Remove skip_index and simultaneous hessian/build#301
Open
luciaquirke wants to merge 2 commits into
Open
Conversation
772a6cd to
e67f5a4
Compare
22f60ce to
702f6d2
Compare
for more information, see https://pre-commit.ci
LouisYRYJ
reviewed
Jun 22, 2026
Comment on lines
+138
to
+139
| # The autocorrelation Hessian is a dense per-module gradient Gram so | ||
| # it computes in one pass and skips the factored eigendecomposition |
Contributor
There was a problem hiding this comment.
Hmmh I think it may be better to modify approxmiate hessians to control whether we do EV computations
Contributor
|
nice, this looks good! I think we can probably unify the hessian approaches even more, but that may be done in another PR. Also important to be aware that everything outside of the autocorr hessian uses Fisher and not empirical Fisher (i.e. uses sampled labels from the logits, not actual data labels) |
This file contains hidden or 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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
We've had this efficiency optimization where the autocorrelation hessian can be computed simultaneously with the index build, but it's a bit convoluted and my understanding is that the amount of data you need for a good autocorrelation estimate is quite low anyway, so the speed gains are limited. Removing for simplicity.