The parameter-to-data ratio that minimizes loss for a fixed compute budget, identified by DeepMind's 2022 Chinchilla paper.
Continue to AI University →