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 11, 15:00 UTC
[fix] fix SM check for Flashinfer TRTLLM MOE (#30314)
Signed-off-by: jiahanc <173873397+jiahanc@users.noreply.github.com>
jiahanc committed
Dec 11, 2025 at 17:00 UTC
0ab23c2b2be1cdbde41b824186f57343f102e306
1 file changed
+1
-1
vllm/model_executor/layers/quantization/utils/flashinfer_utils.py
+1
-1
@@ -290,7 +290,7 @@ def get_flashinfer_moe_backend() -> FlashinferMoeBackend:
290
if flashinfer_moe_backend in backend_map:
291
if (
292
flashinfer_moe_backend == "latency"
293
- and not current_platform.is_device_capability(100)
293
+ and not current_platform.has_device_capability(100)
294
):
295
logger.info_once(
296
"Flashinfer TRTLLM MOE backend is only supported on "