See [link](https://www.paddleocr.ai/main/version3.x/inference_deployment/local_inference/high_performance_inference.html) for details.