PaddleOCRv4 export to nb got core dumped while PaddleOCRv3 is ok

[Houangnt]

[GreatV]

It looks like you're experiencing a core dump issue when exporting a model in PaddleOCR v4, while it works fine in PaddleOCR v3. Based on a related discussion (paddleocr export_model failed!), here are some potential causes and solutions:

Possible Causes:

1. Version Mismatch – PaddleOCR v4 may have changes that introduce incompatibilities with your environment.
2. GPU Memory Usage – Since you are using RTX 3090 x2, ensure that sufficient memory is available.
3. PaddlePaddle Version – Verify that the installed PaddlePaddle version is compatible with PaddleOCR v4.
4. CUDA & cuDNN Compatibility – Mismatches between these libraries can result in crashes during model export.

Suggested Solutions:

1. Check Dependencies

   • Run:

     python -c "import paddle; print(paddle.__version__)"

     Ensure you have the latest PaddlePaddle version required for PaddleOCR v4.

2. Reinstall Dependencies
   Try reinstalling PaddleOCR and PaddlePaddle:

   pip uninstall paddlepaddle paddleocr -y
   pip install paddlepaddle-gpu==<compatible_version> -i https://mirror.baidu.com/pypi/simple
   pip install paddleocr

3. Reduce Batch Size During Export
   If GPU memory is an issue, try a lower batch size when exporting:

   python tools/export_model.py --configs path/to/config.yml --batch_size 1

4. Check CUDA/cuDNN Compatibility
   Ensure CUDA and cuDNN versions are aligned with your PaddlePaddle version:

   nvidia-smi
   nvcc --version

5. Debug with CPU Mode
   If GPU export fails, test exporting on the CPU:

   python tools/export_model.py --device cpu --configs path/to/config.yml

If the issue persists, you can post your error logs, CUDA version, Paddle version, and GPU usage for further debugging. You may also check for updates on the related discussion: paddleocr export_model failed!.

---

Response generated by 🤖 feifei-bot | chatgpt-4o-latest

⚠️ This is an AI-generated response. Please verify any critical information.