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

remove moe_align_block_size torch.zeros in small batch/expert mode (#5298)

Browse Source
This commit is contained in:
Xiaoyu Zhang
2025-04-12 03:13:55 +08:00
committed by GitHub
parent a0a9f6d64f
commit 60bcbf2a35
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/layers/moe/fused_moe_triton/fused_moe.py
View File

@@ -707,7 +707,7 @@ def moe_align_block_size(
dtype=torch.int32, dtype=torch.int32,
device=topk_ids.device, device=topk_ids.device,
) )
cumsum_buffer = torch.zeros( cumsum_buffer = torch.empty(
num_experts + 1, dtype=torch.int32, device=topk_ids.device num_experts + 1, dtype=torch.int32, device=topk_ids.device
) )