Welcome to Magrittetorch, a modern software library for simulating radiation transport, using pytorch as backend for GPU computations. This is a port of Magritte to python, to allow for a redesign of the parallelization scheme. Magrittetorch is currently mainly used for post-processing hydrodynamical simulations of astrophysical models by creating synthetic observations.
Magrittetorch can be used for 3D (N)LTE line radiative transfer and imaging. The software library supports both CPU and GPU parallelization, using pytorch as backend. Model input can be done using astropy Quantities to allow for effortless unit conversions.
Magrittetorch is entirely written in python, and therefore pip installable.
pip install git+https://github.com/Magritte-code/Magritte-torch.gitFind the complete installation instructions in the documentation
The documentation of Magrittetorch can be found here. See the examples in the documentation to learn how to use Magrittetorch to create synthetic observations.
TODO: compare CPU performance vs Magritte, compare GPU performance, scaling behavior
Please report any issues with Magritte-torch here. If you need any further help, please contact Thomas Ceulemans.
Please contact the authors of the papers referenced above if you want to use Magrittetorch in your research. We are currently working on documentation and examples to facilitate its independent use. Until then, please contact us.
Developers
- Thomas Ceulemans
Scientific & Technical advisors
- Frederik De Ceuster
- Leen Decin
Contributors
- Silke Maes
- Jolien Malfait
- Mats Esseldeurs
- Arnout Coenegrachts
- Owen Vermeulen