minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Skipped warning on sm100 (#18000)

Browse Source
This commit is contained in:
Kaixi
2026-01-31 20:21:03 -08:00
committed by GitHub
parent d443d2d2ae
commit 2b2515423a
1 changed files with 9 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/configs/model_config.py
+9 -7
View File
@@ -26,7 +26,7 @@ from transformers import PretrainedConfig
from sglang.srt.environ import envs from sglang.srt.environ import envs
from sglang.srt.layers.quantization import QUANTIZATION_METHODS from sglang.srt.layers.quantization import QUANTIZATION_METHODS
from sglang.srt.server_args import ServerArgs 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 ( from sglang.srt.utils.hf_transformers_utils import (
get_config, get_config,
get_context_length, get_context_length,
@@ -902,12 +902,14 @@ class ModelConfig:
f"supported in ROCm." f"supported in ROCm."
) )
if self.quantization not in optimized_quantization_methods: if self.quantization not in optimized_quantization_methods:
logger.warning( # Don't warn for MXFP4 on SM100 since it has optimized kernels
"%s quantization is not fully " if not (self.quantization == "mxfp4" and is_sm100_supported()):
"optimized yet. The speed can be slower than " logger.warning(
"non-quantized models.", "%s quantization is not fully "
self.quantization, "optimized yet. The speed can be slower than "
) "non-quantized models.",
self.quantization,
)
def _verify_dual_chunk_attention_config(self) -> None: def _verify_dual_chunk_attention_config(self) -> None:
if hasattr(self.hf_config, "dual_chunk_attention_config"): if hasattr(self.hf_config, "dual_chunk_attention_config"):