minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

fix tuning_fused_moe_triton_sep tool per_channel_quant bug (#13027)

Browse Source
This commit is contained in:
Xiaoyu Zhang
2025-11-11 10:33:54 +08:00
committed by GitHub
parent aea88fa7af
commit f18ec927f3
1 changed files with 1 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
benchmark/kernels/fused_moe_triton/tuning_fused_moe_triton_sep.py
+1 -2
View File
@@ -514,7 +514,6 @@ def main(args: argparse.Namespace):
use_fp8_w8a8 = args.dtype == "fp8_w8a8" use_fp8_w8a8 = args.dtype == "fp8_w8a8"
use_int8_w8a8 = args.dtype == "int8_w8a8" use_int8_w8a8 = args.dtype == "int8_w8a8"
use_int8_w8a16 = args.dtype == "int8_w8a16" use_int8_w8a16 = args.dtype == "int8_w8a16"
per_channel_quant = args.per_channel_quant
topk_ids_dir = args.topk_ids_dir topk_ids_dir = args.topk_ids_dir
if args.batch_size is None: if args.batch_size is None:
@@ -602,7 +601,7 @@ def main(args: argparse.Namespace):
use_fp8_w8a8, use_fp8_w8a8,
use_int8_w8a8, use_int8_w8a8,
use_int8_w8a16, use_int8_w8a16,
per_channel_quant, False,
block_shape, block_shape,
) )
print( print(