Ai2's LLM byteification accepted to Nature; new checkpoints extend method to Qwen and Llama

allen_ai · x · 2026-10-08

Ai2 announced that its byteification approach—retrofitting LLMs to operate directly over bytes, the method behind Bolmo—has been accepted to Nature. It gives existing LLMs the strengths of byte-level models without training from scratch, and new checkpoints extend the method from Olmo to Qwen and Llama. Lead author bminixhofer did most of the work during a three-month internship.

Original post →

Ai2's LLM byteification accepted to Nature; new checkpoints extend method to Qwen and Llama

More from Models