Remove GeGLU activation function and golden tests. #4303
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
![copybara-service[bot]](https://avatars.githubusercontent.com/in/44061?s=60&v=4){kind=small}
copybara-service
bot
force-pushed
the
test_686677143
branch
4 times, most recently
from
f982d57
to
9fb2d30
Compare
GeGLU is not a simple activation function, but a gated linear layer used in modern MLPs. Our users are not well served by a baked-in implementation of a linear layer presented as a simple activation function. PiperOrigin-RevId: 689524209
copybara-service
bot
force-pushed
the
test_686677143
branch
from
9fb2d30
to
ad90642
Compare
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Outdated suggestions cannot be applied.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.
Remove GeGLU activation function and golden tests.
GeGLU is not a simple activation function, but a gated linear layer used in modern MLPs.
Our users are not well served by a baked-in implementation of a linear layer presented as
a simple activation function.