EVOLvE: Evaluating and Optimizing LLMs For Exploration