A company is working on a large language model (LLM) and noticed that the LLM's outputs are not as diverse as expected. Which parameter should the company adjust?