SparQ Attention: Bandwidth-Efficient LLM Inference.

Luka Ribar, Ivan Chelombiev, Luke Hudlass-Galley, Charlie Blake, Carlo Luschi, Douglas Orr

Browse the full ICML paper archive.