1 paper
Siyu Sun, Jing Ren, Zhaohe Liao +8
Adapting decoder-only multimodal large language models (MLLMs) for unified multimodal retrieval faces two structural gaps. First, existing methods rely on implicit pooling, which o…