large-scale-training

Prime Intellect releases prime-rl v0.6.0, enabling efficient reinforcement learning at trillion-parameter scale on large Mixture-of-Experts models, with sub-5-minute step times and optimizations for asynchronous RL.