Skip to content

ggml : allow CUDA graphs when using pipeline parallelism #13814

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

Merged
merged 1 commit into from
May 27, 2025

Conversation

slaren
Copy link
Member

@slaren slaren commented May 27, 2025

Due to changes in the graph in each evaluation when pipeline parallelism is enabled, CUDA graphs could not be used. This change ensures that during generation the graphs will not change, allowing the use of CUDA graphs when pipeline parallelism is enabled.

Fixes #13751

@slaren slaren merged commit 952f395 into master May 27, 2025
46 checks passed
@slaren slaren deleted the sl/fix-cuda-graphs-pp branch May 27, 2025 11:05
@ggerganov
Copy link
Member

Not sure, but this change might be causing some of the following issues:

#13879, #13906, #13909

@slaren
Copy link
Member Author

slaren commented May 30, 2025

I have seen the issues, but I cannot reproduce it.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
ggml changes relating to the ggml tensor library for machine learning
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Large performance drop when using pipeline parallelism and layer splitting on multiple GPUs
2 participants