diff --git a/vllm/model_executor/models/qwen3_next.py b/vllm/model_executor/models/qwen3_next.py
index b095c79dc954b..880655da3f0a5 100644
--- a/vllm/model_executor/models/qwen3_next.py
+++ b/vllm/model_executor/models/qwen3_next.py
@@ -462,6 +462,8 @@ class Qwen3NextGatedDeltaNet(nn.Module, MambaBase):
         # ============================================================
         # Part 2: Core Attention (Custom Op)
         # ============================================================
+        # Note: we should not use torch.empty here like other attention backends,
+        # see discussions in https://github.com/vllm-project/vllm/pull/28182
         core_attn_out = torch.zeros(
             (num_tokens, self.num_v_heads // self.tp_size, self.head_v_dim),
             dtype=hidden_states.dtype,