CUDA memory profile

[erhoo82]

What does this PR do ?

Profile CUDA memory breakdown. The reference is available in the below link.
https://pytorch.org/blog/understanding-gpu-memory-1/

The memory profile will generate a pickle file and a user can load the file in the below page.
https://pytorch.org/memory_viz

Changelog

• Added the option to profile a range of training steps to profile the memory usage
• Adjust the training step by the number of micro-batches steps per global step.

Usage

Add the below knobs to the yaml run config.

  memory_profile:                                                                      
     enabled: true                                                                      
     start_step: 2                                                                      
     end_step: 5                                                                        
     rank: 0                                                                            
     output_path: <path/to/out_file>

GitHub Actions CI

The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.

The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR.
To re-run CI remove and add the label again.
To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

• New Feature
• Bugfix
• Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.

Additional Information

• Related to # (issue)

[akoumpa]

Hi @erhoo82,

I'm going to approve this as is, with one suggestion:

I searched for hasattr(self, '_memory_profile_enabled') and found 11 occurrences. Something like self.has_memory_profile_enabled() would result in cleaner code, for example, if the has_memory_profile_enabled was defined in modelPT.py.
However, that's not a hard requirement as the hasattr is widely used in NeMo.

Thank you so much, this is helpful feature to have in NeMo.