From b9fca83256ac47901cc8c5a75259feed8945c7e7 Mon Sep 17 00:00:00 2001 From: Isotr0py Date: Wed, 9 Jul 2025 07:13:58 +0800 Subject: [PATCH] [Bugfix] Fix GLM-4.1-V video prompt update (#20635) Signed-off-by: Isotr0py <2037008807@qq.com> --- vllm/model_executor/models/glm4_1v.py | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/vllm/model_executor/models/glm4_1v.py b/vllm/model_executor/models/glm4_1v.py index a3908e30ec6e3..0996bcf60aa1c 100644 --- a/vllm/model_executor/models/glm4_1v.py +++ b/vllm/model_executor/models/glm4_1v.py @@ -65,7 +65,7 @@ from vllm.multimodal.parse import (ImageSize, MultiModalDataItems, MultiModalDataParser) from vllm.multimodal.processing import (BaseMultiModalProcessor, BaseProcessingInfo, PromptReplacement, - PromptUpdate) + PromptUpdate, PromptUpdateDetails) from vllm.multimodal.profiling import BaseDummyInputsBuilder from vllm.platforms import _Backend from vllm.sequence import IntermediateTensors @@ -1213,7 +1213,10 @@ class Glm4vMultiModalProcessor(BaseMultiModalProcessor[Glm4vProcessingInfo]): placeholder.append(eoi_token_id) placeholder.extend(frame_idx) placeholder.append(eov_token_id) - return placeholder + return PromptUpdateDetails.select_token_id( + placeholder, + embed_token_id=hf_processor.video_token_id, + ) return [ PromptReplacement(