Trust the Model: Compact VLMs as In-Context Judges for Image-Text Data Quality

Daulet Toibazar, Kesen Wang, Sherif Mohamed +3

Vision-language models (VLMs) extend the conventional large language models by integrating visual data, enabling richer multimodal reasoning and significantly broadens the practica…