Show HN: Simulusion – A PyTorch library for training and running world models (github.com)

Simulusion, a newly launched PyTorch library originally known as TorchWM, has consolidated major world model families into a unified API for enhanced accessibility and experimentation in reinforcement learning. Users can effortlessly train various agents, including Dreamer, PlaNet, and IRIS, by simply adjusting configuration parameters while enjoying built-in support for popular environments like Gym and Unity ML-Agents. The library also features straightforward mechanisms for exporting to formats like ONNX and TensorRT, streamlining the application of trained models in different settings. This release is significant for the AI/ML community as it democratizes access to sophisticated world modeling techniques, making it easier for researchers and practitioners to explore, compare, and apply different algorithms without extensive coding efforts. Each model in Simulusion comes with demo capabilities that allow for rapid experimentation, utilizing modest hardware. Furthermore, the library's modular design enables users to customize and extend their architectures easily, ultimately accelerating the pace of innovation in model-based reinforcement learning.