The paper introduces MagNet, an equivariant deep-learning framework designed to represent density functional theory (DFT) total energy, atomic forces, and magnetic forces as functions of atomic and magnetic structures. The method incorporates the principle of equivariance under the three-dimensional Euclidean group into the neural network model, addressing the challenges posed by the subtle magnetic energy landscape and the difficulty of obtaining training data for magnetic materials. Through systematic experiments on various systems, including monolayer magnets, curved nanotube magnets, and moiré-twisted bilayer magnets of Cr$_3$, the authors demonstrate the method's high efficiency, accuracy, and exceptional generalization ability. MagNet shows promise for exploring magnetic phenomena in large-scale materials systems, particularly in studying spin dynamics and magnon dispersion. The work highlights the potential of deep learning in advancing ab initio atomistic simulations for magnetic materials.The paper introduces MagNet, an equivariant deep-learning framework designed to represent density functional theory (DFT) total energy, atomic forces, and magnetic forces as functions of atomic and magnetic structures. The method incorporates the principle of equivariance under the three-dimensional Euclidean group into the neural network model, addressing the challenges posed by the subtle magnetic energy landscape and the difficulty of obtaining training data for magnetic materials. Through systematic experiments on various systems, including monolayer magnets, curved nanotube magnets, and moiré-twisted bilayer magnets of Cr$_3$, the authors demonstrate the method's high efficiency, accuracy, and exceptional generalization ability. MagNet shows promise for exploring magnetic phenomena in large-scale materials systems, particularly in studying spin dynamics and magnon dispersion. The work highlights the potential of deep learning in advancing ab initio atomistic simulations for magnetic materials.