Skip to content

add causal-conv1d in Triton and integrate into vLLM with test code #18206

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

Closed
wants to merge 2 commits into from

Conversation

thoangtrvn
Copy link

@thoangtrvn thoangtrvn commented May 15, 2025

This PRs add causal-conv1d in Triton, making Mamba-based models as Triton-only backend.

There are two kernels implemented

  • causal_conv1d_update_triton: which outperforms the corresponding CUDA kernel in handling decode-only requests
image [data benchmarking two kernels runtime by increasing the number of decode-only requests in a batch] image [data benchmarking runtime processing the same batch of mixed requests, first send the batch to the single Triton kernel, and then using CUDA-split pathway where requests are first separated, with prefill-only requests are sent to one kernel, and decode-only requests are sent to the second kernel]

Also, some other improvement in reducing overhead is incorporated.

Test code is also added

Currently, the code is added as an optional pathway to the CUDA-split via VLLM_USE_TRITON_CONV1D environment variable set to 1.

Copy link

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

Copy link

mergify bot commented May 15, 2025

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @thoangtrvn.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

@mergify mergify bot added the needs-rebase label May 15, 2025
@mergify mergify bot removed the needs-rebase label May 15, 2025
@thoangtrvn thoangtrvn closed this May 15, 2025
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