Meta's PyTorch team has launched Monarch, a framework that simplifies distributed AI workflows across multiple GPUs and machines. It uses a single-controller model to manage computations across a cluster, making large-scale training and reinforcement ...