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.

Sign up for GitHub

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

Jump to bottom

[Core][Bugfix] Use correct device to initialize GPU data during CUDA-graph-capture #11233

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

[Core][Bugfix] Use correct device to initialize GPU data during CUDA-graph-capture #11233

wants to merge 1 commit into from
+23 −15

Conversation

yanburman
Copy link

@yanburman yanburman commented Dec 16, 2024 •
edited by github-actions bot
Loading

Until now, tensors and tiles were created over the default device. When trying to capture a CUDA-graph over other different devices, the code failed since the data was created on a different GPU.

This PR fixes this problem by initializing the data on the correct device.

@github-actions GitHub Actions
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.

🚀

@mgoin mgoin changed the title [Core][Bugfix] Use correct device to initialize GPU data during CUDA-… [Core][Bugfix] Use correct device to initialize GPU data during CUDA-graph-capture Dec 16, 2024
youkaichao
youkaichao approved these changes Dec 16, 2024
View reviewed changes
Copy link
Member

@youkaichao youkaichao left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

makes sense to me. but right now we only use one GPU per process. What's the use case when you have another GPU here?

@yanburman
Copy link
Author

makes sense to me. but right now we only use one GPU per process. What's the use case when you have another GPU here?

We are using vllm as a library and were hitting this. Sounds like a good practice to not assume running on single GPU, especially if it doesn't cost anything

@youkaichao youkaichao enabled auto-merge (squash) December 20, 2024 08:41
@github-actions github-actions bot added the ready ONLY add when PR is ready to merge/full CI is needed label Dec 20, 2024
@youkaichao
Copy link
Member

thanks for the contribution! we can merge it as long as ci passes.

auto-merge was automatically disabled December 22, 2024 06:42

Head branch was pushed to by a user without write access

@yanburman yanburman force-pushed the yan/ido_fix branch 2 times, most recently from 93fac08 to 87d4caf Compare December 25, 2024 08:53
@yanburman
[Core][Bugfix] Use correct device to initialize GPU data during CUDA-…
c62131a 
…graph-capture

Signed-off-by: Yan Burman <[email protected]>
Signed-off-by: Ido Asraff <[email protected]>
@yanburman
Copy link
Author

thanks for the contribution! we can merge it as long as ci passes.

My PR was failing on unrelated stuff. seems fixed now in main. could you please merge?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@youkaichao youkaichao youkaichao approved these changes

@WoosukKwon WoosukKwon Awaiting requested review from WoosukKwon WoosukKwon is a code owner

@robertgshaw2-neuralmagic robertgshaw2-neuralmagic Awaiting requested review from robertgshaw2-neuralmagic robertgshaw2-neuralmagic is a code owner

@njhill njhill Awaiting requested review from njhill njhill is a code owner

@ywang96 ywang96 Awaiting requested review from ywang96 ywang96 is a code owner

@comaniac comaniac Awaiting requested review from comaniac comaniac is a code owner

@alexm-neuralmagic alexm-neuralmagic Awaiting requested review from alexm-neuralmagic alexm-neuralmagic is a code owner

Assignees
No one assigned
Labels
ready ONLY add when PR is ready to merge/full CI is needed
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@yanburman @youkaichao