Inference Optimal VLMs Need Fewer Visual Tokens and More Parameters.

Kevin Y. Li, Sachin Goyal, Joo D. Semedo, J. Zico Kolter

Browse the full ICLR paper archive.