Revert "SM100 Cutlass MLA decode with unrestricted num_heads (< 128) for DeepSeek TP"

[jeejeelee]

Reverts #20769
FIX #20769 (comment)

[github-actions]

👋 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.

🚀

[gemini-code-assist]

Code Review

This pull request reverts the changes from PR #20769, which introduced a new SM100 Cutlass MLA decode implementation. The revert appears to be clean and complete, removing the new feature and its related code across the C++, Python, and build system files. The changes correctly restore the previous state of the affected components.

[LucasWilkinson]

Give me a couple minutes I can push a fix

[LucasWilkinson]

#21020 testing it now

[tlrmchlsmth]

Seems we have a fix @LucasWilkinson? https://github.com/vllm-project/vllm/pull/21020/files

[jeejeelee]

Thank you for the quick fix, I am testing it locally now @LucasWilkinson

[LucasWilkinson]

yes sorry confirmed it works on hopper but waiting for it to build on Blackwell; building this kernel is slowww.... haha

edit:

Checked vllm serve runs when built on hopper
Checked can run VLLM_ATTENTION_BACKEND=CUTLASS_MLA_VLLM_V1 lm_eval --model vllm --model_args pretrained=deepseek-ai/DeepSeek-V2-Lite-Chat,trust_remote_code=true --tasks gsm 8k --batch_size auto