Research
Technical write-ups, papers, and other kinds of releases from the company.
- Consensus: Making Transformers Less Fragile to Train A token-mixing mechanism developed for proteins makes transformer training less sensitive to high learning rates across text, DNA, and protein experiments.
- Odyssey: Scaling Novel Primitives for Better Protein Models Rethinking the primitives of protein modeling: local consensus and joint sequence–structure diffusion, scaled to 102 billion parameters.