train-with-environments

Train models with verifiers environments using hosted RL or prime-rl. Use when asked to configure RL runs, tune key hyperparameters, diagnose instability, set up difficulty filtering and oversampling, or create practical train and eval loops for new environments.