Add spaCy

[Puru35]

Hello!
I'm working on a python nlp project that utilizes the spaCy library, and I need it to work directly on the browser, and using "micropip" to install the required packages isn't cutting it. Is there any possible way to make this happen?

[rth]

There was an attempt to add spaCy in #368. Someone would need to continue that PR and address remaining issues.

[aghasemi]

Hi. Is this now part of version 0.21, as it seems to have been merged? (I guess not, no?)

[shafferchance]

That was for exodide, it seems if you can use 0.21 it would work with it. Going to attempt to follow the documentation on building wheels for packages out of tree and see if that works.

[shafferchance]

Managed to get it working, had to do some flips and unfortunately build cython-blis as generic. Essentially you will need to build the wheels individually and load them in-order but it does seem to be working. Not sure about performance at the moment though.

If you don't use BLIS_ARCH=generic when building the wheel for cython-blis you will get an error like this:

unknown FP unit 'sse'

This is because sse is not fully supported for WASM yet. Thus when attempting to use those headers it fails to resolve. This might change in the future, but it would be a while before production is ready.

[shafferchance]

For a full deps list:

1. cymem-2.0.7
2. murmurhash-1.0.9
3. preshed-3.0.8
4. srsly-2.4.6
5. blis-0.7.9 (This is cython-blis)
6. thinc-8.1.10
7. spacy-3.6.0

All of these are under the explosion organization and should be pretty usable for any of us.

[Vuizur]

This is maybe also relevant, someone published wheels and a demo: https://github.com/SyedAhkam/spacy-wasm