Simulations of energetic materials (EM) typically require vast computational resources due to the extreme dynamics such as strong shocks and nonlinear reactions. Artificial Intelligence (AI) has been shown to be capable of reduction in computation time by thousand folds with minimal impact on accuracy. While previous research in AI acceleration were focused on strong shock loading conditions, there is little work on weak-to-moderate shocks, crucial for safe storage, handling, and understanding deflagration-to-detonation transitions. This work promotes a novel deep learning algorithm, Physics-aware Recurrent Convolutional Neural Network (PARCv2), capable of modeling EM thermo-mechanics in weak shock regimes. We show that the model is capable of predicting shock patterns, hotspot formation, and shear band formation to high accuracy, as well generalizing into unseen initial conditions. We explore the model’s limitations of reduced accuracy in finer details or low impact velocity, and propose a number of avenues of improvements that would guide designs of more accurate and efficient physics-informed machine learning models.