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
Fix pre-commit (#22487)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Aug 8, 2025 at 11:21 UTC
157f9c13687e38b89fdeb20ecdbb75baf8153e0f
1 file changed
+2
-2
vllm/model_executor/models/minicpmo.py
+2
-2
@@ -589,7 +589,7 @@ class MiniCPMO(MiniCPMV2_6):
589
ret = torch.zeros(size, size, device=device, dtype=torch.bool)
590
# Vectorized computation of row indices and chunk boundaries
591
row_indices = torch.arange(size, device=device)
592
- chunk_indices = row_indices // chunk_size
592
+ chunk_indices = row_indices // chunk_size
593
if num_left_chunks < 0:
594
# If num_left_chunks < 0, start is always 0 for all rows
595
start_indices = torch.zeros_like(row_indices)
@@ -597,7 +597,7 @@ class MiniCPMO(MiniCPMV2_6):
597
# Compute start indices vectorially
598
start_chunk_indices = torch.clamp(chunk_indices - num_left_chunks,
599
min=0)
600
- start_indices = start_chunk_indices * chunk_size
600
+ start_indices = start_chunk_indices * chunk_size
601
# Compute ending indices vectorially
602
end_chunk_indices = chunk_indices + 1
603
end_indices = torch.clamp(end_chunk_indices * chunk_size +