A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 7, 09:00 UTC
[Bugfix] Fix misnamed parameter in compressed_tensors_moe.py (#35813)
Signed-off-by: Bill Nell <bnell@redhat.com> Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
bnellnm committed
Mar 3, 2026 at 16:29 UTC
a9b8b13e5cdc52aa7f4472d4d21f178e3805bcdd
1 file changed
+1
-1
vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe.py
+1
-1
@@ -887,7 +887,7 @@ class CompressedTensorsW8A8Fp8MoEMethod(CompressedTensorsMoEMethod):
887
w13,
888
w13_scale,
889
shard_size=layer.intermediate_size_per_partition,
890
- num_experts=layer.num_local_experts,
890
+ num_experts=layer.local_num_experts,
891
is_act_and_mul=self.moe.is_act_and_mul,
892
)
893