A Stitch in Time Saves Nine: Small VLM is a Precise Guidance for Accelerating Large VLMs

Wangbo Zhao, Yizeng Han, Jiasheng Tang +5

Vision-language models (VLMs) have shown remarkable success across various multi-modal tasks, yet large VLMs encounter significant efficiency challenges due to processing numerous…