Search papers, labs, and topics across Lattice.
Peking University
2
0
4
Achieving state-of-the-art accuracy in multimodal RAG while drastically reducing inference latency could redefine efficiency standards in document-based question answering.