Skip to content
  • Installation
  • Using uv

Installing CAREamics with uv#

Follow uv installation guidelines to use uv.

Important

Please refer to the installing PyTorch section of uv documentation should you encounter issues running on GPU. In particular it may be necessary to install torch independently in the venv on Windows (see this section)

Using notebooks#

To run notebooks, install juv:

Installing juv
uv tool install juv

Running CAREamics examples#

CAREamics-examples contains multiple notebook examples that can be directly run with uv:

Run a notebook
juv run notebook.ipynb

Running your own notebook#

To run your own notebook, we recommend adding PEP723 metadata:

Add CAREamics dependency
juv add my_notebook.ipynb "careamics[examples]"
juv run my_notebook.ipynb

You can also pin the CAREamics dependency in order to improve reproducibility:

Pin CAREamics dependency
juv add my_notebook.ipynb "careamics[examples]==0.0.16"

CAREamics as a dependency#

In this section, we will describe creating a project with CAREamics as a dependency, using standard uv:

uv init careamics_project
cd careamics_project
uv add "careamics[examples]>=0.0.16"  # (1)!
  1. Pin the dependency for reproducibility

Installing CAREamics with GPU on Windows#

This section walks you through creating a CAREamics-based projects using GPU on Windows.

  1. Create a pyproject.toml in your project folder:

    pyproject.toml
    [project]
name = "myproject"
version = "0.1.0"
description = "Running CAREamics on Windows with GPU"
readme = "README.md"
requires-python = ">=3.11"
dependencies = [
    "torch>=2.0,<=2.9.0", # (1)!
    "torchvision<=0.24.0",
    "careamics[examples]>=0.0.18",
]

[tool.uv.sources]
torch = [
    { index = "pytorch" },
]
torchvision = [
    { index = "pytorch" },
]

[[tool.uv.index]]
name = "pytorch"
url = "https://download.pytorch.org/whl/cu128" # (2)!
explicit = true

    1. The constraints here will change with time, and are given as indications here.
    2. This will depend on your system (especially GPU driver), check the PyTorch instructions.

  2. Test GPU accessibility by running the following script: 

    uv run python -c "import torch; print([torch.cuda.get_device_properties(i) for i in range(torch.cuda.device_count())])"

  3. Add your training script or notebook to the project and run it with uv (see previous sections).