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

Fix error when calling quantization (#12548)

Browse Source
This commit is contained in:
fzyzcjy
2025-11-03 10:17:43 -08:00
committed by GitHub
parent 952fbe47cb
commit 385599cb04
1 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/quantization/modelopt_quant.py
+3 -1
View File
@@ -56,6 +56,8 @@ try:
from flashinfer import fp4_quantize from flashinfer import fp4_quantize
else: else:
from sgl_kernel import scaled_fp4_quant as fp4_quantize from sgl_kernel import scaled_fp4_quant as fp4_quantize
from flashinfer import fp4_quantize as fp4_quantize_flashinfer
except ImportError: except ImportError:
fp4_quantize = None fp4_quantize = None
@@ -1571,7 +1573,7 @@ class ModelOptNvFp4FusedMoEMethod(FusedMoEMethodBase):
# Quantize before comm, swizzle after. # Quantize before comm, swizzle after.
if x.shape[0] > 0: if x.shape[0] > 0:
x, x_sf = fp4_quantize( x, x_sf = fp4_quantize_flashinfer(
x, layer.w13_input_scale_quant, is_sf_swizzled_layout=False x, layer.w13_input_scale_quant, is_sf_swizzled_layout=False
) )
else: else: