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

fix(quantization): add sgl_kernel fallback for FP4 quantize on Blackwell GPUs (#17816)

Browse Source
This commit is contained in:
MikkoParkkola
2026-01-27 18:43:17 +08:00
committed by GitHub
parent dba264ac73
commit c56d19b977
1 changed files with 5 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/quantization/modelopt_quant.py
+4 -1
View File
@@ -66,12 +66,15 @@ if TYPE_CHECKING:
StandardDispatchOutput, StandardDispatchOutput,
) )
fp4_quantize = None
try: try:
if is_sm120_supported(): if is_sm120_supported():
try:
from flashinfer import fp4_quantize from flashinfer import fp4_quantize
except ImportError:
from sgl_kernel import scaled_fp4_quant as fp4_quantize
else: else:
from sgl_kernel import scaled_fp4_quant as fp4_quantize from sgl_kernel import scaled_fp4_quant as fp4_quantize
except ImportError: except ImportError:
fp4_quantize = None fp4_quantize = None