This dataset consists of graph-based PyTorch .pt files representing CFD-converged flow cases over the NASA Common Research Model (CRM) in the transonic regime (Mach 0.60 to 0.95, Angle of Attack −6° to +18°). Each file encodes a flow condition as a batched graph using DGL. Designed for training graph neural networks (GNNs) and Transformer-based architectures. The dataset is derived from CFD .dat files, which provide pressure coefficient (Cp) values at each node of CRM's surface mesh under various flight conditions. Dataset Split: Train: 90 cases Validation: 18 cases Test: 13 cases Each .pt file contains: A batched DGL graph (from multiple zones of a triangular unstructured surface mesh) Node features: [X, Y, Z, Cp] — 3D coordinates and pressure coefficient Global features: [Mach, AoA] inferred from the filename (e.g., MP75_A3P3.pt → Mach 0.75, AOA 3.3°) Edge features : Euclidean distance between connected nodes Relative angle with respect to the global x-axis ---