Library / SDK
Retrieval & Memory
olmOCR
1Converts 2scanned 3PDFs 4into 5clean 6text
the six Ws · specification
W1
Who
The Allen Institute for AI, Ai2, released olmOCR.
W2
What
olmOCR is a document parsing toolkit built around a fine-tuned vision-language model that converts scanned PDFs and images into clean, structured plain text or markdown.
W3
Where
Hosted on GitHub at allenai/olmocr and on Hugging Face under allenai.
W4
When
Released in February 2025.
W5
Why
It was built to accurately and cheaply convert large PDF corpora into training-ready text for language model pretraining.
W6
With
Runs on the olmOCR-7B vision-language model served via vLLM or SGLang.
W7
Watch
Released under the Apache 2.0 license by the Allen Institute for AI; fully open weights and code with no gating.