Clustering Driven Memory Compression On Device

Compress user-specific memories for LLM personalization by clustering semantically similar memories and merging within clusters, reducing token count while preserving generation quality. Based on Bohdal et al. (ICASSP 2026). Use this skill when the user mentions: - "compress memories for context window" - "reduce memory tokens for on-device LLM" - "cluster and merge user memories" - "personalization with limited context budget" - "memory-efficient prompt construction" - "on-device LLM memory management"

ndpvt-web Updated 14 repo stars

File contents

ndpvt-web/arxiv-claude-skills commit 97397c4a7e

Frequently asked questions

npx get-skill add ndpvt-web/clustering-driven-memory-compression-on-device