Oct 22, 2025 17:14 Add prompt mode presets and remove test script davanstrien +1: Claude HF Staff Changes:
- Added PROMPT_MODES dict with 5 presets: document, image, free, figure, describe
- Added --prompt-mode argument (default: document)
- Custom --prompt still supported (overrides prompt-mode)
- Updated help text and examples to show prompt modes
- Removed deepseek-ocr-vllm-test.py (no longer needed after successful testing)
Prompt modes from DeepSeek-OCR GitHub:
- document: Convert document to markdown with grounding (default)
- image: OCR any image with grounding
- free: Free OCR without layout preservation
- figure: Parse figures from documents
- describe: Generate detailed image descriptions
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com> Jul 31, 2025 16:56 Optimize default settings based on performance testing - Increase default batch size from 8 to 32
- Increase default GPU memory utilization from 0.7 to 0.8
- Update README with new defaults and simple performance tip
- These changes provide ~2-3x speedup based on testing 
