Google OpenRL is an Experimental Self-hosted API for LLM Post-Training Fine-tuning

▤ Summary

Google GKE Labs has launched OpenRL, an open-source self-hosted API for post-training and fine-tuning LLMs on Kubernetes. The project separates reinforcement learning infrastructure from AI research so machine learning teams can scale workflows on their own clusters. It also aims to improve GPU utilization by running multiple RL jobs in parallel and reducing idle time during sequential training loops. The release includes an autoresearch recipe for parallel experiments and integrates with GKE, macOS, Nvidia GPUs, and Tinker-compatible endpoints.

▦ Linked Companies