single mention · post-training
Co-located million-token agentic RL system with resumable sandboxes
The sources on this page only mention it in passing. See rollout & RL infrastructure for the methods filed beside it.
Filed under post-training › rollout & RL infrastructure