[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:
@@ -853,7 +853,7 @@ class ModelOptFp8MoEMethod(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,
|
||||
@@ -967,7 +967,6 @@ class ModelOptFp8MoEMethod(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,
|
||||
@@ -1538,7 +1537,6 @@ class ModelOptNvFp4FusedMoE(FusedMoEMethodBase):
|
||||
layer.w2_weight,
|
||||
topk_weights,
|
||||
topk_ids,
|
||||
inplace=False,
|
||||
activation=layer.activation,
|
||||
global_num_experts=layer.global_num_experts,
|
||||
expert_map=layer.expert_map,
|
||||
|
||||
Reference in New Issue
Block a user