DP/EP Support for gpt-oss with deepep-ht comm kernel on SM100 (#23608)
This commit is contained in:
@@ -311,6 +311,7 @@ class ModelOptFp8MoEMethod(FusedMoEMethodBase):
|
||||
self,
|
||||
prepare_finalize: mk.FusedMoEPrepareAndFinalize,
|
||||
moe: FusedMoEConfig,
|
||||
layer: torch.nn.Module,
|
||||
) -> mk.FusedMoEPermuteExpertsUnpermute:
|
||||
experts = select_cutlass_fp8_gemm_impl(
|
||||
moe,
|
||||
@@ -1032,6 +1033,7 @@ class ModelOptNvFp4FusedMoE(FusedMoEMethodBase):
|
||||
self,
|
||||
prepare_finalize: mk.FusedMoEPrepareAndFinalize,
|
||||
moe: FusedMoEConfig,
|
||||
layer: torch.nn.Module,
|
||||
) -> mk.FusedMoEPermuteExpertsUnpermute:
|
||||
experts = select_nvfp4_gemm_impl(
|
||||
moe,
|
||||
|
||||
Reference in New Issue
Block a user