NousResearch/atropos
A reinforcement learning environment framework for training and evaluating language model decision-making trajectories.
Velocity · 7d
+3.1
★ / day
Trend
→steady
star history
Atropos provides diverse RL environments for collecting and evaluating trajectories from language models. It enables researchers to train LLMs through reinforcement learning, measuring and improving model behavior across different task scenarios. The framework is published by NousResearch and integrates with HuggingFace.