areal-async-rl-language-reasoning

Scale RL training to large models through decoupled rollout and training workers with controlled data staleness.

Independent GitHub snapshot · Automated classification · Original documentation governs