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, 04:45 UTC
[Chore]: qwen3-moe-type-hints-mistake (#19860)
Co-authored-by: xinnan.hou <hxn02029096@alibaba-inc.com>
Xerxes committed
Jun 20, 2025 at 12:43 UTC
e41bf15cd04e6681249ab7d382cef6450a2115f5
1 file changed
+1
-1
vllm/model_executor/models/qwen3_moe.py
+1
-1
@@ -294,7 +294,7 @@ class Qwen3MoeDecoderLayer(nn.Module):
294
positions: torch.Tensor,
295
hidden_states: torch.Tensor,
296
residual: Optional[torch.Tensor],
297
- ) -> torch.Tensor:
297
+ ) -> tuple[torch.Tensor, torch.Tensor]:
298
# Self Attention
299
if residual is None:
300
residual = hidden_states