A collection of 10 Z'/QCD jet datasets used in Meta-learning and data augmentation for mass-generalised jet taggers. Each dataset contains 500K Z' jets and 500K QCD jets. Simulation Events are simulated with MadGraph5 2.8.0, showered in Pythia 8.244, then passed to Delphes 3.4.2 with the default CMS card. For the signal, we use a simple Z' model from FeynRules. In each event, the leading \(R=1\) anti-kt jet is selected. Cuts Parton-level cuts of \(|\eta_J|, |\eta_{Z'}|<2.0\) and \(\not\!\!E_T>1.2\) TeV are applied in MadGraph. Cuts of \(|m_J-m_{Z'}|<m_{Z'}/4\) and \(p_{T}>1.2\) TeV are applied after detector simulation. Preprocessing Jets are preprocessed by first translating all constituents in the rapidity-azimuth plane such that the leading constituent lies at the origin. A rotation is then applied to position the centre of momentum below the origin. Format Each jet is stored as a list of constituent information in the format \((p_T/p_{T,J}, \eta', \phi', g(\texttt{pdg_id}))\) where \(\eta'\) and \(\phi'\) respectively denote pseudorapidity and azimuthal angle coordinates after the mentioned transformations and \(g\) maps Particle Data Group Monte Carlo identifiers to small floats. Specifically: Particle \(\gamma\) \(e^+\) \(e^-\) \(\mu^+\) \(\mu^-\) Neutral Hadron \(\pi^-\) \(\pi^+\) \(K^-\) \(K^+\) \(\bar{p}\) \(p\) pdg_id 22 -11 11 -13 13 0 -211 211 -321 321 -2212 2212 g( pdg_id ) 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.95 1.05 1.15 The arrays are serialised and saved in TFRecord format allowing for efficient interfacing with Tensorflow. To avoid memory issues associated with loading a large number of datasets, each dataset is split into 100 shards.