Bolmo
A fully open-source byte-level language model developed by AI2. It is created by retrofitting existing token-level language models to operate at the byte level, enabling direct processing of UTF-8 byte sequences without a tokenizer.
Timeline (last 90 days)
Ai2 highlighted three projects at COLM 2026: OLMo Hybrid, Asta, and Bolmo.
Bolmo is another open model project presented by Ai2 on this occasion.
The core technology underpinning Ai2's fully open-source byte-level language model Bolmo has been published in the journal Nature.
Bolmo is a fully open-source byte-level language model released by Ai2 that operates directly at the byte level without relying on a traditional tokenizer.
Bolmo's core technology involves retrofitting existing token-based language models to operate at the byte level, rather than training an entirely new byte model from scratch.
Bolmo's model itself, the method, and the checkpoints supporting generalization capability are all released publicly, allowing the community to directly reproduce and verify them.
All Facts (6)
Bolmo is a fully open-source byte-level language model released by Ai2 that operates directly at the byte level without relying on a traditional tokenizer.
50%UnverifiedBolmo's core technology involves retrofitting existing token-based language models to operate at the byte level, rather than training an entirely new byte model from scratch.
50%UnverifiedBolmo's model itself, the method, and the checkpoints supporting generalization capability are all released publicly, allowing the community to directly reproduce and verify them.
50%UnverifiedAi2 highlighted three projects at COLM 2026: OLMo Hybrid, Asta, and Bolmo.
50%UnverifiedBolmo is another open model project presented by Ai2 on this occasion.
50%UnverifiedThe core technology underpinning Ai2's fully open-source byte-level language model Bolmo has been published in the journal Nature.
50%