Dingjie Song

LongLLaVA improves multi-image understanding for various applications.

TRIM method reduces image tokens in multi-modal language models while maintaining performance.

A new framework identifies when multimodal models use inappropriate training data.