Add support for fastino gliner2 models?

[codesoda]

Are you open to adding support for the latest fastino gliner2 models from https://huggingface.co/collections/fastino/gliner2-family?

[fbilhaut]

Hi @codesoda,

Absolutely! I still have to get deeper in the paper and code, but from what I've seen so far, it seems doable in a reasonable timeframe. Would you like to contribute in any way?

[codesoda]

Yes, absolutely. I went through the paper and have made a list of things to be done. Might make sense to just focus on one of token or span modes initially. Even just the simplest of entity extraction as a first pass, then move onto the other extended capabilities.

[fbilhaut]

Hi @codesoda, sorry for the late reply, I had no time recently to work on this project. Still interested in collaborating on this, though. Let me know if I can do anything

[codesoda]

I have it mostly working, I'll try to push it up during the next week.

[fbilhaut]

Hi @codesoda, is it still in your plans to contribute on this? If needed, I should be able to make some time for it. Best

[youssef-tharwat]

Hi @codesoda, is it still in your plans to contribute on this? If needed, I should be able to make some time for it. Best

Looks like not

[lmoe]

@fbilhaut Hey, maybe I can help contributing. I was also curious about Gliner2 as a potential replacement for the mdeberta zero-shot I was using.

I tried to export the Gliner2 large and multi models to ONNX with success: https://github.com/lmoe/gliner2-onnx. It contains a completely LLM generated ARCHITECTURE.md that I didn't proof read, as I just wanted to get this done. 😅
But maybe it helps. Models are available on huggingface to test.

That being said, I hope that fastino just ships a proper ONNX setup natively because this was quite a lot of trial and error for me to get done and I'm sure it's not bug free. However I ran 200 example queries against the PyTorch Gliner2 library and mine, and they all came out equally.

[fbilhaut]

Hi @lmoe, this could indeed be of great help, thank you very much ! Will definitely look into it as soon as possible.