Reproduce “OVAL: Output-Aware Local Page Bases for KV Cache Retrieval” 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
Ashkan Shahbazi, Chayne Thrash, Soheil Kolouri
OVAL builds page summaries from keys and values, combining key fidelity with a local estimate of output sensitivity to guide page retrieval. Its output term assumes uniform within-page attention and isotropic query second moments. On DeepSeek R1 Llama 8B, its mean avg@k is 55.69, versus FreeKV’s 51.83, using the best η selected per setting rather than a fixed practical value.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.