GitHub radar
Parse a document of any length in one pass, no manual splitting
Unlimited-OCR-2 is a Baidu model that reads text from document images (OCR), parsing a file of any length in a single pass while keeping context across pages — no need to cut the document into chunks and stitch the results back together.
The model reads text off document pages and carries understanding of the content from one page to the next, so the whole file is parsed as one piece instead of in chunks.
When it helps
Useful if you regularly work with long scans — books, reports hundreds of pages long, PDFs. Not meant for a single photo or a short file — the whole benefit here is reading straight through many pages at once.
Pros
- Reads an entire document in one pass, no manual stitching of parts
- Speed and memory barely grow as the document gets longer
- Keeps context across pages — doesn't lose links between phrases
- MIT license — free to use and modify
Cons
- Only the author's own benchmarks — not an independent check
- Unclear if handwriting and tables were actually tested
- Needs your own GPU or server — not a hosted service
How to set it up — step by step
- 1Open the model's card at the link from the facts on Hugging Face.
- 2Send the agent the text from the block below — it will open the demo and upload the document.
- 3Drop your own multi-page document or book into the demo and run the parse.
- 4Check the result page by page against the original — especially tables and messy handwriting.
- 5For a local run via Transformers, a laptop with 8GB+ memory may be enough (an estimate, not a vendor guarantee).
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Open the Unlimited-OCR-2 demo on Hugging Face Spaces (https://huggingface.co/spaces/baidu/Unlimited-OCR-2), upload a multi-page document or book, and show how the model parses it page by page.
A regular laptop with about 8 GB of RAM (an estimate based on model size, not stated in the card).
Open on Hugging Face▌ More finds