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 8, 01:15 UTC
[Minor] Add logging when using MXFP4 MXFP8 TRTLLM backend (#34916)
Signed-off-by: frankwang28 <frank.wbb@hotmail.com> Signed-off-by: Frank Wang <41319051+frankwang28@users.noreply.github.com> Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
Frank Wang committed
Feb 19, 2026 at 22:07 UTC
059779231f158b8b570e71aaa5c66f49b41b2fb1
1 file changed
+3
vllm/model_executor/layers/quantization/mxfp4.py
+3
@@ -128,6 +128,9 @@ def get_mxfp4_backend(with_lora_support: bool) -> Mxfp4Backend:
128
and has_flashinfer()
129
and envs.VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8
130
):
131
+ logger.info_once(
132
+ "Using FlashInfer MXFP4 MXFP8 TRTLLM backend for SM100", scope="local"
133
+ )
134
return Mxfp4Backend.SM100_FI_MXFP4_MXFP8_TRTLLM
135
elif current_platform.is_device_capability_family(100) and has_flashinfer():
136
logger.info_once(