Skip to content

[CUDA] Enable full cudagraph for FlashMLA #18581

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

Open
wants to merge 3 commits into
base: main
Choose a base branch
from

Conversation

ProExpertProg
Copy link
Contributor

@ProExpertProg ProExpertProg commented May 23, 2025

Enable fullgraph CUDAGraph capture for the FlashMLA decode case.

Hacks:

  • building the capture metadata
  • prefill batch bypasses compiled code and manually calls eager code

Tested with:

python examples/offline_inference/basic/generate.py --model deepseek-ai/DeepSeek-V2-Lite --trust-remote-code -O {"full_cuda_graph":true}

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.

🚀

@hypnopump
Copy link

Could you give some details on speedup associated with this modification?

Copy link

mergify bot commented May 23, 2025

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

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 23, 2025
direct_call = has_prefill(attn_metadata) and self.full_cuda_graph
if direct_call:
# Skip the outer model layer as inner model is compiled
model_output = self.model.model.forward(
Copy link
Contributor Author

Choose a reason for hiding this comment

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

I'm thinking about setting a context to bypass Inductor backend directly here, but I guess we could also separately compile (and capture) prefill stage - any thoughts?
cc @youkaichao

Copy link
Member

Choose a reason for hiding this comment

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

this is hacky and depends on the fact that self.model.model is the underlying nn.Module we compile (which might not be the case if we are running multi-modality models).

I think we can have some fields in the forward context, like not_use_cudagraph, and set it to true only when we capture full cudagraph but get prefill data in the current batch. then, when we decide to replay the cudagraph, we can check this field.

it should just be changing this line

entry.cudagraph.replay()
to return entry.runnable(*args)

@ProExpertProg
Copy link
Contributor Author

Could you give some details on speedup associated with this modification?

I haven't necessarily profiled this but it's meant to enable the double-batch-overlap optimization (prototype in #18415)

@ProExpertProg ProExpertProg force-pushed the luka/mla-full-cudagraph branch 2 times, most recently from 976e852 to 40e7248 Compare May 30, 2025 20:21
Copy link

mergify bot commented Jun 4, 2025

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

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 Jun 4, 2025
@izhuhaoran
Copy link
Contributor

Hi, any further progress on this pr?

@ProExpertProg
Copy link
Contributor Author

Hi, any further progress on this pr?

Almost ready for review!

@ProExpertProg ProExpertProg force-pushed the luka/mla-full-cudagraph branch from 5e3f7ab to 30562a2 Compare June 4, 2025 21:27
@mergify mergify bot removed the needs-rebase label Jun 4, 2025
@ProExpertProg ProExpertProg force-pushed the luka/mla-full-cudagraph branch from 30562a2 to 9ea599d Compare June 5, 2025 00:30
@ProExpertProg
Copy link
Contributor Author

Currently experiencing some issues when batching (in unit test), need to investigate further.

Copy link

mergify bot commented Jun 5, 2025

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

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 Jun 5, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
Status: No status
Development

Successfully merging this pull request may close these issues.

4 participants