Search papers, labs, and topics across Lattice.
University of Bristol
1
0
3
Compressing multimodal LLM inputs with learned memory-augmented distillation not only speeds up inference by 1.5x, but also *improves* accuracy on audio-visual question answering.
