1/ Introducing RL Swarm’s new backend: GenRL. A modular reinforcement learning library built for distributed, fault-tolerant training - now powering RL Swarm from the ground up. 🧵

2/ Each worker runs its own environment instance, contributes asynchronously to a shared rollout buffer, and updates its model weights independently, so no central controller is required.

3/ GenRL allows RL Swarm to work with any environment, described intuitively through code. This launch incorporates Reasoning Gym out-of-the-box, giving access to >100 community-created environments with no extra configuration required.

4/ What’s new: – Modular GenRL backend – Expanded configuration surface – Prebuilt Docker image for easy deployment – Reasoning Gym environment to enhance model reasoning capabilities – New multi-task swarm

5/ Now live on the Gensyn testnet. You can run RL-Swarm with GenRL today. Full code + setup: