pLM trained using GRPO on LLM agent sequences | Proteinbase