Training and understanding large multimodal systems on our own compute.
The Foundation Models center trains large language and multimodal systems end to end, studying how capabilities emerge with scale, data and architecture.
Our models and datasets are released openly for the research community.
A 70B model trained on the AI Supercomputer Lab.
Predicting capability from compute and data.
Serving large models affordably.
Collaborate on research, join as a student, or partner with us.