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 13, 13:15 UTC
[Bugfix] fix qwen3-next crash (#28202)
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
Jiangyun Zhu committed
Nov 11, 2025 at 16:24 UTC
f0359fffa434a4fce981389f9dff93a2a4c2b13e
1 file changed
+1
-1
vllm/model_executor/models/qwen3_next.py
+1
-1
@@ -587,7 +587,7 @@ class Qwen3NextGatedDeltaNet(nn.Module, MambaBase):
587
self.conv1d.bias,
588
self.activation,
589
conv_state_indices=non_spec_state_indices_tensor[
590
- : attn_metadata.num_decodes
590
+ : attn_metadata.num_actual_tokens
591
],
592
validate_data=True,
593
)