Tag
Google announces a major update to Tunix, its post-training library, with an asynchronous decoupled rollout engine to scale agentic reinforcement learning on JAX/TPU, eliminating idle time and improving throughput.