feat: implement core RL training infrastructure, including GRPO training, evaluation utilities, custom environments, and Modal-based execution scripts.

3807ea3

Humanlearning commited on 13 days ago