Liquid chromophores constitute an important class of molecules, with applications from solvent free dyes to increasing the efficiency of solar cells, owing to their interesting optical properties. However, to further understand and model these optical properties, one must first understand the structural and in particular dynamical properties of liquid chromophores at the molecular level, which at the moment are not well understood. In this project we aim to develop neuro-evolution potentials (NEP), a type of neural network-based machine learning force field, for these chromophore systems. NEPs are highly accurate and efficient, which enables inferring the dynamics of large systems, up to millions of atoms, larger than what is possible with density functional theory (DFT). GPU-acceleration is fundamental to this project as it enables faster development of our force field models, but it also unlocks the crucial ability to train an ensemble of models for a particular system. Such an ensemble can then be used to gauge the uncertainty of the model through, for instance, bagging. Uncertainty estimates are important for validating the model, and can also be used to improve the model through active learning, in which the model is automatically retrained when it encounters structures upon which it hasn't been trained.