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, 02:15 UTC
[Bugfix] Fix GLM-4.1-V video prompt update (#20635)
Signed-off-by: Isotr0py <2037008807@qq.com>
Isotr0py committed
Jul 9, 2025 at 07:13 UTC
b9fca83256ac47901cc8c5a75259feed8945c7e7
1 file changed
+5
-2
vllm/model_executor/models/glm4_1v.py
+5
-2
@@ -65,7 +65,7 @@ from vllm.multimodal.parse import (ImageSize, MultiModalDataItems,
65
MultiModalDataParser)
66
from vllm.multimodal.processing import (BaseMultiModalProcessor,
67
BaseProcessingInfo, PromptReplacement,
68
- PromptUpdate)
68
+ PromptUpdate, PromptUpdateDetails)
69
from vllm.multimodal.profiling import BaseDummyInputsBuilder
70
from vllm.platforms import _Backend
71
from vllm.sequence import IntermediateTensors
@@ -1213,7 +1213,10 @@ class Glm4vMultiModalProcessor(BaseMultiModalProcessor[Glm4vProcessingInfo]):
1213
placeholder.append(eoi_token_id)
1214
placeholder.extend(frame_idx)
1215
placeholder.append(eov_token_id)
1216
- return placeholder
1216
+ return PromptUpdateDetails.select_token_id(
1217
+ placeholder,
1218
+ embed_token_id=hf_processor.video_token_id,
1219
+ )
1220
1221
return [
1222
PromptReplacement(