Metraipolis › archive › dc_e9e992b24f46

TaskPress: Query-Agnostic KV Cache Compression via Task-Guided Pruning — arXiv:2608.03276

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.03276
License (as archived)
CC0-1.0
Archived
2026-10-31T12:15:41.626Z
SHA-256 of the archived text
0e74649555daa14581b3c3b336a5adde801d3d7fa86525961cd75c6082c728f1
Stable URL of this page
https://metraipolis.com/documento/dc_e9e992b24f46

How to cite

TaskPress: Query-Agnostic KV Cache Compression via Task-Guided Pruning — arXiv:2608.03276 — arXiv (en), https://arxiv.org/abs/2608.03276 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_e9e992b24f46 (Metraipolis archive dc_e9e992b24f46, sha256 0e74649555daa14581b3c3b336a5adde801d3d7fa86525961cd75c6082c728f1)
BibTeX
@misc{dce9e992b24f46,
  title = {TaskPress: Query-Agnostic KV Cache Compression via Task-Guided Pruning — arXiv:2608.03276},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.03276},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_e9e992b24f46; sha256 0e74649555daa14581b3c3b336a5adde801d3d7fa86525961cd75c6082c728f1},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.03276. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Long-context inference with large language models is constrained by the linear growth of the key-value cache to sequence length. While pruning offers mitigation, prevailing methods determine query-specific token importance that cannot be reused across unseen queries. In contrast, we introduce TaskPress, a framework for task-guided, query-agnostic KV cache eviction. Instead of optimizing the cache for a single query, TaskPress constructs a reusable memory representation conditioned on a high-level task guide. The guide functions as a meta-query during prefill to filter irrelevant tokens before downstream queries are issued. In addition, TaskPress leverages quantization scale factors as a zero-cost signal for detecting influential representation outliers, providing an efficient proxy for token importance. Experiments on conducted on various tasks with long context input demonstrate that TaskPress efficiently creates a compact, reusable cache across diverse queries.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-05 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2608.03276 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-31T12:05:52.772Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-05?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).