Explore cost-optimal training for LLMs by balancing compute budgets between model size, training data, and inference efficiency. Learn how scaling laws like Chinchilla reduce costs.