Reproduce “WAKMan: Workload Aware KVCache Manager” 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
Suhas Kumar Morisetty, Vaibhavi Lokegaonkar, Sachin Mathew Varghese, Wesley Harris
WAKMan adapts LLM prefix-cache eviction by replaying recent block hashes through shadow LRU and S3FIFO policies, then selecting the policy with higher simulated hit rate for the next interval. Across 20 synthetic one-hour mixes, its mean and maximum regret were 2.04 and 11.0 percentage points, lower than either static policy; on pure multi-turn traffic, S3FIFO did better.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.