RSGPT for synthetic data

We proposed Retro Synthesis Generative Pre-Trained Transformer (RSGPT), a retrosynthesis planning model , leveraging the architecture of LLaMA2. In this context, frags_dic.pkl is provided. This is a part of the {Reactants in the template: [Matched molecular fragments]} that we use in the data generation stage for reference, so that you can use your own templates and molecular fragment libraries to efficiently generate data. Our submolecules from PubChem, ChEMBL, and Enamine datasetes are provided in the .txt files.