Reproduce “EchoKV: Self-Corrective KV Cache Eviction via Iterative Query Feedback” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
JianKun Zou, Tongtong Yuan
EchoKV revises a provisional key-value cache workspace during long-context inference, using early generated queries to rescore the full prompt cache before completing lookahead decoding. This feedback is intended to preserve evidence that an initial workspace may omit. On Qwen2.5-7B-Instruct at cache budget 128, it scores 57.10 on LongBench HotpotQA versus LAQ’s 56.76, while aggregate results remain comparable to LAQ.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.