From 2b2515423a6482bc8e0f2e9e1a7f4c33146da1bd Mon Sep 17 00:00:00 2001 From: Kaixi Date: Sun, 1 Feb 2026 05:21:03 +0100 Subject: [PATCH] Skipped warning on sm100 (#18000) --- python/sglang/srt/configs/model_config.py | 16 +++++++++------- 1 file changed, 9 insertions(+), 7 deletions(-) diff --git a/python/sglang/srt/configs/model_config.py b/python/sglang/srt/configs/model_config.py index 32032bded..12fac064f 100644 --- a/python/sglang/srt/configs/model_config.py +++ b/python/sglang/srt/configs/model_config.py @@ -26,7 +26,7 @@ from transformers import PretrainedConfig from sglang.srt.environ import envs from sglang.srt.layers.quantization import QUANTIZATION_METHODS from sglang.srt.server_args import ServerArgs -from sglang.srt.utils import is_hip, retry +from sglang.srt.utils import is_hip, is_sm100_supported, retry from sglang.srt.utils.hf_transformers_utils import ( get_config, get_context_length, @@ -902,12 +902,14 @@ class ModelConfig: f"supported in ROCm." ) if self.quantization not in optimized_quantization_methods: - logger.warning( - "%s quantization is not fully " - "optimized yet. The speed can be slower than " - "non-quantized models.", - self.quantization, - ) + # Don't warn for MXFP4 on SM100 since it has optimized kernels + if not (self.quantization == "mxfp4" and is_sm100_supported()): + logger.warning( + "%s quantization is not fully " + "optimized yet. The speed can be slower than " + "non-quantized models.", + self.quantization, + ) def _verify_dual_chunk_attention_config(self) -> None: if hasattr(self.hf_config, "dual_chunk_attention_config"):