Search papers, labs, and topics across Lattice.
University of Electronic Science and Technology of China
1
0
2
Pruning 77.8% of visual tokens without losing performance could revolutionize the efficiency of multimodal large language models.
