From Bytes to Ideas: Language Modeling with Autoregressive U-Nets
This addresses the problem of inflexible tokenization for NLP researchers and practitioners, offering a novel approach that could enhance model adaptability, though it appears incremental in performance gains.
The paper tackles the rigidity of fixed tokenization in language models by introducing an autoregressive U-Net that learns multi-scale embeddings from raw bytes, enabling predictions at varying granularities from bytes to words. The model ties strong BPE baselines in shallow hierarchies and shows promising trends in deeper ones, while also handling character-level tasks and cross-lingual knowledge transfer.
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.