diff --git a/vllm/utils/flashinfer.py b/vllm/utils/flashinfer.py
index 3fd3759590..159d19bfad 100644
--- a/vllm/utils/flashinfer.py
+++ b/vllm/utils/flashinfer.py
@@ -220,6 +220,8 @@ def force_use_trtllm_attention() -> bool | None:
 
 def can_use_trtllm_attention(num_qo_heads: int, num_kv_heads: int) -> bool:
     """Check if the current configuration supports TRTLLM attention."""
+    if force_use_trtllm_attention() is False:
+        return False
     has_trtllm = supports_trtllm_attention()
     return has_trtllm and (num_qo_heads % num_kv_heads == 0)