Bt bloom #1221
Merged
Bt bloom #1221
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
|
can i ask when we can get this PR merged to optimize bloom? pytorch just added attention bias memory efficient optimization. pytorch/pytorch#104310 |
|
Contributor
|
Thank you, I'll make sure it is merged this week! |
fxmarty
approved these changes
Aug 11, 2023
Contributor
fxmarty
left a comment
fxmarty
left a comment
There was a problem hiding this comment.
LGTM, thanks a lot for the addition @baskrahmer it is working smoothly!
About the untangle batch_size from self.num_heads, was it necessary for numerical equivalence?
This was referenced Aug 11, 2023
Contributor
Author
|
@fxmarty thanks for the review. IIRC the untangling of the batch size is there since the key/value cache of Bloom is structured a bit differently than other autoregressive language models. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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Outdated suggestions cannot be applied.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.
What does this PR do?
What the title says :)
In order to make the tests with key-value cache work, the input format had to be changed in order to be compatible with the Bloom cache format. Perhaps this could be automated by adding to Transformers directly (for the generate method this is done in huggingface/transformers#20213)
Fixes #1175
Duplicate PR of #1187 but I messed up a bit with rebasing there, so I closed that one ;)
Before submitting