MQA Implementation for 2B models #114
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
There was a problem hiding this comment.
Thanks very much, MQA is one of the most important low-hanging fruit to implement right now! Looks pretty good overall, have a look at the comment about avoiding branching.
Tagging @pculliton to check the model exporting + vocab size change and @jan-wassenberg on any perf suggestions.
jan-wassenberg
requested changes
Mar 21, 2024
There was a problem hiding this comment.
This LGTM, if the performance looks good/better (I'm curious how much) and generation looks correct + @jan-wassenberg LGTMs can probably move forward with merging to dev.
|
I tested the weights converted from gemma_pytorch (2b-it and 7b-it) and the generation looks fine. |
jan-wassenberg
approved these changes
Mar 22, 2024
austinvhuang
added
the
copybara-import
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Outdated suggestions cannot be applied.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.
This PR implements "Multi-Query Attention" for the 2B models and modifies vocabulary size to be the same as gemma_pytorch (mentioned in #103). It works fine with weights converted from gemma_pytorch but will lead to the original gemma.cpp weights are unusable.
It needs more testing, and I'll use it to test the fine-tuned weights.