FTorch

Brief description

It is desirable to be able to run machine learning (ML) models directly in Fortran. ML models are often trained in some other language (say, Python) using a popular frameworks (say, PyTorch) and saved. We want to run inference on this model without having to call a Python executable. To achieve this we use the existing Torch C++ interface, LibTorch.

FTorch provides a library enabling a user to directly couple their PyTorch models to Fortran code. There are also installation instructions for the library and examples of performing coupling.

We support running on both CPU and GPU, and have tested the library on UNIX and Windows based operating systems

Publications and Presentations

FTorch is published in JOSS. To cite it in your work please refer to:

Atkinson et al., (2025). FTorch: a library for coupling PyTorch models to Fortran. Journal of Open Source Software, 10(107), 7602, https://doi.org/10.21105/joss.07602

The following presentations contain information about FTorch:

• Coupling Machine Learning to Numerical (Climate) Models
  Platform for Advanced Scientific Computing, Zurich - June 2024
  Slides
• Blending Machine Learning and Numerical Simulation, with Applications to Climate Modelling
  Durham HPC days, Durham - May 2024
  Slides
• Reducing the overheads for coupling PyTorch machine learning models to Fortran
  ML & DL Seminars, LSCE, IPSL, Paris - November 2023
  Slides - Recording
• Reducing the Overhead of Coupled Machine Learning Models between Python and Fortran
  RSECon23, Swansea - September 2023
  Slides - Recording

License

The FTorch source code, related files and documentation are distributed under an MIT License which can be viewed here.

Projects using FTorch

The following projects make use of FTorch.
If you use our library in your work please let us know.

• DataWave CAM-GW - Using FTorch to couple neural net parameterisations of gravity waves to the CAM atmospheric model.
• MiMA Machine Learning - Implementing a neural net parameterisation of gravity waves in the MiMA atmospheric model. Demonstrates that nets trained near-identically offline can display greatly varied behaviours when coupled online. See Mansfield and Sheshadri (2024) - DOI: 10.1029/2024MS004292
• Convection parameterisations in ICON - Implementing machine-learnt convection parameterisations in the ICON atmospheric model showing that best online performance occurs when causal relations are eliminated from the net. See Heuer et al (2024) - DOI: 10.1029/2024MS004398
• In the GloSea6 Seasonal Forecasting Model - Replacing a BiCGStab bottleneck in the code with a deep learning approach to speed up execution without compromising model accuracy. See Park and Chung (2025) - DOI: 10.3390/atmos16010060