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 6, 05:45 UTC
[Bugfix] Store Qwen3Next A_log in fp32 (#37810)
Signed-off-by: effortprogrammer <yhjhoward7@gmail.com> Co-authored-by: Roger Wang <hey@rogerw.io>
Hojin Yang committed
Mar 23, 2026 at 16:36 UTC
54ab804e87dd534c0a8cb8ad8eec7ad5d75775ae
1 file changed
+1
vllm/model_executor/models/qwen3_next.py
+1
@@ -501,6 +501,7 @@ class Qwen3NextGatedDeltaNet(nn.Module, MambaBase):
501
self.A_log = nn.Parameter(
502
torch.empty(
503
divide(self.num_v_heads, self.tp_size),
504
+ dtype=torch.float32,
505
)
506
)
507