GitLab

- Refactor VRAM detection logic for better readability and efficiency
- Add fallback mechanism for unknown VRAM sizes
- Improve device checking in get_vram function

- Remove PIL usage across multiple files
- Convert image processing functions to use NumPy arrays
- Update crop_img function to work with NumPy arrays
- Modify image loading and resizing to use NumPy and OpenCV
- Clean up unused imports and comments related to PIL

- Update `clean_memory.py` to use `torch_npu.npu` instead of `torch.npu`
- Update `model_utils.py` to use `torch_npu.npu` instead of `torch.npu`
- Simplify NPU availability check and bfloat16 support in `pdf_parse_union_core_v2.py`

- Update clean_memory function to support both CUDA and NPU devices
- Implement get_device function to centralize device selection logic
- Modify model initialization and memory cleaning to use the selected device
- Update RapidTableModel to support both RapidOCR and PaddleOCR engines

- Add NPU support for memory cleaning and model initialization
- Optimize table model initialization and prediction process
- Update memory utils to support NPU
- Add language parameter for table model

- Update VRAM checking logic in app.py and model_utils.py
- Add None and type checks for VRAM values
- Adjust concurrency limit calculation in app.py
- Modify clean_vram function to handle cases with no VRAM information

- Add get_concurrency_limit function to calculate concurrency limit based on VRAM
- Update clean_vram function and rename to get_vram for better clarity
- Apply concurrency limit to the to_markdown function in the Gradio app