AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Semantic Cache Distillation: Efficient State Transfer via Reuse and Selective Patching

Qianli Ma, Zhiqing Tang, Hanshuai Cui, Zhi Yao, Weijia Jia

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Distributed LLM inference systems · 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
add Patch at transition layers to Reuse-only (ablation)SCD (Mistral-7B) · CMRC2018F1: 0.6421 -> 0.7850Improved
add Patch at transition layers to Reuse-only (ablation)SCD (Mistral-7B) · WikiText-2perplexity: 6.11 -> 4.65Improved
add Patch at transition layers to Reuse-only (ablation)SCD (Mistral-7B) · CMRC2018TTFT (ms): 118.2 -> 125.2Hurt
patch 1 layer instead of none (ablation)SCD (Mistral-7B) · WikiText-2perplexity: 6.788 -> 5.971Improved
patch top-3 layers instead of none (ablation)SCD (Mistral-7B) · WikiText-2perplexity: 6.788 -> 5.326Improved
patch top-5 layers instead of none (ablation)SCD (Mistral-7B) · WikiText-2perplexity: 6.788 -> 5.017Improved
patch top-8 layers instead of none (ablation)SCD (Mistral-7B) · WikiText-2perplexity: 6.788 -> 4.816Improved
patch 8 layers instead of 1 adds latency (ablation)SCD (Mistral-7B) · WikiText-2TTFT (ms): 115.9 -> 145.9Hurt
add Patch (Reuse-only vs SCD) latency breakdown recompute phase (ablation)SCD (Mistral-7B) · WikiText-2total TTFT (ms): 112.9 -> 124.1Hurt
replace DroidSpeak-style selective recompute with SCDMistral-7B pair · CMRC2018F1: 0.7052 -> 0.7850Improved
replace DroidSpeak-style selective recompute with SCDMistral-7B pair · CMRC2018TTFT (ms): 198.7 -> 125.2Improved
quantize transferred KV to 4-bit instead of BF16Raw KV transfer · CMRC2018F1: 0.2725 -> 0.1289Hurt

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