https://hgpu.org/?p=25676
Parallel Actors and Learners: A Framework for Generating Scalable RL Implementations