Samples and metadata supporting 'Architecture-Based Energy Consumption Dataset'

The Architecture-Based Energy Consumption Dataset (ABCD) is a large-scale experimental dataset capturing power, current, and voltage measurements collected during the execution of deep neural network architectures under diverse configurations.The dataset comprises raw per-run measurement files and processed, aggregated summaries that encode architectural topology, training parameters, and derived statistical descriptors of energy-related behaviour. It was generated as part of a PhD thesis at Newcastle University and is intended to support research on energy-aware machine learning, architectural generalisation, and performance prediction for deep neural networks, particularly in resource-constrained or edge-computing contexts.Due to the size of the full dataset, this record provides representative samples and comprehensive metadata. The complete dataset is available in Zenodo via the DOI below, while the supporting code can be accessed in GitHub.