Halide perovskites are promising materials for solar cells; however, a critical issue hindering their practical application is their stability. To facilitate improvements in stability through modeling, it is essential to comprehend the dynamics at the atomic scale. Given the complexity of these systems, especially in the case of materials containing organic molecules, like FAPI, understanding them can be significantly enhanced by employing neural network-based methods. This project focuses on the training and optimization of Neural Evolution Potentials (NEPs) for this purpose. Utilizing pre-generated CPU-based data, we will leverage GPU resources primarily for the training of NEPs. The GPUs will accelerate the iterative refinement of the NEPs, ensuring both high accuracy and efficiency. By improving the predictive power of our NEPs, we aim to gain crucial insights into the stability mechanisms of halide perovskites, thereby advancing their viability in solar cell applications.