diff --git a/vllm/model_executor/models/ovis2_5.py b/vllm/model_executor/models/ovis2_5.py index 6dbed78a6f..19dd047407 100644 --- a/vllm/model_executor/models/ovis2_5.py +++ b/vllm/model_executor/models/ovis2_5.py @@ -402,12 +402,11 @@ class Ovis2_5MultiModalProcessor(BaseMultiModalProcessor[Ovis2_5ProcessingInfo]) hf_processor_mm_kwargs: Mapping[str, object], out_mm_kwargs: MultiModalKwargsItems, ) -> list[PromptReplacement]: - tokenizer = self.info.get_tokenizer() - vocab = tokenizer.get_vocab() + hf_processor = self.info.get_hf_processor() placeholder = { - "image": vocab[IMAGE_TOKEN], - "video": vocab[VIDEO_TOKEN], + "image": hf_processor.get_token_value("image_token"), + "video": hf_processor.get_token_value("video_token"), } def get_replacement_ovis(item_idx, modality: str): diff --git a/vllm/transformers_utils/processors/ovis2_5.py b/vllm/transformers_utils/processors/ovis2_5.py index 11ac0360e7..b9c2d11240 100644 --- a/vllm/transformers_utils/processors/ovis2_5.py +++ b/vllm/transformers_utils/processors/ovis2_5.py @@ -78,7 +78,6 @@ class Ovis2_5Processor(ProcessorMixin): @cached_property def extra_special_tokens(self): - vocab = self.tokenizer.get_vocab() required_tokens = { "image_token": "", "video_token": "