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

Remove contiguous before Flashinfer groupwise fp8 gemm (#6804)

Browse Source
This commit is contained in:
Baizhou Zhang
2025-06-01 18:30:54 -07:00
committed by GitHub
parent c429919def
commit 6a47b73024
1 changed files with 6 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
python/sglang/srt/layers/quantization/fp8_utils.py
View File

@@ -166,11 +166,13 @@ def flashinfer_gemm_w8a8_block_fp8_linear(
input_2d, block_size[1], column_major_scales=False input_2d, block_size[1], column_major_scales=False
) )
x_scale_input = x_scale.T.contiguous()
weight_scale_input = weight_scale.T.contiguous()
output = gemm_fp8_nt_groupwise( output = gemm_fp8_nt_groupwise(
q_input, weight, x_scale_input, weight_scale_input, out_dtype=input_2d.dtype q_input,
weight,
x_scale,
weight_scale,
scale_major_mode="K",
out_dtype=input_2d.dtype,
) )
if bias is not None: if bias is not None: