Remove requirement to be in training mode to capture activations

[lucmos]

Types of changes

• [x] Bug fix (non-breaking change which fixes an issue)
• [ ] New feature (non-breaking change which adds functionality)
• [ ] Breaking change (fix or feature that would cause existing functionality to change)
• [ ] Docs change / refactoring / dependency upgrade

Motivation and Context / Related issue

In some contexts it is necessary to compute gradients during validations and/or testing.

However, there is currently an explicit check that -- even if the gradients are manually enabled -- activations are captured solely during training.

This leads to the folllowing error:

How Has This Been Tested (if it applies)

Manual testing on local project.

Checklist

• [x] The documentation is up-to-date with the changes I made.
• [x] I have read the CONTRIBUTING document and completed the CLA (see CONTRIBUTING).
• [ ] All tests passed, and additional code has been covered with new tests.

[facebook-github-bot]

Hi @lucmos!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks!

[facebook-github-bot]

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks!

[HuanyuZhang]

Thanks for the contribution to Opacus. Just one question, it is not clear to me when per-sample gradient is needed beyond training. Could you provide some examples? Thanks!

[lucmos]

Thanks for you work on Opacus! 🍻

I think they may be needed whenever the gradients are enabled. Why would one need batch-gradients but not sample-gradients?

As a specific example atm I can only provide the research project I am working on, it's about a completely different topic (out of distribution generalization) -- still I think Opacus could be useful beyond it's original scope in differential privacy 🙂