xx_begin_0"), val = tensor<int32, [3]>([0, 1024, 0])];
            tensor<int32, [3]> xx_end_0 = const()[name = string("xx_end_0"), val = tensor<int32, [3]>([1, 1536, 24])];
            tensor<bool, [3]> xx_end_mask_0 = const()[name = string("xx_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 24]> xx_cast_fp16 = slice_by_index(begin = xx_begin_0, end = xx_end_0, end_mask = xx_end_mask_0, x = bcx_cast_fp16)[name = string("xx_cast_fp16")];
            tensor<fp16, [1, 512, 24]> new_cast_fp16 = mul(x = b_23_cast_fp16, y = xx_cast_fp16)[name = string("new_cast_fp16")];
            tensor<int32, [4]> input_101_begin_0 = const()[name = string("input_101_begin_0"), val = tensor<int32, [4]>([4, 0, 0, 0])];
            tensor<int32, [4]> input_101_end_0 = const()[name = string("input_101_end_0"), val = tensor<int32, [4]>([5, 1, 512, 6])];
            tensor<bool, [4]> input_101_end_mask_0 = const()[name = string("input_101_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_101_squeeze_mask_0 = const()[name = string("input_101_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 512, 6]> input_101_cast_fp16 = slice_by_index(begin = input_101_begin_0, end = input_101_end_0, end_mask = input_101_end_mask_0, squeeze_mask = input_101_squeeze_mask_0, x = conv)[name = string("input_101_cast_fp16")];
            tensor<int32, [6]> var_1523_pad_0 = const()[name = string("op_1523_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 6, 0])];
            string var_1523_mode_0 = const()[name = string("op_1523_mode_0"), val = string("constant")];
            fp16 const_31_to_fp16 = const()[name = string("const_31_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 30]> var_1523_cast_fp16 = pad(constant_val = const_31_to_fp16, mode = var_1523_mode_0, pad = var_1523_pad_0, x = new_cast_fp16)[name = string("op_1523_cast_fp16")];
            tensor<int32, [6]> var_1525_pad_0 = const()[name = string("op_1525_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 24])];
            string var_1525_mode_0 = const()[name = string("op_1525_mode_0"), val = string("constant")];
            fp16 const_32_to_fp16 = const()[name = string("const_32_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 512, 30]> var_1525_cast_fp16 = pad(constant_val = const_32_to_fp16, mode = var_1525_mode_0, pad = var_1525_pad_0, x = input_101_cast_fp16)[name = string("op_1525_cast_fp16")];
            tensor<fp16, [1, 512, 30]> win_17_cast_fp16 = add(x = var_1523_cast_fp16, y = var_1525_cast_fp16)[name = string("win_17_cast_fp16")];
            tensor<int32, [3]> var_1527_begin_0 = const()[name = string("op_1527_begin_0"), val = tensor<int32, [3]>([0, 0, 24])];
            tensor<int32, [3]> var_1527_end_0 = const()[name = string("op_1527_end_0"), val = tensor<int32, [3]>([1, 512, 30])];
            tensor<bool, [3]> var_1527_end_mask_0 = const()[name = string("op_1527_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 512, 6]> var_1527_cast_fp16 = slice_by_index(begin = var_1527_begin_0, end = var_1527_end_0, end_mask = var_1527_end_mask_0, x = win_17_cast_fp16)[name = string("op_1527_cast_fp16")];
            string var_1533_pad_type_0 = const()[name = string("op_1533_pad_type_0"), val = string("valid")];
            int32 var_1533_groups_0 = const()[name = string("op_1533_groups_0"), val = int32(512)];
            tensor<int32, [1]> var_1533_strides_0 = const()[name = string("op_1533_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1533_pad_0 = const()[name = string("op_1533_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1533_dilations_0 = const()[name = string("op_1533_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 7]> core_backbone_layers_7_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_conv_weight_to_fp16"), val = tensor<fp16, [512, 1, 7]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63598272)))];
            tensor<fp16, [1, 512, 24]> var_1533_cast_fp16 = conv(dilations = var_1533_dilations_0, groups = var_1533_groups_0, pad = var_1533_pad_0, pad_type = var_1533_pad_type_0, strides = var_1533_strides_0, weight = core_backbone_layers_7_conv_conv_weight_to_fp16, x = win_17_cast_fp16)[name = string("op_1533_cast_fp16")];
            tensor<fp16, [1, 512, 24]> var_1535_cast_fp16 = mul(x = c_cast_fp16, y = var_1533_cast_fp16)[name = string("op_1535_cast_fp16")];
            tensor<int32, [3]> input_103_perm_0 = const()[name = string("input_103_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> core_backbone_layers_7_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63605504)))];
            tensor<fp16, [1, 24, 512]> input_103_cast_fp16 = transpose(perm = input_103_perm_0, x = var_1535_cast_fp16)[name = string("transpose_6")];
