Update thinc dependency + maybe some extras

[chriss2401]

Currently DaCy references thinc v8.0.13 via spacy-wrap. This uses pydantic v1, which can be a pain for downstream users. Current latest is v9.1.1

Some additional dependency upgrades worth considering:

1. Spacy 3.2.1 -> Can be updated to >3.8.2 (v4 is also coming soon it seems)

• Spacy in v3.8.2 introduced better options for memory management, so this would be a benefit of simply updating and nudging users to use this feature. Check here. This will also enable numpy v2.0 support.

Other smaller updates can be made (i.e. tqdm, spacy-experimental), although they are definitely not crucial and I have not checked whether their updates offer something significant.

[KennethEnevoldsen]

So should be fine, though do note that the dacy pipelines has a spacy dependency >=3.5.2,<3.6.0

https://huggingface.co/chcaa/da_dacy_large_trf

I am actually unsure if this dependency is included in the wheel. You would have to unpack it to see. I suspect they can work with newer dependencies.

---

Other than that I think it looks reasonable.

Currently, Dacy installs from HuggingFace when downloading a model (which can be a bit weird because calling Dacy load can change your dependencies. I did consider just adding the models as optional dependencies, e.g. like:

pip install dacy[large]

[chriss2401]

After downloading and unpacking the small model , I can see that indeed it has the same dependency:

But as you say it might work with newer v3+ dependencies as well. Should I just update the dependencies and start testing it with my current project?

[KennethEnevoldsen]

yes - if it works fine we can update the dependencies on HF