Changed the model size calculation using ByteCounter

[rohitgr7]

What does this PR do?

Fixes #10074
all credits to @ruro :)
Tests are already there for get_model_size_mb

Does your PR introduce any breaking changes? If yes, please list them.

Before submitting

• Was this discussed/approved via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests? (not for typos and docs)
• Did you verify new and existing tests pass locally with your changes?
• Did you list all the breaking changes introduced by this pull request?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is welcome to review the PR.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:

• Is this pull request ready for review? (if not, please submit in draft mode)
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory and the description concisely explains the PR
• Add labels and milestones (and optionally projects) to the PR so it can be classified

Did you have fun?

Make sure you had fun coding 🙃

[tchaton]

Hey @rohitgr7,

I believe we should also have a version for DeepSpeed FSDP which reduces the number of bytes across processes.

Best,
T.C

[rohitgr7]

Hey @rohitgr7,

I believe we should also have a version for DeepSpeed FSDP which reduces the number of bytes across processes.

Best, T.C

will have a look. maybe in a separate one?

[rohitgr7]

@tchaton couldn't find a way to support this for Deepspeed/FSDP. Discussed this with @SeanNaren too and seems like its right not practically possible since weights are sharded and are not kept in a readable format. I'd suggest for now we can merge this for v1.5 and later we can explore more.

[kaushikb11]

LGTM! Could you add it to the class that it's not supported by DeepSpeed and Sharded plugins.

[tchaton]

LGTM !

[codecov]

Codecov Report

Merging #10123 (16d6c9a) into master (a1fe5ba) will decrease coverage by 1%.
The diff coverage is 100%.

❗ Current head 16d6c9a differs from pull request most recent head 249fd8e. Consider uploading reports for the commit 249fd8e to get more accurate results

@@           Coverage Diff            @@
##           master   #10123    +/-   ##
========================================
- Coverage      92%      92%    -1%     
========================================
  Files         181      182     +1     
  Lines       16408    17108   +700     
========================================
+ Hits        15175    15700   +525     
- Misses       1233     1408   +175     

[SeanNaren]

Looks good! can we add an issue around FSDP/DeepSpeed support for summarization of model size? As @tchaton mentioned it is possible, just will require some thought/implementation