Reproduce “Rethinking KV Cache Eviction for Agentic LLM Serving” 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
Jiakun Fan, Han Meng, Yuxuan Xia, Hao Mark Chen, Dong Li, Dimitrios S. Nikolopoulos
Penelope manages agent KV-cache eviction by preserving shared prefixes, releasing early private-history blocks, and retaining later suffixes for reuse. When histories return, it recomputes missing segments. In a Qwen3-8B single-request test, early eviction made the next request 2.93× faster at 128K context. Across replayed agent traces, end-to-end gains varied; FreeInference JCT improved about 2%.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.