            tensor<fp16, [1, 24, 512]> linear_46_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_7_conv_out_proj_weight_to_fp16, x = input_103_cast_fp16)[name = string("linear_46_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1540_cast_fp16 = abs(x = linear_46_cast_fp16)[name = string("op_1540_cast_fp16")];
            tensor<int32, [1]> var_1542_axes_0 = const()[name = string("op_1542_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1542_cast_fp16 = reduce_max(axes = var_1542_axes_0, keep_dims = var_33, x = var_1540_cast_fp16)[name = string("op_1542_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_77_cast_fp16 = maximum(x = var_1542_cast_fp16, y = var_34_to_fp16)[name = string("s_77_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_81_cast_fp16 = real_div(x = linear_46_cast_fp16, y = s_77_cast_fp16)[name = string("y_81_cast_fp16")];
            fp16 var_35_promoted_35_to_fp16 = const()[name = string("op_35_promoted_35_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1545_cast_fp16 = pow(x = y_81_cast_fp16, y = var_35_promoted_35_to_fp16)[name = string("op_1545_cast_fp16")];
            tensor<int32, [1]> var_1547_axes_0 = const()[name = string("op_1547_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1547_keep_dims_0 = const()[name = string("op_1547_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1547_cast_fp16 = reduce_mean(axes = var_1547_axes_0, keep_dims = var_1547_keep_dims_0, x = var_1545_cast_fp16)[name = string("op_1547_cast_fp16")];
            fp32 var_1548_epsilon_0 = const()[name = string("op_1548_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1548_cast_fp16 = inverse(epsilon = var_1548_epsilon_0, x = s_77_cast_fp16)[name = string("op_1548_cast_fp16")];
            fp16 var_1549_to_fp16 = const()[name = string("op_1549_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1550_cast_fp16 = mul(x = var_1548_cast_fp16, y = var_1549_to_fp16)[name = string("op_1550_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1551_cast_fp16 = real_div(x = var_1550_cast_fp16, y = s_77_cast_fp16)[name = string("op_1551_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1552_cast_fp16 = add(x = var_1547_cast_fp16, y = var_1551_cast_fp16)[name = string("op_1552_cast_fp16")];
            fp32 var_1553_epsilon_0 = const()[name = string("op_1553_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1553_cast_fp16 = rsqrt(epsilon = var_1553_epsilon_0, x = var_1552_cast_fp16)[name = string("op_1553_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1554_cast_fp16 = mul(x = y_81_cast_fp16, y = var_1553_cast_fp16)[name = string("op_1554_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_7_post_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_post_operator_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64129856)))];
            tensor<fp16, [1, 24, 512]> a_cast_fp16 = mul(x = core_backbone_layers_7_post_operator_norm_weight_to_fp16, y = var_1554_cast_fp16)[name = string("a_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_123_cast_fp16 = add(x = x_119_cast_fp16, y = a_cast_fp16)[name = string("x_123_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1558_cast_fp16 = abs(x = x_123_cast_fp16)[name = string("op_1558_cast_fp16")];
            tensor<int32, [1]> var_1560_axes_0 = const()[name = string("op_1560_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1560_cast_fp16 = reduce_max(axes = var_1560_axes_0, keep_dims = var_33, x = var_1558_cast_fp16)[name = string("op_1560_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_79_cast_fp16 = maximum(x = var_1560_cast_fp16, y = var_34_to_fp16)[name = string("s_79_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_83_cast_fp16 = real_div(x = x_123_cast_fp16, y = s_79_cast_fp16)[name = string("y_83_cast_fp16")];
            fp16 var_35_promoted_36_to_fp16 = const()[name = string("op_35_promoted_36_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1563_cast_fp16 = pow(x = y_83_cast_fp16, y = var_35_promoted_36_to_fp16)[name = string("op_1563_cast_fp16")];
            tensor<int32, [1]> var_1565_axes_0 = const()[name = string("op_1565_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1565_keep_dims_0 = const()[name = string("op_1565_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1565_cast_fp16 = reduce_mean(axes = var_1565_axes_0, keep_dims = var_1565_keep_dims_0, x = var_1563_cast_fp16)[name = string("op_1565_cast_fp16")];
            fp32 var_1566_epsilon_0 = const()[name = string("op_1566_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1566_cast_fp16 = inverse(epsilon = var_1566_epsilon_0, x = s_79_cast_fp16)[name = string("op_1566_cast_fp16")];
