[Kernel][Performance] Add FlashInfer cutedsl NVFP4 GEMM backend (#42235)
Signed-off-by: Mohammad Miadh Angkad <176301910+mmangkad@users.noreply.github.com>
This commit is contained in:
@@ -956,6 +956,15 @@ class AsyncTPPass(VllmFusionPatternMatcherPass):
|
||||
a_scale_view=a_scale_view,
|
||||
)
|
||||
)
|
||||
self.register(
|
||||
FlashInferAllGatherFP4Pattern(
|
||||
self.model_dtype,
|
||||
self.device,
|
||||
"cute-dsl",
|
||||
use_8x4_sf_layout=False,
|
||||
a_scale_view="float8",
|
||||
)
|
||||
)
|
||||
# NVFP4 reduce-scatter does not need scale communication: FP4
|
||||
# scales are consumed by the local GEMM and only BF16 partial
|
||||
# outputs are reduced. Keep this PR scoped to the all-gather
|
||||
|
||||
Reference in New Issue
Block a user