From Bytes to Ideas: Language Modeling with Autoregressive U-Nets

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Videau, Mathurin, Idrissi, Badr Youbi, Leite, Alessandro, Schoenauer, Marc, Teytaud, Olivier, Lopez-Paz, David
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908410884128768
author Videau, Mathurin
Idrissi, Badr Youbi
Leite, Alessandro
Schoenauer, Marc
Teytaud, Olivier
Lopez-Paz, David
author_facet Videau, Mathurin
Idrissi, Badr Youbi
Leite, Alessandro
Schoenauer, Marc
Teytaud, Olivier
Lopez-Paz, David
contents Tokenization imposes a fixed granularity on the input text, freezing how a language model operates on data and how far in the future it predicts. Byte Pair Encoding (BPE) and similar schemes split text once, build a static vocabulary, and leave the model stuck with that choice. We relax this rigidity by introducing an autoregressive U-Net that learns to embed its own tokens as it trains. The network reads raw bytes, pools them into words, then pairs of words, then up to 4 words, giving it a multi-scale view of the sequence. At deeper stages, the model must predict further into the future -- anticipating the next few words rather than the next byte -- so deeper stages focus on broader semantic patterns while earlier stages handle fine details. When carefully tuning and controlling pretraining compute, shallow hierarchies tie strong BPE baselines, and deeper hierarchies have a promising trend. Because tokenization now lives inside the model, the same system can handle character-level tasks and carry knowledge across low-resource languages.
format Preprint
id arxiv_https___arxiv_org_abs_2506_14761
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle From Bytes to Ideas: Language Modeling with Autoregressive U-Nets
Videau, Mathurin
Idrissi, Badr Youbi
Leite, Alessandro
Schoenauer, Marc
Teytaud, Olivier
Lopez-Paz, David
Computation and Language
Artificial Intelligence
Tokenization imposes a fixed granularity on the input text, freezing how a language model operates on data and how far in the future it predicts. Byte Pair Encoding (BPE) and similar schemes split text once, build a static vocabulary, and leave the model stuck with that choice. We relax this rigidity by introducing an autoregressive U-Net that learns to embed its own tokens as it trains. The network reads raw bytes, pools them into words, then pairs of words, then up to 4 words, giving it a multi-scale view of the sequence. At deeper stages, the model must predict further into the future -- anticipating the next few words rather than the next byte -- so deeper stages focus on broader semantic patterns while earlier stages handle fine details. When carefully tuning and controlling pretraining compute, shallow hierarchies tie strong BPE baselines, and deeper hierarchies have a promising trend. Because tokenization now lives inside the model, the same system can handle character-level tasks and carry knowledge across low-resource languages.
title From Bytes to Ideas: Language Modeling with Autoregressive U-Nets
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2506.14761