[Kernel]Support W4A8 Grouped GEMM on Hopper (#29691)

Signed-off-by: czhu-cohere <conway.zhu@cohere.com>
2025-12-08 22:29:06 -05:00
parent ea657f2078
commit f6227c22ab
22 changed files with 2045 additions and 101 deletions
--- a/vllm/model_executor/layers/fused_moe/modular_kernel.py
+++ b/vllm/model_executor/layers/fused_moe/modular_kernel.py
@@ -367,7 +367,7 @@ class FusedMoEPrepareAndFinalize(ABC):
 class FusedMoEPermuteExpertsUnpermute(ABC):
    """
    An abstract base class for the [Permute-Experts-Unpermute] step described
-    above.
+        above.
    """

    def __init__(