Changed default TokenAndPositionEmbedding initializer to 'uniform' #1237
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
|
I have no idea what |
This is some manually triggered GPU testing we maintainers will run for some PRs. No need on this one. |
|
Thanks! Yeah we should definitely be using the same initializer as Nice catch! |
mattdangerw
approved these changes
Sep 11, 2023
|
Hmm, looks like this is causing some test failures where we asserted the default while testing configs, but I actually believe that will no longer be the case with #1238. I will just rebase this PR, I think everything should pass after that. |
mattdangerw
force-pushed
the
tp-embeddings-initializer
branch
from
f8d7675 to
cf31bfe
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Outdated suggestions cannot be applied.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.
TokenAndPositionEmbeddingadvertises itself as a combination of aEmbeddingandPositionEmbedding, but the inconsistency in initializer leads to a surprising degradation of performance, e.g. replacing the custom layer in this example withTokenAndPositionEmbeddingmakes results significantly worse (unless you manually specify the initializer). AFAIK there is no theoretical reason why glorot initialization should be here.