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
[BugFix][MM] Fix Nonetype error when video is cache in qwen2.5-omni-thinker (#26004)
Signed-off-by: wwl2755 <wangwenlong2755@gmail.com>
Wenlong Wang committed
Oct 1, 2025 at 01:03 UTC
84d57342b66eea5dbdf4a5f79c4422759d19f805
1 file changed
+9
-3
vllm/model_executor/models/qwen2_5_omni_thinker.py
+9
-3
@@ -323,9 +323,15 @@ class Qwen2_5OmniThinkerMultiModalProcessor(
323
mm_item_counts = mm_items.get_all_counts()
324
self._validate_mm_kwargs(mm_kwargs, mm_item_counts)
325
326
- use_audio_in_video = (all(
327
- item["use_audio_in_video"].data
328
- for item in mm_kwargs["video"]) if "video" in mm_kwargs else False)
326
+ use_audio_in_video = False
327
+ if "video" in mm_kwargs:
328
+ video_items = [
329
+ item for item in mm_kwargs["video"] if item is not None
330
+ ]
331
+ # only check video items (if there are any)
332
+ if video_items:
333
+ use_audio_in_video = all(item["use_audio_in_video"].data
334
+ for item in video_items)
335
336
if is_update_applied:
337
mm_placeholders = self._find_mm_placeholders(