            fp16 var_1567_to_fp16 = const()[name = string("op_1567_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1568_cast_fp16 = mul(x = var_1566_cast_fp16, y = var_1567_to_fp16)[name = string("op_1568_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1569_cast_fp16 = real_div(x = var_1568_cast_fp16, y = s_79_cast_fp16)[name = string("op_1569_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1570_cast_fp16 = add(x = var_1565_cast_fp16, y = var_1569_cast_fp16)[name = string("op_1570_cast_fp16")];
            fp32 var_1571_epsilon_0 = const()[name = string("op_1571_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1571_cast_fp16 = rsqrt(epsilon = var_1571_epsilon_0, x = var_1570_cast_fp16)[name = string("op_1571_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1572_cast_fp16 = mul(x = y_83_cast_fp16, y = var_1571_cast_fp16)[name = string("op_1572_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_7_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64130944)))];
            tensor<fp16, [1, 24, 512]> input_105_cast_fp16 = mul(x = core_backbone_layers_7_ffn_norm_weight_to_fp16, y = var_1572_cast_fp16)[name = string("input_105_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_7_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64132032)))];
            tensor<fp16, [1, 24, 2304]> linear_47_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w1_weight_to_fp16, x = input_105_cast_fp16)[name = string("linear_47_cast_fp16")];
            tensor<fp16, [2304, 512]> core_backbone_layers_7_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [2304, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66491392)))];
            tensor<fp16, [1, 24, 2304]> linear_48_cast_fp16 = linear(bias = linear_3_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w3_weight_to_fp16, x = input_105_cast_fp16)[name = string("linear_48_cast_fp16")];
            fp16 var_1581_to_fp16 = const()[name = string("op_1581_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 24, 2304]> var_1582_cast_fp16 = mul(x = linear_47_cast_fp16, y = var_1581_to_fp16)[name = string("op_1582_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_1585_cast_fp16 = tanh(x = var_1582_cast_fp16)[name = string("op_1585_cast_fp16")];
            fp16 var_1586_to_fp16 = const()[name = string("op_1586_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 24, 2304]> var_1587_cast_fp16 = add(x = var_1585_cast_fp16, y = var_1586_to_fp16)[name = string("op_1587_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> var_1588_cast_fp16 = mul(x = var_1582_cast_fp16, y = var_1587_cast_fp16)[name = string("op_1588_cast_fp16")];
            tensor<fp16, [1, 24, 2304]> input_cast_fp16 = mul(x = var_1588_cast_fp16, y = linear_48_cast_fp16)[name = string("input_cast_fp16")];
            tensor<fp16, [512, 2304]> core_backbone_layers_7_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [512, 2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68850752)))];
            tensor<fp16, [1, 24, 512]> linear_49_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w2_weight_to_fp16, x = input_cast_fp16)[name = string("linear_49_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1593_cast_fp16 = abs(x = linear_49_cast_fp16)[name = string("op_1593_cast_fp16")];
            tensor<int32, [1]> var_1595_axes_0 = const()[name = string("op_1595_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1595_cast_fp16 = reduce_max(axes = var_1595_axes_0, keep_dims = var_33, x = var_1593_cast_fp16)[name = string("op_1595_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_81_cast_fp16 = maximum(x = var_1595_cast_fp16, y = var_34_to_fp16)[name = string("s_81_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_85_cast_fp16 = real_div(x = linear_49_cast_fp16, y = s_81_cast_fp16)[name = string("y_85_cast_fp16")];
            fp16 var_35_promoted_37_to_fp16 = const()[name = string("op_35_promoted_37_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1598_cast_fp16 = pow(x = y_85_cast_fp16, y = var_35_promoted_37_to_fp16)[name = string("op_1598_cast_fp16")];
            tensor<int32, [1]> var_1600_axes_0 = const()[name = string("op_1600_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1600_keep_dims_0 = const()[name = string("op_1600_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1600_cast_fp16 = reduce_mean(axes = var_1600_axes_0, keep_dims = var_1600_keep_dims_0, x = var_1598_cast_fp16)[name = string("op_1600_cast_fp16")];
            fp32 var_1601_epsilon_0 = const()[name = string("op_1601_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1601_cast_fp16 = inverse(epsilon = var_1601_epsilon_0, x = s_81_cast_fp16)[name = string("op_1601_cast_fp16")];
            fp16 var_1602_to_fp16 = const()[name = string("op_1602_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1603_cast_fp16 = mul(x = var_1601_cast_fp16, y = var_1602_to_fp16)[name = string("op_1603_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1604_cast_fp16 = real_div(x = var_1603_cast_fp16, y = s_81_cast_fp16)[name = string("op_1604_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1605_cast_fp16 = add(x = var_1600_cast_fp16, y = var_1604_cast_fp16)[name = string("op_1605_cast_fp16")];
