works on
kv cache compression 1large language models 1low-rank approximation 1transformer inference 1tucker decomposition 1

From the 1 of 1 linked paper with an AI index.

collaborators

1 paper

cs.LG2026

A JoLT for the KV cache: Near-lossless KV cache compression via joint Lagrangian allocation of Tucker ranks and a rotated residual for llms

Rahul Krishnan, Volker Schulz

The paper introduces JoLT, a method that compresses the key‑value cache of transformer models by applying a partial Tucker decomposition on token and feature dimensions and adding…