A power-law relationship, discovered by training many models of different sizes, that predicts a model's loss from the compute, parameters, and data used to train it.
Continue to AI University →