GenRecal: Generation after Recalibration from Large to Small Vision-Language Models

GenRecal is a novel distillation framework designed to adapt and transfer knowledge between diverse vision-language models (VLMs), enabling smaller models to achieve performance on par with larger counterparts while overcoming computational challenges. The framework shows significant improvement in model performance through extensive experimental validation.