This file contains a set-to-sequence model trained on the PROCAT dataset of product catalogues. It consists of a Set Transformer module that encodes the set of product offers, and a permutation learning module in the form of a Pointer Network that then implicitly clusters them into complementary sections and outputs the target catalogue structure. References to the authors of the individual used models can be found in the paper describing the dataset, which is under review for the NeurIPS 2021 Dataset track as of June 2021, referenced in the repository with code for repeated experiments, which can be found here: https://github.com/mateuszjurewicz/procat The model requires using the appropriate version of PyTorch and importing the proper model definition, as shown in the provided jupyter notebooks.