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 15, 21:00 UTC
[Bugfix] Fix use_direct_call condition in FusedMoE layer for (#14382)
Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>
Tyler Michael Smith committed
Mar 6, 2025 at 17:17 UTC
958adce4787561317ca2632ed8de4c9e9a02d60d
1 file changed
+1
-1
vllm/model_executor/layers/fused_moe/layer.py
+1
-1
@@ -359,7 +359,7 @@ class FusedMoE(torch.nn.Module):
359
360
# For smuggling this layer into the fused moe custom op
361
self.use_direct_call = self.dp_size == 1
362
- if self.use_direct_call:
362
+ if not self.use_direct_call:
363
compilation_config = vllm_config.compilation_config
364
if prefix in compilation_config.static_forward_context:
365
raise ValueError("Duplicate layer name: {}".format(prefix))