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

Jamba: revert skipped test #31094

Closed
wants to merge 3 commits into from
Closed

Jamba: revert skipped test #31094

wants to merge 3 commits into from

Conversation

gante
Copy link
Member

@gante gante commented May 28, 2024

What does this PR do?

Fixes #30828 🤗

Jamba: reverts skipped test, it is now passing on main
Mamba: all_generative_model_classes is not set on Mamba's tests, as it requests the attention outputs in the generate calls (which mamba doesn't have). Added a comment regarding that on MambaModelTest

(feel free to merge if you approve)

@gante gante requested a review from amyeroberts May 28, 2024 18:19
@HuggingFaceDocBuilderDev

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

@amyeroberts
Copy link
Collaborator

Just so I understand, are mamba and jamba generative models? I thought so. Is there a reason why we can't have custom generative tests for these models which don't require attention?

@gante
Copy link
Member Author

gante commented May 29, 2024

@amyeroberts both are generative.

Jamba has the generative tests enabled (it has attention outputs):

all_generative_model_classes = (JambaForCausalLM,) if is_torch_available() else ()

Mamba doesn't have. I agree it would be nice to have some tests -- or better yet, for the tests to be more flexible, don't check attention if the model doesn't have attentions. Would you be okay with a) adding a TODO in the code; b) opening an issue to make the tests more flexible?

@amyeroberts
Copy link
Collaborator

Would you be okay with a) adding a TODO in the code; b) opening an issue to make the tests more flexible?

@gante In this case no. As mamba is pretty popular, has very unusual logic, this isn't just something flaky e.g. hardware, and generates we should be testing its generation logic properly

@gante gante closed this Jun 18, 2024
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.

Unable to run generation tests for Mamba & Jamba models
3 participants