Unlimited-OCR-2: Baidu open-sources an OCR model for unlimited-horizon parsing
AI model · https://huggingface.co/baidu/Unlimited-OCR-2
First seen by Oriel on .
AI summary
Unlimited-OCR-2 uses Sparse Reference Sliding Window Attention to parse ultra-long documents while keeping per-step decoding latency and memory bounded as document length grows.
- SR-SWA compresses text-visual attention interactions into a fixed-size working reference set, with 512 tokens given as an example.
- The model supports multi-page inference in one request, preserving page order and enabling cross-page context.
- The page reports state-of-the-art results on MPDocBench and the hundred-page MarathonDocBench, without providing scores in the supplied text.
- Inference is supported through Transformers for quick tests and SGLang for serving; SGLang settings trade memory use against cross-page merging across page windows.
- The documented setup includes an NVIDIA A800-80GB test configuration and memory guidance for serving concurrent requests.
#ocr #document parsing #attention #sglang
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hacker News: 7, latest observation ; +3 over observed period since
Likes
Total likes reported by the source.
- Hugging Face Models: 668, latest observation ; +542 over observed period since
Downloads (30d)
Rolling 30-day downloads. A change in this value is not new downloads during the observed period.
- Hugging Face Models: 428,120, latest observation ; +427,607 over observed period since
Comments
Comment count reported by the source.
- Hacker News: 0, latest observation ; 0 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Models | baidu/Unlimited-OCR-2 | https://huggingface.co/baidu/Unlimited-OCR-2 Read discussion |
| Hacker News | 50259228 | https://huggingface.co/baidu/Unlimited-OCR-2 Read discussion |