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 15, 17:15 UTC
[Bugfix] Fixed the issue of not being able to input video and image simultaneously (#15387)
Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>
Chauncey committed
Mar 25, 2025 at 11:48 UTC
10b34e36b9be7f68e236ec9c78425b6304be4cd1
1 file changed
+6
-6
vllm/entrypoints/chat_utils.py
+6
-6
@@ -556,11 +556,11 @@ class MultiModalItemTracker(BaseMultiModalItemTracker[object]):
556
raise ValueError(\
557
"Only one message can have {'type': 'image_embeds'}")
558
mm_inputs["image"] = image_embeds_lst[0]
559
- elif "image" in items_by_modality:
559
+ if "image" in items_by_modality:
560
mm_inputs["image"] = items_by_modality["image"] # A list of images
561
- elif "audio" in items_by_modality:
561
+ if "audio" in items_by_modality:
562
mm_inputs["audio"] = items_by_modality["audio"] # A list of audios
563
- elif "video" in items_by_modality:
563
+ if "video" in items_by_modality:
564
mm_inputs["video"] = items_by_modality["video"] # A list of videos
565
return mm_inputs
566
@@ -589,11 +589,11 @@ class AsyncMultiModalItemTracker(BaseMultiModalItemTracker[Awaitable[object]]):
589
raise ValueError(
590
"Only one message can have {'type': 'image_embeds'}")
591
mm_inputs["image"] = image_embeds_lst[0]
592
- elif "image" in items_by_modality:
592
+ if "image" in items_by_modality:
593
mm_inputs["image"] = items_by_modality["image"] # A list of images
594
- elif "audio" in items_by_modality:
594
+ if "audio" in items_by_modality:
595
mm_inputs["audio"] = items_by_modality["audio"] # A list of audios
596
- elif "video" in items_by_modality:
596
+ if "video" in items_by_modality:
597
mm_inputs["video"] = items_by_modality["video"] # A list of videos
598
return mm_inputs
599