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

Fix MXFP8 FlashInfer CUTLASS scale selection (#28553)

Browse Source
This commit is contained in:
Mohammad Miadh Angkad
2026-06-17 16:44:23 -07:00
committed by GitHub
parent a663500ea9
commit d773b49e5b
1 changed files with 3 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/quantization/fp8.py
+3 -4
View File
@@ -790,10 +790,9 @@ class Fp8LinearMethod(LinearMethodBase):
) )
if self.use_mxfp8: if self.use_mxfp8:
if ( if get_fp8_gemm_runner_backend().is_flashinfer_cutlass():
get_fp8_gemm_runner_backend().is_flashinfer_cutlass() weight_scale = layer.weight_scale_inv_swizzled
or get_fp8_gemm_runner_backend().is_flashinfer_trtllm() elif get_fp8_gemm_runner_backend().is_flashinfer_trtllm():
):
weight_scale = layer.weight_scale_inv_shuffled weight_scale = layer.weight_scale_inv_shuffled
else: else:
weight_scale = layer.weight_scale_inv weight_scale = layer.weight_scale_inv