High-Fidelity KV Cache Summarization Using Entropy and Low-Rank Reconstruction
(jchandra.com)
Interesting Approach. Curious about the latency tradeoff: OLS + SVD are much heavier than Top-K.Have you benchmarked end-to-end inference latency?
reply[dead]
reply[dead]
reply