A modular deep reinforcement learning framework in PyTorch for research and application, featuring ready-to-use algorithms and reproducible experiments.
SLM Lab is a modular deep reinforcement learning framework built in PyTorch that provides ready-to-use algorithms and tools for training AI agents. It solves the problem of complex and non-reproducible RL experimentation by offering a structured, configuration-driven approach. The framework is also the companion library for the book "Foundations of Deep Reinforcement Learning."
Researchers, students, and practitioners in reinforcement learning who need a reproducible and easy-to-configure framework for experimenting with RL algorithms. It is particularly useful for those working with PyTorch and Gymnasium environments.
Developers choose SLM Lab for its modular design, extensive algorithm support, and strong emphasis on reproducibility. Its integration with cloud services and automatic analysis tools reduces the overhead of managing RL experiments, making it efficient for both learning and advanced research.
Modular Deep Reinforcement Learning framework in PyTorch. Companion library of the book "Foundations of Deep Reinforcement Learning".
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Includes validated implementations of PPO, SAC, DQN, and others on 70+ environments, reducing implementation time for common RL tasks.
Saves spec files and git SHAs for each run, ensuring exact experiment reproduction as highlighted in the README's reproducibility feature.
Built-in support for dstack GPU training and HuggingFace result sharing streamlines cloud workflows, as detailed in the Cloud Training section.
Uses JSON spec files to define experiments, eliminating code changes and making it easy to manage multiple runs, per the Easy Configuration feature.
Major updates like v5.0 to Gymnasium break compatibility with the companion book's code, requiring users to manually check out older versions, as noted in the README warning.
Heavily relies on specific tools like uv, dstack, and HuggingFace, which may not integrate well with existing pipelines or preferred alternatives.
Requires multiple steps for installation, cloud configuration, and dependency management, including uv tooling and .env file setup, as seen in the Quick Start and Cloud Training sections.