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 15, 06:00 UTC
[Misc] Fix misleading ROCm warning (#19486)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Jee Jee Li committed
Jun 12, 2025 at 00:12 UTC
04a55612dd6f7423a7138f6e5a4b837b63f37d65
1 file changed
+6
-1
vllm/attention/ops/triton_flash_attention.py
+6
-1
@@ -25,9 +25,14 @@ Not currently supported:
25
import torch
26
27
from vllm.platforms import current_platform
28
-from vllm.platforms.rocm import on_gfx1x
28
from vllm.triton_utils import tl, triton
29
30
+# Avoid misleading ROCm warning.
31
+if current_platform.is_rocm():
32
+ from vllm.platforms.rocm import on_gfx1x
33
+else:
34
+ on_gfx1x = lambda *args, **kwargs: False
35
+
36
torch_dtype: tl.constexpr = torch.float16
37
38