Change hook-based non-linearity replacement to module-swap-based. #805
Merged
jhnwu3 merged 4 commits intosunlabuiuc:masterfrom Jan 31, 2026
Merged
Change hook-based non-linearity replacement to module-swap-based. #805jhnwu3 merged 4 commits intosunlabuiuc:masterfrom
jhnwu3 merged 4 commits intosunlabuiuc:masterfrom
Conversation
Logiquo
added
component: interpret
Collaborator
Author
|
This has also been tested on the mortality prediction stagenet mimic4 interpretability experiment and gives the same numbers |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Contributor: Yongda Fan (yongdaf2@illinois.edu)
Contribution Type: Interpretability
Description
Originally the DeepLift & GIM requires the model to have hook to replace the activation function, which is both cumbersome and error-prone (e.g. the old stagenet_mha forget to add a hook for one of the softmax).
The new method only requires that all model non-linearity is defined as
nn.Module, and DeepLift and GIM will dynamtically swap them if needed. This makes both the model definination cleanear and fix a few small issues regarding GIM with stagenet attn.