FineTune-LLM-OnlineRL

Game: Xiangqi

Main Idea - Fine-tuning LLM Agent with Online RL (PPO & LoRA) :

1. Pre-trained LLMs are used as starting policy for RL agent
2. Observations from environments are converted to text
3. Text observations triggers an action and subsequently updates the RL agent’s policy

Other Methods Implemented:

1. Random
2. Greedy
3. DQN
4. DDQN