LibreFalconOCR270m

Falcon OCR (0.27B early-fusion document OCR model, version 1.5) weights mirrored for LibreYOLO's ocr task.

Source

Mirrored from tiiuae/Falcon-OCR at commit 0bc28f89c90fbe8bdb063097fab144d4ce4c19b5 (code, paper arXiv:2603.27365). Copyright 2026 Technology Innovation Institute (TII), UAE. Licensed under the Apache License 2.0.

Modifications

No learned parameters were changed. model.safetensors, config.json, model_args.json and the tokenizer files are byte-identical to the upstream snapshot. The upstream remote-code Python files, assets and evaluation files are not included: LibreYOLO runs its own PyTorch port of the architecture and never executes remote code.

Usage

from libreyolo import LibreVLM

model = LibreVLM("falcon-ocr")                # whole page in one pass
result = model.predict("page.png")
print(result.ocr.markdown)                    # page text, LaTeX formulas, HTML tables

regions = LibreVLM("falcon-ocr", layout=True).predict("page.png")
print(regions.ocr.labels, regions.ocr.texts)  # PP-DocLayoutV3 regions, read one by one

License

Apache License 2.0. See LICENSE and NOTICE.