New Paper: Dynamic Multi-Byte Prediction Speeds Up Hierarchical Byte-Level LMs

madeofAjala · x · 2026-08-28

A new paper, Dynamic Multi-Byte Prediction With Hierarchical Language Models, is now online. Hierarchical byte-level LMs are gaining adoption thanks to their tokenizer-free advantage, but lag behind subword models in inference speed; the paper introduces dynamic multi-byte prediction to close that gap.

Original post →

More from Research

Research channel →