Scaling Laws and Compute-Optimal Training Beyond Fixed Training Durations Alexander Hägele

Scale has become a main ingredient in obtaining strong machine learning models. As a result, understanding a model's scaling properties is key to effectively