Update vit.py to include LayerNorm in the MLP head which is missing #317
Conversation
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
I have included the LayerNorm layer in the classification head, which needed to be added to the implementation. Also, an unnecessary layer norm was included, which I have removed. For reference, you have included LayerNorm in the ditill_mlp attribute of the class DistillWrapper in distill.py. Thank you.
Owner
|
@vivekh2000 hey Vivek, thanks for this PR so i think the layernorm should be kept at its current position, as most transformers these days end with a final norm will definitely remove the layernorm from the distill wrapper though (or make it an option, in the case that some of the variants do not have the final norm) |
19eb6d4 to
5e808f4
Compare
cbf6723 to
5cf8384
Compare
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Outdated suggestions cannot be applied.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.
I have included the
LayerNormlayer in the classification head to identify it with the original implementation. It will be easy to compare now.Also, an isolated
LayerNormwas included, which I have removed and inserted in the MLP head, again to make it more intuitive. For reference, you have includedLayerNormin theself.ditill_mlpattribute of theclass DistillWrapperindistill.py.I hope these changes are acceptable.Thank you for the excellent code base.