githublsk

@jcwchen 谢谢您，我在onnx-extensions中做了如下修改：  但是还是报如下错误，赶着上线，麻烦您帮忙解决一下，真是万分感谢 

@wolfshow Oh,yes, do you have a plan to support multi-language like layoutxlm for layoutmv3? because layoutxlm model is a little heavy

@[wolfshow](https://github.com/wolfshow) Thank you for your answer, I have a question that can I use LayoutLMv3 English model for finetuning my Chinese task?

> @githublsk The RE model is often trained with keys and values with ground truth labels. So, basically, you need two models to finish the SER and RE tasks. Thank...

> 可以详细说明一下问题吗？ PaddleOCRv4 server端文字检测模型转ONNX准确率并未下降，ONNX转TRT后精度下降70%左右，可以参照之前有几个开发者提出了同样问题： https://github.com/PaddlePaddle/PaddleOCR/issues/10917 https://github.com/PaddlePaddle/PaddleOCR/issues/11419 麻烦帮忙解答一下，非常感谢 转换的命令： 

> 可以详细说明一下问题吗？ 能否帮忙解答一下？多谢

> 可以尝试使用Fastdeploy进行推理，切换后端为trt： > > ``` > python infer.py --det_model ch_PP-OCRv4_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv4_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device gpu --backend trt > ``` > > 参考：https://github.com/PaddlePaddle/FastDeploy/tree/develop/examples/vision/ocr/PP-OCR/cpu-gpu/python > > 但需要注意，目前没有在FP16上做过精度验证，确实可能存在精度损失的情况，建议先使用FP32。...

> thanks in advance Yes, just for two times