[Bugfix] Fix CUDA arch flags for MoE permute

[minosfuture]

Purpose

Fix CUDA arch flag for MOE_PERMUTE_SRC.

This resolves the following error

  File "/data/users/yming/gitrepos/vllm/vllm/model_executor/layers/fused_moe/modular_kernel.py", line 545, in _maybe_chunk_fused_experts
    return self._do_fused_experts(
           ^^^^^^^^^^^^^^^^^^^^^^^
  File "/data/users/yming/gitrepos/vllm/vllm/model_executor/layers/fused_moe/modular_kernel.py", line 492, in _do_fused_experts
    self.fused_experts.apply(
  File "/data/users/yming/gitrepos/vllm/vllm/model_executor/layers/fused_moe/cutlass_moe.py", line 314, in apply
    run_cutlass_moe_fp8(
  File "/data/users/yming/gitrepos/vllm/vllm/model_executor/layers/fused_moe/cutlass_moe.py", line 200, in run_cutlass_moe_fp8
    output.copy_(ops.shuffle_rows(c3, c_map).view(M * topk, K),
RuntimeError: CUDA error: the provided PTX was compiled with an unsupported toolchain.
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1

Test Plan

pytest tests/kernels/test_shuffle_rows.py

Test Result

UT passed

(Optional) Documentation Update

[gemini-code-assist]

Code Review

This pull request fixes a critical bug in CMakeLists.txt by correcting the CUDA architecture flags for the MoE permute kernel, resolving a runtime error. The fix is direct and effective, ensuring the kernel runs on the intended hardware.

[gemini-code-assist]

The CUDA arch flags are now correctly set to ${CUDA_ARCHS}. This ensures that the MoE permute kernel is compiled with the appropriate architecture flags, resolving the original error. This change is critical to ensure the kernel runs on the intended hardware.

    CUDA_ARCHS "${CUDA_ARCHS}")

[houseroad]

I guess someone just copied it over, lol

[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.

🚀

[minosfuture]

cc: @tlrmchlsmth @houseroad

[houseroad]

Looks good. Thanks for the fix.