Metraipolis › archive › dc_dbf58aa013b2
Mask-Guided KV Cache Eviction in Block Diffusion Language Models — arXiv:2610.06996
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.06996
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T11:20:08.404Z
- SHA-256 of the archived text
d1219a152abe03b835606602081ae7aeb6c893275e76e326bb8ddd289761d515- Stable URL of this page
https://metraipolis.com/documento/dc_dbf58aa013b2
How to cite
Mask-Guided KV Cache Eviction in Block Diffusion Language Models — arXiv:2610.06996 — arXiv (en), https://arxiv.org/abs/2610.06996 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_dbf58aa013b2 (Metraipolis archive dc_dbf58aa013b2, sha256 d1219a152abe03b835606602081ae7aeb6c893275e76e326bb8ddd289761d515)
BibTeX
@misc{dcdbf58aa013b2,
title = {Mask-Guided KV Cache Eviction in Block Diffusion Language Models — arXiv:2610.06996},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.06996},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_dbf58aa013b2; sha256 d1219a152abe03b835606602081ae7aeb6c893275e76e326bb8ddd289761d515},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.06996. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Block diffusion language models keep a large key-value (KV) cache throughout generation and attend to it at every denoising step, limiting both memory capacity and generation speed. Reducing these costs requires deciding which past tokens to use for denoising the current block (selection) and which to keep in memory for future blocks (eviction). We propose MaskAhead, a training-free method that solves both tasks with a single mask-query-based ranking mechanism. Current-block masks guide selection, while probes of upcoming masked blocks guide eviction. Both rank KV entries by their estimated contribution to the attention output. Our quantized variant, Q-MaskAhead, computes selection and attention directly from low-bit KV, largely preserving the selected entries. Experiments on Fast-dLLM-v2, DreamReasoner, and LLaDA2.0-mini cover long-generation reasoning, long-prompt question answering, and needle-in-a-haystack retrieval. On long-prompt QA, MaskAhead reduces KV memory by $9.5\times$ on average with a 1.2-point mean F1 loss relative to dense inference. Q-MaskAhead increases the reduction to $20.1\times$ with a 2.3-point mean F1 loss. In a batch-32 systems profile, MaskAhead achieves $1.23\times$ end-to-end and $1.68\times$ decode-stage speedups over dense inference. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI); Computation and Language (cs.CL). Pagina del lavoro: https://arxiv.org/abs/2610.06996 — 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 2026-10-07T11:15:24.657Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).