diff --git a/vllm/model_executor/models/idefics3.py b/vllm/model_executor/models/idefics3.py index cf7d51c86c..ad94719241 100644 --- a/vllm/model_executor/models/idefics3.py +++ b/vllm/model_executor/models/idefics3.py @@ -310,7 +310,10 @@ class Idefics3MultiModalProcessor(BaseMultiModalProcessor[Idefics3ProcessingInfo prompt_ids = self._apply_hf_processor_tokens_only(prompt_ids) return BatchFeature(dict(input_ids=[prompt_ids]), tensor_type="pt") - mm_kwargs = {"input_data_format": "channels_last", **mm_kwargs} + image_processor = self.info.get_hf_processor().image_processor + if getattr(image_processor, "backend", "pil") == "pil": + mm_kwargs = {"input_data_format": "channels_last", **mm_kwargs} + processed_outputs = super()._call_hf_processor( prompt, mm_data,