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
[Misc] Fix Qwen3-VL `video_grid_thw` typing (#25646)
Signed-off-by: Roger Wang <hey@rogerw.io>
Roger Wang committed
Sep 25, 2025 at 03:16 UTC
7be9ffcd9f5c1a93705caac7d556794dd2608099
1 file changed
+1
-1
vllm/model_executor/models/qwen3_vl.py
+1
-1
@@ -1249,7 +1249,7 @@ class Qwen3VLForConditionalGeneration(nn.Module, SupportsMultiModal,
1249
rope_type="rope_3d")
1250
else:
1251
video_embeds = self.visual(pixel_values_videos,
1252
- grid_thw=grid_thw)
1252
+ grid_thw=grid_thw_list)
1253
1254
# Split concatenated embeddings for each video item.
1255
# Using prod on grid_thw_list instead of grid_thw.prod avoids CUDA sync