Start / Install

Install

This page installs DOROTHY and checks that it works. DOROTHY needs Python 3.10 or later; pip installs PyTorch 2.4 or later, NumPy 2.0 or later and the other dependencies.

Step 1: Install PyTorch for your GPU

If you will train on an NVIDIA GPU, first install the PyTorch build for your CUDA version from pytorch.org. On a CPU or on Apple silicon, skip this step: pip installs a PyTorch that works there.

Step 2: Install DOROTHY

git clone https://github.com/joshspeagle/dorothy.git
cd dorothy
pip install -e .

This installs the dorothy package and the dorothy command. To also install the test and lint tools, run pip install -e ".[dev]" instead.

Step 3: Check the install

dorothy --version

The command prints dorothy and the version number.

Step 4: Check the device

python -c "from dorothy.utils import resolve_device; print(resolve_device())"

This prints the device that DOROTHY uses when the device is set to auto. It prints cuda for an NVIDIA GPU, else mps for an Apple silicon GPU, else cpu. If it prints cpu on a machine with an NVIDIA GPU, reinstall PyTorch as in step 1.

To choose a device, set device in the configuration, or pass --device to dorothy train, dorothy evaluate or dorothy predict (Commands).

Next

The quick start trains, evaluates and uses a model on a small synthetic catalogue, so it checks the whole install.