OCR and documents · Nanonets
Nanonets OCR2 3B
Qwen2.5-VL-3B tuned to emit Markdown, tables and LaTeX.
Datasheet
Maker | Nanonets |
|---|---|
Parameters | 3.76 BVerifiedsafetensors.total from the HF API Source |
Licence | Not declared on Hugging FaceVerifiedRestricts commercial use or is not declared. Read the licence before you build on it. |
Access | Public download |
Task (HF pipeline) | image-text-to-text |
First published on HF | 2025-10-13 |
Last updated on HF | 2025-10-16 |
Weights in memory | 7.5 GB at 16-bit · 2.3 GB at ~4.85 bitsEstimateParameters × bits ÷ 8. Excludes context cache and runtime overhead; not every model has a 4-bit build. |
Benchmark
69.5%
olmOCR-Bench, overall · #4 of 4 scored in this category
Nanonets-OCR2-3B row (±1.1). Source
The description summarises the model card at https://huggingface.co/nanonets/Nanonets-OCR2-3B. Parameters and licence are read from the Hugging Face API; the weights-in-memory line is our arithmetic, not a measurement.