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

Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
2025-07-31 11:41:12 +08:00
parent 61445453df
commit 0f7919fca0
1 changed files with 1 additions and 1 deletions
--- a/vllm/model_executor/layers/fused_moe/deepep_ll_prepare_finalize.py
+++ b/vllm/model_executor/layers/fused_moe/deepep_ll_prepare_finalize.py
@ -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]

    def __init__(self,
                 buffer: deep_ep.Buffer,