Dynamic Memory Compression: Retrofitting LLMs for Accelerated Inference.

Piotr Nawrot, Adrian Lancucki, Marcin Chochowski, David Tarjan, Edoardo M. Ponti

Browse the full ICML paper archive.