Sublime

Gemini Pretraining

It discusses the scaling of large language models, focusing on compute budgets, scaling laws, inference efficiency, and implications for training and model size optimization in various applications.