EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[fix] fix modelopt fp4 on b200 (#8195)

Browse Source
This commit is contained in:
JieXin Liang
2025-07-21 08:39:57 +08:00
committed by GitHub
parent 93d124ef5a
commit 7eebd44047
1 changed files with 4 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
python/sglang/srt/layers/quantization/petit.py
View File

@@ -21,6 +21,9 @@ from sglang.srt.layers.quantization.petit_utils import (
verify_petit_nvfp4_supported, verify_petit_nvfp4_supported,
) )
from sglang.srt.layers.quantization.utils import is_layer_skipped from sglang.srt.layers.quantization.utils import is_layer_skipped
from sglang.srt.utils import is_hip
_is_hip = is_hip()
# Initialize logger for the module # Initialize logger for the module
logger = logging.getLogger(__name__) logger = logging.getLogger(__name__)
@@ -104,7 +107,7 @@ class PetitNvFp4Config(QuantizationConfig):
@classmethod @classmethod
def is_petit_nvfp4_compatible(cls, quant_config: Dict[str, Any]) -> bool: def is_petit_nvfp4_compatible(cls, quant_config: Dict[str, Any]) -> bool:
quant_method = quant_config.get("quant_method", "").lower() quant_method = quant_config.get("quant_method", "").lower()
return quant_method == "modelopt" return _is_hip and quant_method == "modelopt"
def is_layer_excluded(self, prefix: str, exclude_modules: list): def is_layer_excluded(self, prefix: str, exclude_modules: list):
for pattern in exclude_modules: for pattern in exclude_modules: