skandermoalla / torchrl

Provides a development environment to develop on pytorch/tensordict and pytorch/torchrl.

TorchRL Development Environment

Overview

A development environment for pytorch/tensordict and pytorch/rl.

Code and development environment

• Docker/OCI-container for AMD64 machines (+ NVIDIA GPUs): This option works for machines with AMD64 CPUs and NVIDIA GPUs. E.g. Linux machines (EPFL HaaS servers, VMs on cloud providers), Windows machines with WSL, and clusters running OCI-compliant containers, like the EPFL Run:ai (Kubernetes) clusters.

  Follow the instructions in installation/docker-amd64-cuda/README.md to install the environment.

• Conda for osx-arm64 This option works for macOS machines with Apple Silicon and can leverage MPS acceleration.

  Follow the instructions in installation/conda-osx-arm64-mps/README.md to install the environment.

Contributing

We use pre-commit hooks to ensure high-quality code. Make sure it's installed on the system where you're developing (it is in the dependencies of the project, but you may be editing the code from outside the development environment. If you have conda you can install it in your base environment, otherwise, you can install it with brew). Install the pre-commit hooks with

# When in the PROJECT_ROOT.
pre-commit install --install-hooks

Then every time you commit, the pre-commit hooks will be triggered. You can also trigger them manually with:

pre-commit run --all-files