Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[BUG] Fix gemma 2 flash attention 2 #31932

Closed
wants to merge 1 commit into from

Conversation

Boubou78000
Copy link

There was a bug line 815:

It was using the version library and, using the version function argument.

AttributeError: 'str' object has no attribute 'parse'

@tanliboy
Copy link

I got the same problem while trying to fine-tune Gemma 2 with the latest flash-attn. Thanks for fixing it!

@Boubou78000
Copy link
Author

Boubou78000 commented Jul 13, 2024

I got the same problem while trying to fine-tune Gemma 2 with the latest flash-attn. Thanks for fixing it!

I just fixed ONE of the problem.
There are more 😭

@vanilladucky
Copy link

vanilladucky commented Jul 15, 2024

Hi, I saw this case as being closed and was wondering if this bug not going to get fixed? My implementation with flash attention for Gemma 2 is not working because of this problem.

I was using the flash attention library v==2.6.1

@Boubou78000
Copy link
Author

Boubou78000 commented Jul 15, 2024

Hi, I saw this case as being closed and was wondering if this bug not going to get fixed? My implementation with flash attention for Gemma 2 is not working because of this problem.

I was using the flash attention library v==2.6.1

  1. Install the dev version from source
  2. Not the end of the problems

@vanilladucky
Copy link

Ah yes, managed to get it fixed, thank you very much!

@thepowerfuldeez
Copy link

Hi everyone! Can't find the fix for that in the latest main, nor in the discussion from the link above. Why this PR is closed?

@thepowerfuldeez
Copy link

UPD: found update in main

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants