AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

EpiCache: Episodic KV Cache Management for Long-Term Conversation on Resource-Constrained Environments

Minsoo Kim, Arnav Kundu, Han-Byul Kim, Richa Dixit, Minsik Cho

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: KV cache eviction and compression · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
episodic KV compression with block prefill instead of KVzipQwen3-4B, RealTalk · RealTalk cross-episode queries, 8K budgetGPT score: 35.9 -> 50.8Improved
episodic clustering and episode-specific caches vs KVzip in block prefillQwen3-4B, RealTalk · Avg across query types, 8K budgetGPT score: 36.0 -> 49.8Improved
add sensitivity-aware layer-wise budget allocation (ablation)EpiCache, Qwen3-4B · RealTalk, 8K budgetGPT score: 49.8 -> 53.9 (alpha=2)Improved
make layer allocation too sharp (alpha 8 vs 2 to 4) (ablation)EpiCache, Qwen3-4B · RealTalk, 8K budgetGPT score: 53.9 -> 49.8Hurt
segment conversation by utterances instead of words or tokens (ablation)EpiCache, Qwen3-4B · RealTalk, 8K budgetGPT score: 47.5 -> 49.8Improved
use Qwen3-Embedding instead of LLM embedding layer (ablation)EpiCache, Qwen3-4B · RealTalk, 8K budgetGPT score: 43.0 -> 49.8Improved
scale embedding encoder from 0.6B to 4B (ablation)EpiCache, Qwen3-4B · RealTalk, 8K budgetGPT score: 49.8 -> 50.6No clear effect
increase number of episodes E from 2 to 8 (ablation)EpiCache, Qwen3-4B · RealTalk, 8K budgetGPT score: 47.9 -> 51.3Improved
sensitivity profiling with 1 vs 100 calibration samples, various domains (ablation)EpiCache layer-wise allocation · LoCoMo and RealTalk, Qwen3-8B, 6K budgetF1 / GPT scoreNo clear effect
vary K-Means initialization and generation sampling seeds (ablation)EpiCache · LoCoMo and RealTalk, Qwen3-8Bscore: std below 0.5No clear effect
apply KVzip post-prefill eviction in block prefill settingKVzip, LLaMA-3.2-3B · LongConvQALongConvQA scoreHurt
use block prefill with episodic KV and 8K budget instead of full KVLLaMA3.2-3B, 90K-token history · 300-turn conversation efficiencypeak memory (GB): 36.3 -> 9.6Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy