Skip to content

Conversation

@ilmarkov
Copy link
Contributor

@ilmarkov ilmarkov commented Apr 3, 2025

As mentioned in the comment to the original PR enabling custom allreduce on ROCM, device capability on RocM is not sequential so the way we enable custom allreduce for MI300 should be changed.
This PR adds use_custom_allreduce method to platform class letting enable custom allreduce for ROCM platform on gfx94 architectures.

@github-actions
Copy link

github-actions bot commented Apr 3, 2025

👋 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.

🚀

vllm/config.py Outdated
Copy link
Member

Choose a reason for hiding this comment

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

This condition should be inverted?

Suggested change
if current_platform.use_custom_allreduce():
if not current_platform.use_custom_allreduce():

vllm/config.py Outdated
Comment on lines 1626 to 1627
Copy link
Member

Choose a reason for hiding this comment

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

Suggested change
"Disable the custom all-reduce kernel because it is not "
"supported on on current platform.")
"Disabled the custom all-reduce kernel because it is not "
"supported on current platform.")

@ilmarkov ilmarkov force-pushed the imarkov/fix_ca_enable branch from fa234a2 to 7a2f7c2 Compare April 3, 2025 12:33
Signed-off-by: ilmarkov <imarkov@redhat.com>
@ilmarkov ilmarkov force-pushed the imarkov/fix_ca_enable branch from 7a2f7c2 to f93e47f Compare April 3, 2025 12:38
@tlrmchlsmth tlrmchlsmth enabled auto-merge (squash) April 3, 2025 14:49
@github-actions github-actions bot added the ready ONLY add when PR is ready to merge/full CI is needed label Apr 3, 2025
@vllm-bot vllm-bot merged commit ef608c3 into vllm-project:main Apr 4, 2025
61 of 62 checks passed
Alex4210987 pushed a commit to LeiWang1999/vllm-bitblas that referenced this pull request Apr 5, 2025
…#16010)

Signed-off-by: ilmarkov <imarkov@redhat.com>
Co-authored-by: ilmarkov <imarkov@redhat.com>
Signed-off-by: xinyuxiao <xinyuxiao2024@gmail.com>
lulmer pushed a commit to lulmer/vllm that referenced this pull request Apr 7, 2025
…#16010)

Signed-off-by: ilmarkov <imarkov@redhat.com>
Co-authored-by: ilmarkov <imarkov@redhat.com>
Signed-off-by: Louis Ulmer <ulmerlouis@gmail.com>
lk-chen pushed a commit to lk-chen/vllm that referenced this pull request Apr 29, 2025
…#16010)

Signed-off-by: ilmarkov <imarkov@redhat.com>
Co-authored-by: ilmarkov <imarkov@redhat.com>
shreyankg pushed a commit to shreyankg/vllm that referenced this pull request May 3, 2025
…#16010)

Signed-off-by: ilmarkov <imarkov@redhat.com>
Co-authored-by: ilmarkov <imarkov@redhat.com>
RichardoMrMu pushed a commit to RichardoMrMu/vllm that referenced this pull request May 12, 2025
…#16010)

Signed-off-by: ilmarkov <imarkov@redhat.com>
Co-authored-by: ilmarkov <imarkov@redhat.com>
Signed-off-by: Mu Huai <tianbowen.tbw@antgroup.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ready ONLY add when PR is ready to merge/full CI is needed

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants