Skip to content

[Misc] Expand SUPPORTED_HIDDEN_SIZES for DeepEP low-latency kernels #21818

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

Merged
merged 2 commits into from
Jul 31, 2025
Merged

[Misc] Expand SUPPORTED_HIDDEN_SIZES for DeepEP low-latency kernels #21818

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
27efe54
Done
jeejeelee Jul 29, 2025
25ee7a3
Merge branch 'vllm-project:main' into add-deepep-hidden-size
jeejeelee Jul 30, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion vllm/model_executor/layers/fused_moe/deepep_ll_prepare_finalize.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,7 @@ class DeepEPLLPrepareAndFinalize(mk.FusedMoEPrepareAndFinalize):

# DeepEP low-latency kernels are compiled only for certain
# specific hidden sizes.
SUPPORTED_HIDDEN_SIZES = [2048, 2560, 4096, 5120, 7168]
SUPPORTED_HIDDEN_SIZES = [2048, 2560, 4096, 5120, 6144, 7168]
Copy link
Contributor

Choose a reason for hiding this comment

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

high

Adding 6144 to SUPPORTED_HIDDEN_SIZES enables this new configuration, but the pull request lacks corresponding tests to validate its functionality. The existing tests for low-latency DeepEP kernels in tests/kernels/moe/test_deepep_moe.py do not cover this new hidden size.

To ensure the correctness of the implementation for this new hidden size and to prevent future regressions, please add a test case for k=6144 to test_low_latency_deep_ep_moe. This will verify that the pre-compiled kernel for this size works as expected.


def __init__(self,
buffer: deep_ep.Buffer,
Expand Down