#model-selection

8B models handle classification well. 70B models handle summarization. Code-specialized models beat generalists at code. Match the model to the task.