LLM Collaboration With Multi-Agent Reinforcement Learning