Author

Nitzan Hodos

1 paper indexed here

We haven’t gathered this author’s papers yet. Follow them and we’ll fetch their work.

Not the right person? Other researchers publish under this name.

Preprint Aug 2026

AnchorKV: Anchor-Residual KV Cache Compression

AnchorKV, a compression scheme that shrinks the key-value cache by $20\times$ without discarding a single token, and consistently preserves accuracy across models and datasets, while keeping the entire context at a fraction of its cost.

M. Khalaf, Yara Shamshoum, Nitzan Hodos et al. · 0 citations