{
  "annotation_id": "fb77d63b-2ee9-4e7d-a4f4-4a12e8bfd0cc",
  "date_created": "2026-02-17T05:53:12.331000Z",
  "date_modified": "2026-02-17T05:53:12.331000Z",
  "file_hash": "7fd35e6c55e058cb799ca675747a59bf48c471e10658c550dd8a131f2d7871f1",
  "private": false,
  "record": {
    "abstract": "Growing context lengths in transformer-based language models have made the key-value (KV) cache a critical inference bottleneck. While many KV cache pruning methods have been proposed, they have not yet been adopted in major inference engines due to speed--accuracy trade-offs. We introduce KVzap, a fast, input-adaptive approximation of KVzip that works in both prefilling and decoding. On Qwen3-8B, Llama-3.1-8B-Instruct, and Qwen3-32B across long-context and reasoning tasks, KVzap achieves $2$--$4\\times$ KV cache compression with negligible accuracy loss and achieves state-of-the-art performance on the KVpress leaderboard. Code and models are available at https://github.com/NVIDIA/kvpress.",
    "arxiv_id": "2601.07891",
    "authors": [
      "Simon Jegou",
      "Maximilian Jeblick"
    ],
    "categories": [
      "cs.LG",
      "cs.AI",
      "cs.CL"
    ],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "title": "KVzap: Fast, Adaptive, and Faithful KV Cache Pruning",
    "url": "https://arxiv.org/abs/2601.07891",
    "version": "v1"
  },
  "schema_id": "dorsal/arxiv",
  "source": {
    "execution_id": "6debbc8c-2e9a-4bdd-b998-afe453050f06",
    "id": "arXiv Dataset",
    "type": "Model",
    "variant": "snapshot-2026-01-17",
    "version": "0.1.0"
  },
  "user_id": 1000002
}