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 issue of using only the part of video frame [Nemotron Nano] (#26186)
Signed-off-by: Eugene Khvedchenia <ekhvedchenia@nvidia.com>
Eugene Khvedchenya committed
Oct 4, 2025 at 03:21 UTC
5a05f26603703e15b254f69dc9c8b30e52f453ea
1 file changed
+1
-1
vllm/model_executor/models/nano_nemotron_vl.py
+1
-1
@@ -208,7 +208,7 @@ def video_to_pixel_values(
208
)
209
# dynamic_preprocess returns tensors already; take the single tile
210
assert len(pil_frame) >= 1
211
- frames_tensors.append(pil_frame[0])
211
+ frames_tensors.append(pil_frame[-1])
212
213
return torch.stack(frames_tensors)
214