Installation Guide

da4ml is available on PyPI as binary wheel and can be installed using pip.

pip install da4ml

Requirements

Binary Wheels

  • python>=3.10 (3.13+ is recommended)

  • numpy>=2.0

  • Linux on x86_64 or MacOS on ARM64 platform

Building from Source

  • python>=3.10 (3.13+ is recommended)

  • numpy

  • A C++20 compliant compiler with OpenMP support

  • meson-python>1.9.0

git clone https://github.com/calad0i/da4ml.git
pip install ./da4ml

Alternatively, you can configure and build da4ml with meson directly:

meson setup build/cp31*
meson compile -C build/cp31*

Warning

If you are building an editable installation, the flag [--no-build-isolation] must be used. Since editable installation with meson is implemented by dynamic hooking of the build_ext command to recompile the C++ extension in-place, the build isolation will break the dynamic hooking and cause ninja to fail upon importing. You must also install meson-python manually before running pip in this case.

Note

Technically, you do not need numpy>=2.0, since that is due to the use of nanobind’s abi3 feature. However, since we do not see any good reason for using older numpy versions, we require numpy>=2.0 for the binary wheels for simplicity. If you want to build from source with older numpy versions, you can modify the meson.build file to override the limited_api flag.