High-Fidelity KV Cache Summarization Using Entropy and Low-Rank Reconstruction

(jchandra.com)

13 points by jchandra2 days ago

2 comments

  • vivahir2152 days ago
    Interesting Approach. Curious about the latency tradeoff: OLS + SVD are much heavier than Top-K.Have you benchmarked end-to-end inference latency?
  • jchandra2 days ago
    [dead]