Inference Compute-Optimal Video Vision Language Models

Peiqi Wang, ShengYun Peng, Xuewen Zhang +5

This work investigates the optimal allocation of inference compute across three key scaling factors in video vision language models: language model size, frame count, and the numbe…