Fix TransformerBridge backward hook cleanup#1324
Merged
jlarson4 merged 2 commits intoMay 26, 2026
Merged
Conversation
Collaborator
|
Excellent work on this @SamuelePunzo, I appreciate the per-site choices |
jlarson4
added a commit
that referenced
this pull request
May 28, 2026
* Fix type of HookedTransformerConfig.device (#1230) * Fix type of HookedTransformerConfig.device This is typed as `Optional[str]` but sometimes returns `torch.device`. Updated the code to just return the `str` instead of wrapping with a device. I'm not confident that every function which takes a device will always be passed a string, so I didn't change functions like warn_if_mps. Found while working on #1219 * more cleanup * 3.0 CI Bugs (#1261) * Fixing `utils` imports * skip gated notebooks on PR from forks * Updating notebooks * Ensure LLaMA only runs when HF_TOKEN is available --------- Co-authored-by: jlarson4 <jonahalarson@comcast.net> * Fix TransformerBridge backward hook cleanup (#1324) * Fix TransformerBridge backward hook cleanup * Preserve backward hooks in run_with_cache * Fix FactoredMatrix indexing returning empty result for -1 index FactoredMatrix.__getitem__ converts an integer index `v` into the matrix (ldim/rdim) dimensions to `slice(v, v + 1)`. For `v == -1` this becomes `slice(-1, 0)`, which is an empty slice, so indexing the last row/column with a negative index silently returns a (0, ...) tensor instead of the requested element. Other negative indices (-2, -3, ...) are unaffected because `v + 1` stays negative. Use `None` as the slice stop when `v == -1` so the final element is kept. Adds regression tests covering negative indices on each matrix dimension. --------- Co-authored-by: Brendan Long <self@brendanlong.com> Co-authored-by: jlarson4 <jonahalarson@comcast.net> Co-authored-by: Samuele_Punzo <90847990+SamuelePunzo@users.noreply.github.com>
jlarson4
added a commit
that referenced
this pull request
May 29, 2026
* Fix type of HookedTransformerConfig.device (#1230) * Fix type of HookedTransformerConfig.device This is typed as `Optional[str]` but sometimes returns `torch.device`. Updated the code to just return the `str` instead of wrapping with a device. I'm not confident that every function which takes a device will always be passed a string, so I didn't change functions like warn_if_mps. Found while working on #1219 * more cleanup * 3.0 CI Bugs (#1261) * Fixing `utils` imports * skip gated notebooks on PR from forks * Updating notebooks * Ensure LLaMA only runs when HF_TOKEN is available --------- Co-authored-by: jlarson4 <jonahalarson@comcast.net> * Fix TransformerBridge backward hook cleanup (#1324) * Fix TransformerBridge backward hook cleanup * Preserve backward hooks in run_with_cache * Updated broken Slack link --------- Co-authored-by: Brendan Long <self@brendanlong.com> Co-authored-by: jlarson4 <jonahalarson@comcast.net> Co-authored-by: Samuele_Punzo <90847990+SamuelePunzo@users.noreply.github.com>
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
2 participants
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.
Fix
TransformerBridgebackward-hook cleanup so temporary backward hooks do not leak after context exit.HookPoint.remove_hooks()defaults to forward-only cleanup, andTransformerBridgewas calling it without an explicit direction in several cleanup paths. Thatmeant non-permanent backward hooks could remain attached after
with model.hooks(...)exited, andreset_hooks()on bridge components could also miss backwardhooks.
This change:
TransformerBridge.hooks(...)TransformerBridge.run_with_hooks(...)run_with_cache(...)teardown by cleaning up only the forward cache hooks it addsreset_hooks(), andrun_with_cache(...)This fixes the bug reported in the linked issue. No additional dependencies are required.
Fixes #1323
Type of change
Screenshots
Not applicable.
Checklist: