1 paper
Shichu Sun, Yichen Zhang, Haolin Song +6
Visual encoding followed by token condensing has become the standard architectural paradigm in multi-modal large language models (MLLMs). Many recent MLLMs increasingly favor globa…