Pytorch Lightning 2 compatibility

[percevalw]

Going a bit further than #15, this PR relaxes all the upper bounds of the requirements:

• python: keep >=3.6 (but most projects use python>=3.7 anyway)
• einops: from ~=0.3.0 to >=0.3.0
• tqdm: from ~=4.56.0 to >=4.56.0
• pytorch_lightning: from >=1.1.7,<=1.4.9 to >=1.1.7
• torchmetrics: from ==0.7.3 to >=0.7.3
• rich_logger: from >=0.1.4 to >=0.3.1
• sentencepiece: from ~=0.1.95 to >=0.1.95
• xxhash: from ~=2.0.0 to >=2.0.0
• regex: from ~=2020.11.13 to >=2020.11.13
• parse: from ~=1.19.0 to >=1.19.0

and ensures that models can still be trained under pytorch-ligthning==1.* or pytorch-ligthning==2.*.

I've tested this under python 3.7, 3.8 and 3.10, with oldest & latest requirements.

Additionally, from rich-logger>=0.3.1, nested tqdm progress bars are rendered nicely with rich, so I added them back.

@camila-ud do you want to pull & check if this works for you ?

[camila-ud]

Hello @percevalw, thanks for this update!

I've tested the integration with medkit, and it seems to work fine.
Would it be possible to release a new version of nlstruct including this PR ?

[percevalw]

Yes ofc ! I was waiting for your feedback, I'll do it asap

[ghisvail]

Great to hear this is moving forward. Thank you- 🙏

As a side note:

python: keep >=3.6 (but most projects use python>=3.7 anyway)

Not really. Versions 3.6 and 3.7 are no longer supported and are being progressively dropped by libraries.

[ghisvail]

@camila-ud we would also need #17, similar situation as with iamsystem sadly.

[camila-ud]

Yes, indeed. The license should also be updated in order to be compatible with more projects.

I would be grateful if you could include the modifications from #17 in the release.

I am just finishing the details of the operation, but I would like to integrate it soon.
The nested entities are very useful ;)

[percevalw]

It depends. Indeed, py37 is no longer supported, but some cluster environments have incompatibilities with py38 and above (for example spark 2 clusters). I prefer to support these versions as long as there are no major incompatibilities. For reference, here are the pip package download statistics by python version (excluding universal/source package downloads) over the last month, with py37 in second place!

Version	Counts
3.1	3865
3.2	1971
3.3	2058507
3.4	8053025
3.5	36266512
3.6	1181840918
3.7	5582678149
3.8	6151245728
3.9	5534895827
3.10	5386164720
3.11	2383864971
3.12	468044346
3.13	1380700

[percevalw]

@ghisvail @camila-ud the new version 0.2.0 has just been released, thank you both for your contributions !

[ghisvail]

Awesome, thanks for addressing our issues so quickly.

[percevalw]

My pleasure ! Consider leaving a star if you end up finding it useful in your work