Manage GPU clusters for running LLMs