New KV cache compaction technique cuts LLM memory 50x without accuracy loss

MIT researchers developed Attention Matching, a KV cache compaction technique that compresses LLM memory by 50x in seconds — ...
Semiconductor Engineering

Functional Compaction for Functional Test Sequences (Purdue University, I. Pomeranz)

A new technical paper titled “Functional Compaction for Functional Test Sequences” was published by IEEE Fellow Irith Pomeranz at Purdue University. Abstract: “The occurrence of silent data corruption ...