Skip to content

Installation

AnyPINN requires Python 3.10 or later.

Install the CLI

The anypinn CLI is the main entry point. Install it globally so it's available from any directory:

uv tool install anypinn

pip install anypinn

Verify the installation:

anypinn --version

Use uvx for one-off runs

If you don't want to install globally, you can run the CLI directly with uvx:

uvx anypinn create my-project

Dependencies

Each project scaffolded by anypinn create includes its own pyproject.toml with pinned dependencies. Running uv sync inside the project directory installs everything you need:

  • PyTorch: tensor computation and automatic differentiation
  • NumPy / SciPy: numerical utilities
  • torchdiffeq: ODE integration for synthetic data generation
  • PyTorch Lightning: optional, included when --lightning is selected
  • Pandas: CSV data loading
  • Matplotlib / Seaborn: result visualization

You do not need to install these manually. The generated project handles it.

GPU Support

AnyPINN uses PyTorch under the hood. If you have a CUDA-capable GPU, install the appropriate PyTorch build for your system. See the PyTorch installation guide for platform-specific instructions.

When using Lightning, GPU training is automatic: the Trainer detects available hardware and uses it.