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

[Bug Fix] Fix broken sgemm_lora_a_graph_fwd due to invalid torch.mm() call (#24760)

Browse Source
This commit is contained in:
flutist-alibaba
2026-05-12 22:29:01 +03:00
committed by GitHub
parent 1419bcc39e
commit 00d8730d05
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/lora/torch_ops/graph_lora_ops.py
+1 -1
View File
@@ -65,7 +65,7 @@ def sgemm_lora_a_graph_fwd(
x_seq = torch.where(batch_token_mask, inputs, 0) x_seq = torch.where(batch_token_mask, inputs, 0)
w_seq = weights[lora_idx] w_seq = weights[lora_idx]
output.add_(scaling_tensor[lora_idx] * torch.mm(x_seq, w_seq.t(), 0)) output.add_(scaling_tensor[lora_idx] * torch.mm(x_seq, w_seq.t()))
return output return output