[Moe Refactor] Make Inplace Flag for FusedMoEModularKernel part of the constructor (#33375)

Signed-off-by: Bill Nell <bnell@redhat.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
This commit is contained in:
bnellnm
2026-02-05 18:07:18 +00:00
committed by GitHub
co-authored by Robert Shaw
parent 1ee95841bd
commit a57c8228ff
37 changed files with 132 additions and 109 deletions
@@ -854,7 +854,7 @@ class Fp8MoEMethod(FusedMoEMethodBase):
self.moe_quant_config = self.get_fused_moe_quant_config(layer)
if self.moe_quant_config:
assert self.experts_cls is not None
self.moe_mk, self.use_inplace = make_fp8_moe_kernel(
self.moe_mk = make_fp8_moe_kernel(
moe_quant_config=self.moe_quant_config,
moe_config=self.moe,
fp8_backend=self.fp8_backend,
@@ -958,10 +958,6 @@ class Fp8MoEMethod(FusedMoEMethodBase):
def supports_eplb(self) -> bool:
return True
@property
def allow_inplace(self) -> bool:
return True
@property
def is_monolithic(self) -> bool:
return self.fp8_backend == Fp8MoeBackend.FLASHINFER_TRTLLM
@@ -1032,7 +1028,6 @@ class Fp8MoEMethod(FusedMoEMethodBase):
layer.w2_weight,
topk_weights,
topk_ids,
inplace=self.use_inplace,
activation=layer.activation,
global_num_experts=layer.global_num_experts,
expert_map=layer.expert_map,