Metraipolis › archive › dc_ef8ee212d47c
RIFT: KV Cache Compression for Draft Models with Complementary Compressibility Training — arXiv:2610.31098
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.31098
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-23T15:20:38.207Z
- SHA-256 of the archived text
058f32aa48dd746d5b14ce64b5b6561dfc3a5df0904ba9b4840543c37671e9b3- Stable URL of this page
https://metraipolis.com/documento/dc_ef8ee212d47c
How to cite
RIFT: KV Cache Compression for Draft Models with Complementary Compressibility Training — arXiv:2610.31098 — arXiv (en), https://arxiv.org/abs/2610.31098 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ef8ee212d47c (Metraipolis archive dc_ef8ee212d47c, sha256 058f32aa48dd746d5b14ce64b5b6561dfc3a5df0904ba9b4840543c37671e9b3)
BibTeX
@misc{dcef8ee212d47c,
title = {RIFT: KV Cache Compression for Draft Models with Complementary Compressibility Training — arXiv:2610.31098},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.31098},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ef8ee212d47c; sha256 058f32aa48dd746d5b14ce64b5b6561dfc3a5df0904ba9b4840543c37671e9b3},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.31098. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Speculative decoding is a popular decode acceleration method. As such, it is particularly relevant to long-output generation, which requires an increasingly large KV cache. KV cache compression offers a way to constrain costs in long-context drafting. However, compression methods that focus on prefill information can lose relevance as the output length grows. We term this phenomenon decode staleness and show that it degrades draft model accuracy under selection-based KV cache compression. We introduce RIFT, a family of selection-based KV cache compression strategies that balance prefill information and recent generated tokens. We show that refreshing the KV cache selection during token generation significantly improves the drafter's quality at the cost of additional decode latency. To alleviate the refresh costs, we propose an adaptive KV refresh strategy that reduces the number of score calculations while retaining the drafter's quality. We further demonstrate how a compressibility-driven drafter training phase, based on sparse attention masks, can improve the drafter's quality under a compressed KV cache. Combined with compressibility training, RIFT enables Qwen3.5-4B, drafting for Qwen3.5-122B, to retain over 98% of its full-cache acceptance length on 32K-token inputs, 4K-token outputs, and an 8K-token KV cache budget. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno). Categorie arXiv: Computation and Language (cs.CL). Pagina del lavoro: https://arxiv.org/abs/2610.31098 — 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 2026-10-23T15:09:38.349Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).