LLM Digest

Story

arxiv_cs_cl · Nov 4, 2026 · paper

Source brief

Where Should a KV-Cache Repair Budget Go? Distractor Structure and Contextual Compatibility in Non-Prefix Reuse

arxiv.orgNov 4, 2026
original source linked

In brief

Non-prefix KV-cache reuse encodes retrieved chunks independently and concatenates their caches. To find where a small repair budget should be spent, we vary the chunks around the answer-bearing one (the target) across three 7B-8B models and four datasets of dialogue memory, open-

Read the original at arxiv.org →Open in live feed Read that day’s brief

Where Should a KV-Cache Repair Budget Go? Distractor Structure and Contextual Compatibility in Non-Prefix Reuse

Earlier in this thread 4 items

Where Should the Evaluation Budget Go in LLM-Driven Agent Scaffold Evolution?

DeepSeek Cut Its KV-Cache HBM Need 75% and SSD Need 87.5%. Micron and Sandisk Investors Should Pay Attention

KVShareArena: KV-Cache Reuse Across Contexts and Model Checkpoints

DARE: Difficulty Analysis for Repair Enhancement