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 14, 05:30 UTC
[Qwen] Remove cuda hard-code in qwen3 next (#25243)
Signed-off-by: Icey <1790571317@qq.com>
Icey committed
Sep 19, 2025 at 20:25 UTC
838d7116ba59db528647b29f0d000742f4af9d4b
1 file changed
+1
-1
vllm/model_executor/models/qwen3_next.py
+1
-1
@@ -306,7 +306,7 @@ class Qwen3NextGatedDeltaNet(nn.Module, MambaBase):
306
eps=self.layer_norm_epsilon,
307
group_size=None,
308
norm_before_gate=True,
309
- device=torch.cuda.current_device(),
309
+ device=current_platform.current_device(),
310
dtype=config.torch_dtype,
311
)
312