This section presents a complete two-agent training run and examples of more advanced usage.
Quick Start#
Length-Ratio Training
A two-agent training example with a length-ratio reward, covering the dataset, reward function, configuration, and training loop.
Advanced Usage#
Resuming Training
Resuming an interrupted run or extending a finished one from training checkpoints, with all segments logged to one W&B run.
Multi-Turn Feedback
Two-turn training in which an external transition reports each agent’s response length before the agents revise.
Environments#
Richer examples can be found in the environments, which apply CoMLRL to writing, coding, travel planning, and tool use.