How to correctly set up 2 stage inference?

[ryx2]

Please provide the following information to quickly locate the problem

• 系统环境/System Environment：Debian 11
• 版本号/Version：Paddle：paddleocr==2.7.0.3 PaddleOCR：paddlepaddle-gpu==2.6.0 问题相关组件/Related components：
• 运行指令/Command Code：
  ocr=PaddleOCR(det_model_dir=X, rec_model_dir=X, cls_model_dir=../files/paddle_models/cls/ch_ppocr_mobile_v2.0_cls_infer/) ... ocr.ocr(im, cls=True)
• 完整报错/Complete Error Message：

ValueError: (InvalidArgument) Broadcast dimension mismatch. Operands could not be broadcast together with the shape of X = [960, 6, 512] and the shape of Y = [256, 1, 512]. Received [960] in X is not equal to [256] in Y at i:0.
  [Hint: Expected x_dims_array[i] == y_dims_array[i] || x_dims_array[i] <= 1 || y_dims_array[i] <= 1 == true, but received x_dims_array[i] == y_dims_array[i] || x_dims_array[i] <= 1 || y_dims_array[i] <= 1:0 != true:1.] (at /paddle/paddle/phi/kernels/funcs/common_shape.h:86)
  [operator < elementwise_add > error]

I have been using the tools/export_model.py to create the inference models, for an example like this:
python3 tools/export_model.py -c configs/det/det_r50_db++_icdar15.yml -o Global.pretrained_model=/home/raymondxu/Downloads/det_r50_db++_icdar15_train/best_accuracy Global.save_inference_dir=/home/raymondxu/Downloads/exp_det_dbpp, and similarly for the rec, I pick another model and export.
However, no matter which 2 models i pick for det and rec, I get some sort of shape error.

currently, all examples are for pp-ocr, or for det and rec models being run individually. But, I am running this on a server, so I would like larger models for each to be run.
Question: How can I correctly integrate larger detection and recognition models in PaddleOCR for server use without encountering shape mismatch errors? Are there specific models or configurations that are known to work together seamlessly?