update ray dependency to <2.0.0

[mattip]

ray 1.9.1 was released last week, and with it windows support has moved from "alpha" to "beta" quality. It would help the ray project if more people could try out the newer version. spacy-ray pins the maximum ray version to ray<1.0.0. Are there known problems preventing bumping that up to ray<=1.9.1 ?

[klimpt]

The mentioned dependency requirement ray<1.0.0 currently prevents using python>=3.9, which is a real obstacle for using spacy-ray.

[adrianeboyd]

We'd be interested to hear how you're currently using spacy-ray? It's currently definitely more on the end of work-in-progress than a stable integration, and we haven't had a lot of time to work on it recently (as you clearly see). I know that there are some major issues, like saving the trained model isn't integrated correctly as far as I remember.

The last time I tested it (six months ago or so?), spacy-ray definitely did not work out-of-the-box with ray>=1.0.0 and it looked like it was going to take some effort to update it.

[klimpt]

I was just evaluating it and tried some benchmarks for the speed-up (and had to downgrade in order to run them). You're right, that model-saving is not working and should be fixed first. I thought the dependency requirement could be fixed with a reasonable small investment, but I understand this is not the case. Thanks for the quick response.