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

Use the correct wrapper for fp4_quantize (#27956)

Browse Source 
Co-authored-by: Brayden Zhong <brayden@radixark.ai>
This commit is contained in:
Brayden Zhong
2026-06-11 20:23:09 -07:00
committed by GitHub
co-authored by Brayden Zhong
parent 40894be3c3
commit 8bfcc0c39c
1 changed files with 1 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/models/deepseek_v2.py
+1 -2
View File
@@ -286,8 +286,7 @@ class DeepseekV2MLP(nn.Module):
and self.swiglu_limit is None and self.swiglu_limit is None
and not isinstance(x, tuple) and not isinstance(x, tuple)
): ):
from flashinfer import fp4_quantize from sglang.srt.layers.quantization.fp4_utils import fp4_quantize
from sglang.srt.layers.quantization.nvfp4_gemm_swiglu_nvfp4_quant import ( from sglang.srt.layers.quantization.nvfp4_gemm_swiglu_nvfp4_quant import (
nvfp4_gemm_swiglu_nvfp4_quant, nvfp4_gemm_swiglu_nvfp4_quant,
) )