sparseml integration

[mathemusician]

Closes #196

Add option to use sparseml. This is still a little rough around the edges, but I welcome the feedback! Implementation on Google Colab.

Still needs:

1. Implementation of MODELS_PATH and RECIPE_PATH through Hydra instead of environment variable
2. "test" stage needs to work, only training and evaluation work at the moment
3. stylistic coding changes

Will continue to add more changes after my reactions engineering exam tomorrow. Feel free to give me pointers or links to tutorials if you see any changes I need to make.

[codecov]

Codecov Report

Merging #197 (2788bea) into master (6c9b495) will decrease coverage by 2.94%.
The diff coverage is 49.57%.

@@            Coverage Diff             @@
##           master     #197      +/-   ##
==========================================
- Coverage   90.21%   87.27%   -2.95%     
==========================================
  Files          71       73       +2     
  Lines        1543     1658     +115     
==========================================
+ Hits         1392     1447      +55     
- Misses        151      211      +60     

Flag	Coverage Δ
unittests	87.27% <49.57%> (-2.95%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
lightning_transformers/core/callback.py	28.57% <25.67%> (-12.61%)	⬇️
lightning_transformers/core/loggers.py	87.50% <87.50%> (ø)
...transformers/task/nlp/text_classification/model.py	97.61% <87.50%> (-2.39%)	⬇️
lightning_transformers/utilities/imports.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6c9b495...2788bea. Read the comment docs.

[mathemusician]

Wow, that's a lot of red. I think I may be able to get rid of the dependency to pytorch-lightning-bolts, which isn't even listed as a dependency yet. Also, @SeanNaren I've tested sparseml with torch 1.9, and it works!? I'll do a more thorough test after my test tomorrow and add the appropriate changes.

[mathemusician]

The logger and callback tests need certain imports that aren't in requirements.txt. But I wasn't sure if we wanted to add more import statements. The logger test needs wandb; and the sparseml callback test needs pytorch 1.7-1.8.

[mathemusician]

Should I add wandb to requirements.txt? Also, I'm going to update the docs with the different ways to use sparseml soon

[SeanNaren]

Thanks for your hard work @mathemusician hopefully tests should pass now, and we can merge :)