diff --git a/vllm/attention/ops/triton_flash_attention.py b/vllm/attention/ops/triton_flash_attention.py
index a26e713b1c624..49070e4c7ae6a 100644
--- a/vllm/attention/ops/triton_flash_attention.py
+++ b/vllm/attention/ops/triton_flash_attention.py
@@ -25,9 +25,14 @@ Not currently supported:
 import torch
 
 from vllm.platforms import current_platform
-from vllm.platforms.rocm import on_gfx1x
 from vllm.triton_utils import tl, triton
 
+# Avoid misleading ROCm warning.
+if current_platform.is_rocm():
+    from vllm.platforms.rocm import on_gfx1x
+else:
+    on_gfx1x = lambda *args, **kwargs: False
+
 torch_dtype: tl.constexpr = torch.float16