- Add ProgressTracker class for real-time progress parsing from mineru stdout - Implement intelligent stall-based timeout (5min no progress triggers warning) - Add vlm-direct backend mode to directly connect to vLLM server - Improve API mode with streaming response, HTTP session reuse, and auto-retry - Add heartbeat updates during long-running operations - Support dynamic timeout based on page count This addresses the deadlock issue where tasks hang indefinitely when using MinerU API backend, even after the API returns 200 OK. |
||
|---|---|---|
| .. | ||
| app | ||
| flow | ||
| llm | ||
| nlp | ||
| prompts | ||
| res | ||
| svr | ||
| utils | ||
| __init__.py | ||
| benchmark.py | ||
| raptor.py | ||
| settings.py | ||