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

Disable triton FA #323

Closed
wants to merge 1 commit into from
Closed

Disable triton FA #323

wants to merge 1 commit into from

Conversation

hliuca
Copy link
Collaborator

@hliuca hliuca commented Dec 12, 2024

The CK version has better performance and more features.

@hliuca hliuca requested a review from gshtras December 12, 2024 19:27
@gshtras
Copy link
Collaborator

gshtras commented Dec 12, 2024

Could you provide some numbers to back this claim? Namely, models with/without MoE, and vision models on Instinct MI 300, 308, 325, and Navi 3,4
In our experience triton is generally better and more robust for the general case.

@hliuca
Copy link
Collaborator Author

hliuca commented Dec 12, 2024

It seems "export VLLM_USE_TRITON_FLASH_ATTN=0" is used everywhere... and the doc here, https://github.com/powderluv/vllm-docs

If it is better to set Triton as default, we can close this PR. Thank you.

@hliuca hliuca closed this Dec 12, 2024
@hliuca hliuca deleted the disable_triton_fa branch December 12, 2024 20:24
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.

2 participants