[Bugfix][Qwen] fixes the weights dtype in qwen3_next: it is actually a bfloat16 (#27030)

Signed-off-by: Tao He <linzhu.ht@alibaba-inc.com>

vllm/model_executor/models/qwen3_next.py

@@ -325,7 +325,6 @@ class Qwen3NextGatedDeltaNet(nn.Module, MambaBase):
         self.A_log = nn.Parameter(
             torch.empty(
                 divide(self.num_v_heads, self.tp_size),
-                dtype=torch.float32,
             )
         )