xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-07-10 00:37:19 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fix for issue 17396 (#18773)

Browse Source 
Signed-off-by: Fred Reiss <frreiss@us.ibm.com>
This commit is contained in:
Fred Reiss 2025-05-31 04:58:17 -07:00 committed by GitHub
parent 0f5e0d567e
commit f2c3f66d59
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 5 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/lora/ops/torch_ops/lora_ops.py
View File

@ -36,10 +36,13 @@ def bgmv_expand(inputs: torch.Tensor,
if outputs.shape[0] == 1 and output_tensor.shape[0] != 1: if outputs.shape[0] == 1 and output_tensor.shape[0] != 1:
limit = 1 limit = 1
# LoRA adapter and model may add different amounts of padding to output
common_len = min(outputs.shape[1], output_tensor.shape[1])
if add_inputs: if add_inputs:
output_tensor[:, :outputs.shape[1]] += outputs[:limit, :] output_tensor[:, :common_len] += outputs[:limit, :common_len]
else: else:
output_tensor[:, :outputs.shape[1]] = outputs[:limit, :] output_tensor[:, :common_len] = outputs[:limit, :common_len]
def sgmv_shrink( def sgmv_shrink(