Installation

FloDR is on PyPI and needs Python 3.11 or newer. It trains with PyTorch and uses the GPU when one is available.

pip install flodr

With uv, add it to a project with uv add flodr or install it into the current environment with uv pip install flodr.

GPU

pip installs the PyTorch build of the package index, which may not match your CUDA driver. For a specific CUDA version install PyTorch first, following pytorch.org, and FloDR after it.

import flodr

flodr.default_device()  # "cuda" when PyTorch sees a GPU, otherwise "cpu"

FloDR runs on the CPU as well, several times slower. See Performance.

From source

For development, clone the repository and install it with its locked dependencies and the test tools.

git clone https://github.com/daenuprobst/flodr
cd flodr
uv sync --extra dev
uv run pytest

The tests run on the CPU and take a few minutes.