Search papers, labs, and topics across Lattice.
University of Exeter
1
0
3
LVLMs can run 2.3x faster with only a 2% accuracy drop, thanks to a new pruning method that understands which visual tokens are most relevant to the text.
