Metraipolis › archive › dc_5bc4945d2ef8
Dynamic Multi-Byte Prediction With Hierarchical Language Models — arXiv:2608.15454
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.15454
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-18T17:20:07.293Z
- SHA-256 of the archived text
dac2b833227d743b2259989b86feb6c38f2229974ac2eb43a86836134defe23e- Stable URL of this page
https://metraipolis.com/documento/dc_5bc4945d2ef8
How to cite
Dynamic Multi-Byte Prediction With Hierarchical Language Models — arXiv:2608.15454 — arXiv (en), https://arxiv.org/abs/2608.15454 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_5bc4945d2ef8 (Metraipolis archive dc_5bc4945d2ef8, sha256 dac2b833227d743b2259989b86feb6c38f2229974ac2eb43a86836134defe23e)
BibTeX
@misc{dc5bc4945d2ef8,
title = {Dynamic Multi-Byte Prediction With Hierarchical Language Models — arXiv:2608.15454},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.15454},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_5bc4945d2ef8; sha256 dac2b833227d743b2259989b86feb6c38f2229974ac2eb43a86836134defe23e},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.15454. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Byte-level hierarchical language models (LMs) have recently emerged as a robust alternative to their popular counterparts that use subword tokenization. However, generating one byte at a time remains a bottleneck for inference speed. To address this, we introduce multi-byte prediction (MBP), which generates multiple bytes in parallel, speeding up inference with minimal performance impact and no additional parameters. MBP builds on the popular multi-token prediction (MTP) paradigm with two crucial innovations. First, we introduce a variable-length prediction window that aligns with the latent tokens, or segments, of a hierarchical LM. Second, we implement a novel attention-masking scheme that enables parallel byte prediction without violating causality. We show that multi-byte prediction strikes a Pareto-optimal trade-off across multiple generative tasks, instruction following, question answering, summarization, and machine translation, achieving the best trade-off between performance and inference throughput. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-18 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2608.15454 — 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 2026-10-18T16:29:13.804Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-18?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).