Add InfoLM

[stancld]

What does this PR do?

Fixes #849

Before submitting

• Was this discussed/approved via a Github issue? (no need for typos and docs improvements)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure to update the docs?
• Did you write any new necessary tests?

I've prepared a personal repository with scripts/instructions for reproducing test results -> will be also useful for #852 and #854 if the original package installation will not be available.

PR review

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

Did you have fun?

Make sure you had fun coding 🙃

[codecov]

Codecov Report

Merging #915 (5a8f6f0) into master (c7b7c46) will decrease coverage by 0%.
The diff coverage is 94%.

@@          Coverage Diff           @@
##           master   #915    +/-   ##
======================================
- Coverage      94%    94%    -0%     
======================================
  Files         182    185     +3     
  Lines        8147   8389   +242     
======================================
+ Hits         7654   7880   +226     
- Misses        493    509    +16     

[stancld]

@justusschock @SkafteNicki Thanks a lot for your reviews, I'll resolve the remaining issues tomorrow.

@Borda Not sure how to fix the issue with those daemonic processes having children -- gonna probe this in more depth tomorrow as well :]

[SkafteNicki]

Hi @stancld,
I did a bit of digging and it seem the problem is using multiple number of workers in the dataloader, as this will try to spawn multiple processes within each ddp process (they are spawned for testing) which is not allowed. It seems to be working for BERTScore so I am not sure why this is not working for this metric

[stancld]

@Borda Mind have a look at some of your previous comments if they're now settled? :]

[stancld]

Hi @stancld, I did a bit of digging and it seem the problem is using multiple number of workers in the dataloader, as this will try to spawn multiple processes within each ddp process (they are spawned for testing) which is not allowed. It seems to be working for BERTScore so I am not sure why this is not working for this metric

I think the problem arises when dist_sync_on_step=True, and I suspect this configuration is not tested for BERTScore as it uses a non-standard test suite as suggested in #1110 -> gonna fix bertscore testing once this merged :]

[Borda]

@stancld good with me 🎉 how about the very last comments?
cc: @justusschock @SkafteNicki 🐰

[SkafteNicki]

LGTM

[Borda]

@stancld mind checking the remaining comments and mar as resolved if it so... then it ll be merged 🎉