The PyTorch Foundation announced Ray as a hosted project and introduced PyTorch Monarch, a framework that simplifies distributed AI workloads by treating GPU clusters as single logical devices. Ray joins DeepSpeed and vLLM to form a complete open-source stack for model development, training, and inference. The conference highlighted open research initiatives from Stanford (Marin) and AI2 (Olmo-Thinking) focused on transparency and reproducibility in foundation model development, with full disclosure of datasets, code, and training processes.
Sort: