An RL accelerator framework with reusable types for taking an existing RL training loop and achieving maximum utilization of existing computational resources.
Create a new conda environment for this project and activate it.
conda create -n parllel python=3.9
conda activate parllel
Install pytorch (or ML framework of your choice, coming soon). The process depends on your hardware, but some common cases are handled by installing yml files.
Linux with CUDA 11.3+: conda env update --file torch_cuda11.yml
Mac OS on Apple Silicon: conda env update --file torch_m1.yml
Install other requirements.
pip install -r requirements.txt
Install parllel repo itself.
pip install -e .
To run the examples, you must also install the development requirements.
pip install -r requirements_dev.txt
If you already had hera-gym installed in development mode, you will now need to reinstall it, as it has been replaced by a fresh copy of hera_gym from gitlab.