TGLF neural network
Neural network surrogate models of the TGLF quasilinear plasma turbulent transport simulator in various parameter spaces.
Overview
Neural networks as surrogate models that emulate the trapped gyroLandau
fluid (TGLF) quasilinear plasma turbulent transport simulator and that is valid in the operating space of the Spherical Tokamak for Energy Production (STEP).
Who it’s for
TGLF neural network is for the integrated modelling community.
Software details
The TGLF neural networks are trained on the experimental space or the theoretical space of current and future devices. They have been trained using the ADEPT codebase which is not open source. The training set will not be open sourced at this stage.
We wish to open source the model weights (the output of the training exercise) and metadata about the training space and inputs and outputs so that the surrogates can be coupled to and evaluated in external applications.
Licence
Open Source
Release date
21 July 2025
Programming language
Python
Dependencies
The weights could be read in JAX or pytorch.
Compatible platforms
Linux / Windows / Mac