            fp32 var_1606_epsilon_0 = const()[name = string("op_1606_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1606_cast_fp16 = rsqrt(epsilon = var_1606_epsilon_0, x = var_1605_cast_fp16)[name = string("op_1606_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1607_cast_fp16 = mul(x = y_85_cast_fp16, y = var_1606_cast_fp16)[name = string("op_1607_cast_fp16")];
            tensor<fp16, [512]> core_backbone_layers_7_post_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_post_ffn_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71210112)))];
            tensor<fp16, [1, 24, 512]> f_cast_fp16 = mul(x = core_backbone_layers_7_post_ffn_norm_weight_to_fp16, y = var_1607_cast_fp16)[name = string("f_cast_fp16")];
            tensor<fp16, [1, 24, 512]> x_cast_fp16 = add(x = x_123_cast_fp16, y = f_cast_fp16)[name = string("x_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1611_cast_fp16 = abs(x = x_cast_fp16)[name = string("op_1611_cast_fp16")];
            tensor<int32, [1]> var_1613_axes_0 = const()[name = string("op_1613_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 24, 1]> var_1613_cast_fp16 = reduce_max(axes = var_1613_axes_0, keep_dims = var_33, x = var_1611_cast_fp16)[name = string("op_1613_cast_fp16")];
            tensor<fp16, [1, 24, 1]> s_cast_fp16 = maximum(x = var_1613_cast_fp16, y = var_34_to_fp16)[name = string("s_cast_fp16")];
            tensor<fp16, [1, 24, 512]> y_cast_fp16 = real_div(x = x_cast_fp16, y = s_cast_fp16)[name = string("y_cast_fp16")];
            fp16 var_35_promoted_38_to_fp16 = const()[name = string("op_35_promoted_38_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 24, 512]> var_1616_cast_fp16 = pow(x = y_cast_fp16, y = var_35_promoted_38_to_fp16)[name = string("op_1616_cast_fp16")];
            tensor<int32, [1]> var_1618_axes_0 = const()[name = string("op_1618_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1618_keep_dims_0 = const()[name = string("op_1618_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 24, 1]> var_1618_cast_fp16 = reduce_mean(axes = var_1618_axes_0, keep_dims = var_1618_keep_dims_0, x = var_1616_cast_fp16)[name = string("op_1618_cast_fp16")];
            fp32 var_1619_epsilon_0 = const()[name = string("op_1619_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 24, 1]> var_1619_cast_fp16 = inverse(epsilon = var_1619_epsilon_0, x = s_cast_fp16)[name = string("op_1619_cast_fp16")];
            fp16 var_1620_to_fp16 = const()[name = string("op_1620_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 24, 1]> var_1621_cast_fp16 = mul(x = var_1619_cast_fp16, y = var_1620_to_fp16)[name = string("op_1621_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1622_cast_fp16 = real_div(x = var_1621_cast_fp16, y = s_cast_fp16)[name = string("op_1622_cast_fp16")];
            tensor<fp16, [1, 24, 1]> var_1623_cast_fp16 = add(x = var_1618_cast_fp16, y = var_1622_cast_fp16)[name = string("op_1623_cast_fp16")];
            fp32 var_1624_epsilon_0 = const()[name = string("op_1624_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 24, 1]> var_1624_cast_fp16 = rsqrt(epsilon = var_1624_epsilon_0, x = var_1623_cast_fp16)[name = string("op_1624_cast_fp16")];
            tensor<fp16, [1, 24, 512]> var_1625_cast_fp16 = mul(x = y_cast_fp16, y = var_1624_cast_fp16)[name = string("op_1625_cast_fp16")];
            tensor<fp16, [512]> core_backbone_embedding_norm_weight_to_fp16 = const()[name = string("core_backbone_embedding_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71211200)))];
            tensor<fp16, [1, 24, 512]> hidden = mul(x = core_backbone_embedding_norm_weight_to_fp16, y = var_1625_cast_fp16)[name = string("op_1626_cast_fp16")];
            int32 var_1628_axis_0 = const()[name = string("op_1628_axis_0"), val = int32(0)];
            tensor<fp16, [5, 1, 512, 6]> conv_out = stack(axis = var_1628_axis_0, values = (var_429_cast_fp16, var_546_cast_fp16, var_873_cast_fp16, var_1200_cast_fp16, var_1527_cast_fp16))[name = string("op_1628_cast_fp16")];
            int32 var_1630_axis_0 = const()[name = string("op_1630_axis_0"), val = int32(0)];
            tensor<fp16, [3, 1, 2, 29, 64]> k_out = stack(axis = var_1630_axis_0, values = (var_764_cast_fp16, var_1091_cast_fp16, var_1418_cast_fp16))[name = string("op_1630_cast_fp16")];
            int32 var_1632_axis_0 = const()[name = string("op_1632_axis_0"), val = int32(0)];
            tensor<fp16, [3, 1, 2, 29, 64]> v_out = stack(axis = var_1632_axis_0, values = (var_767_cast_fp16, var_1094_cast_fp16, var_1421_cast_fp16))[name = string("op_1632_cast_fp16")];
        } -> (hidden, conv_out, k_out, v_out);
}