The PyTorch Ecosystem Working Group announces ten new projects joining the PyTorch Ecosystem Landscape: Perforated (data-efficiency training library), AReaL (async RL infrastructure for LLM/VLM agents), TorchJD (multi-loss training via Jacobian Descent), RLinf (RL framework for embodied/agentic AI), Miles (large-scale post-training framework by RadixArk), SMG (Rust-based model-routing gateway), FiftyOne (multimodal data platform for physical AI), TokenSpeed (LLM inference engine separating control and execution planes), VisualTorch (PyTorch model visualization library), and TorchSurv (survival analysis toolkit used in FDA's regulatory science catalog). Each project gets a short description of its purpose, architecture, and links to learn more, along with instructions on how other projects can apply for inclusion in the landscape.
