Analysis dataset used for the report "Deriving Constraints on di-Higgs SMEFT couplings with Neural Simulation-based Inference" (arXiv link). We provide dihiggs signal events gg -> hh -> (bb)(γγ) and QCD background events pp -> bbγγ at both a 14 TeV and 100 TeV collider setup. All events are generated with MadGraph 3.5.1, Pythia 8.306, and Delphes 3.5.0. For the hard process (pp -> hh) for the signal events, we use the SMEFT@NLO MadGraph model to incorporate the effects of three dimension-6 SMEFT operators (c_p, c_{dp}, c_{tp}). We select 10 values ("morphing basis points") for the vector (c_p, c_{dp}, c_{tp}) (see Table IV of the accompanying report). At each morphing basis point, we generate a number of signal events; for each event, we also calculate the weights to the other 9 basis points. All of this data is stored in the .h5 files. We use MadSpin to carry out the higgs decays. These datasets can be used as input into the MadMiner SampleAugmenter function, which uses the weights associated with the signal events to generate detector-level events at arbitrary (c_p, c_{dp}, c_{tp}) points. All scripts used to generate and process this data are available at this repository. Please refer to the accompanying paper for more details on the analysis.