TorchRL provides highly modular and extensible approach to experimenting with Reinforcement Learning. It allows for a registry based approach to running experiments, allows easy checkpointing, and updating hyper parameter sets. All this is accessible via a programmatic interface.

• Modularity in the RL pipeline
• Clean implementations of fundamental ideas
• Fast Experimentation
• Scalability
• Low bar and High ceiling

pip install torchrl

Install from source for the latest changes that have not been published to PyPI.

pip install https://github.com/activatedgeek/torchrl/tarball/master