Make optimizer state public #2561
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #2561 +/- ##
==========================================
- Coverage 82.38% 82.37% -0.01%
==========================================
Files 825 825
Lines 105659 105643 -16
==========================================
- Hits 87042 87026 -16
Misses 18617 18617 ☔ View full report in Codecov by Sentry. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
For some use cases, it helps to do surgery on the internal optimizer state (in my case - I dynamicall add parameters, and want to add zero's as the momenta for these).
This could be done by going from optimizer -> record -> state, doing the modifications, and going back to an optimizer. A dedicate API could also work but just exposing fields seems simplest.
Also remove the Backend bound on WeightDecay - it doesn't make much sense to have WeightDecay be a FloatElement, and it complicates all bounds. AdamW already used a f32 for decay, so matching that.