Comment by progval
It's not open-source (nor open-weight): https://huggingface.co/nanonets/Nanonets-OCR-s/discussions/2
It's not open-source (nor open-weight): https://huggingface.co/nanonets/Nanonets-OCR-s/discussions/2
IMO weights being downloadable doesn't mean it's open weight.
My understanding:
- Weight available: You can download the weights.
- Open weight: You can download the weights, and it is licensed freely (e.g. public domain, CC BY-SA, MIT).
- Open source: (Debated) You can download the weights, it is licensed freely, and the training dataset is also available and licensed freely.
> You're right. The Apache-2.0 license was mistakenly listed, and I apologize for the confusion. Since it's a derivative of Qwen-2.5-VL-3B, it will have the same license as the base model (Qwen RESEARCH LICENSE AGREEMENT). Thanks for pointing this out.
Interestingly, another OCR model based on Qwen2.5-VL-3B just dropped which also publishes as Apache 2. It's right next to Nanonets-OCR-s on the HF "Trending" list.
https://huggingface.co/echo840/MonkeyOCR/blob/main/Recogniti...
Hi, author of the model here. It is an open-weight model, you can download it from here: https://huggingface.co/nanonets/Nanonets-OCR-s