Ai2's byte-level LM retrofitting method published in Nature, extended to Qwen and Llama
PontiEdoardo · x · 2026-10-08
Allen AI announced that its method for retrofitting language models to operate over bytes — the approach behind Bolmo — has been accepted to Nature.
- The technique converts existing token-based LMs to work directly on bytes, removing the tokenizer bottleneck
- Originally developed for Bolmo, it has now been generalized to other model families
- New open checkpoints extending the method from Olmo to Qwen and Llama, plus all data, are being released
More from Research
- CrystalJev cuts materials screening cost 30x by reading foundation models as fast decision-makers — CatAstro_Piyush · 2026-10-08
- OpenSLA unifies sensor, language and action in one model across 116K people and 79 sensor modalities — yang-ai-lab · 2026-10-08
- Math professor grades OpenAI's 722 results: mostly B/C level, one D-level shock — khademinori · 2026-10-08
- Why temp=0 LLM inference still isn't deterministic: floating-point order and parallelism — ducha_aiki · 2026-10-08
- Cambridge's S2PD uses serial diffusion to make video generation physically consistent — kastnerkyle · 2026-10-08
- A sober take on AI mining centuries of math: real medium-term gains, no autonomous-math miracle — _onionesque · 2026-10-08