Keep FFN output layer in float32 for T5 models #1239
Draft
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
|
Hey @guillaumekln, thanks for the great library. Any idea as to when you might merge this to main? I'm currently utilizing the binaries of this PR built during CI and would love to get the performance improvements of 3.16+. |
|
I updated this branch if you want to try new binaries: https://github.com/OpenNMT/CTranslate2/actions/runs/5644199537?pr=1239 I'm still not sure we will merge this branch. The change is specific to a single model (T5) and I feel like it will get in the way of other features in the future. If you have enough memory you could now try to use "bfloat16" with T5 models. |
|
Thank you for the update. It's working well so far. |
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Outdated suggestions cannot be applied.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.
No description provided.