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

[V1] Supports scheduling asynchronousization on V1 version #11133

Open
wants to merge 1 commit into
base: main
Choose a base branch
from

Conversation

lixiaolx
Copy link

@lixiaolx lixiaolx commented Dec 12, 2024

Hello @robertgshaw2-neuralmagic @njhill Based on the previous discussion #10634, I implemented a version of asynchronous scheduling. I measured the request locally with bs=256, which dropped from the original 5-6ms to 150-170us. During the entire process, gpu-util continued to be 100%.

image

Now that the verification of correctness has been completed, I will continue to improve the pr
TODO:

    • Support logits output
    • Synchronize the latest code to support multiple cards with tp>1
    • Repair corner cases
    • Clean up the code (such as setting the async-schedule switch, removing nvtx, etc.)
    • Continue to optimize and open the cudagraph token gap, which is currently 150-170us (when cudagrah is not used, prepare_input is processed in advance in the new stream and can be 50-70us).
      But after turning on cudagraph, when using the new stream to process input, the precision is not aligned. Can we discuss this? Is there anything we noticed in cudagrpah?

In addition to the above, everyone is welcome to make suggestions

Copy link

👋 Hi! Thank you for contributing to the vLLM project.
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 do one of these:

  • Add ready label to the PR
  • Enable auto-merge.

🚀

Copy link

mergify bot commented Dec 12, 2024

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

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 Dec 12, 2024
@mergify mergify bot added the v1 label Feb 5, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant