The program is designed for building and managing a pre-training pipeline for neural network models using unsupervised self-learning based on contrastive representations, as well as supervised learning with flexible data management. The program forms visual representations of dermoscopic images for subsequent fine-tuning on classification tasks without complete sample labeling. It is used in research organizations for developing training methods on partially labeled medical data. Functional capabilities: configuration of independent augmentation sets with geometric, photometric, and color transformations; management of image sets from multiple sources; configuration of encoders with weight updates; automation of the training cycle with cosine scheduler and warm-up phase; saving checkpoints at specified intervals. Computer type: IBM PC-compatible. OS: Windows, Linux.