A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
Remove redundant row_indices unsqueeze operation in MiniCPMO (#22528)
Signed-off-by: zitian.zhao <zitian.zhao@tencentmusic.com>
ZiTian Zhao committed
Aug 11, 2025 at 00:20 UTC
8c50d62f5a51799c2ecc1ad25380a5a6dd7c7180
1 file changed
-1
vllm/model_executor/models/minicpmo.py
-1
@@ -605,7 +605,6 @@ class MiniCPMO(MiniCPMV2_6):
605
max=size)
606
# Create column indices for broadcasting
607
col_indices = torch.arange(size, device=device).unsqueeze(0)
608
- row_indices = row_indices.unsqueeze(1)
608
start_indices = start_indices.unsqueeze(1)
609
end_indices = end_indices.unsqueeze(1)
610
# Vectorized mask creation