ORDerly Transformer Models for chemical tasks

Transformer models trained on tasks in organic chemistry on ORDerly benchmark datasets.ORDerly_retro: Retrosynthesis prediction (prediction reactants given a desired product)ORDerly_forward_separated: Forward reaction prediction (predict reaction products given reactants, solvents, and agents), with reactants separated by > from the solvents and agents in the reaction string.ORDerly_forward_mixed: Forward reaction prediction (predict reaction products given reactants, solvents, and agents), with reactants, solvents and agents mixed together in the reaction string.non-uspto-eval: Evaluation of transformer models trained on USPTO data on non-uspto data available in the Open Reaction Database.Full details can be found in our paper: https://chemrxiv.org/engage/chemrxiv/article-details/64ca5d3e4a3f7d0c0d78ca42Neurips workshop paper: https://openreview.net/forum?id=R8FQMsECISCode: https://github.com/sustainable-processes/orderlyThe supplementary datasets used for this work can be found here: https://doi.org/10.6084/m9.figshare.23502372.v3Transformer model architecture is from Molecular Transformer: https://pubs.acs.org/doi/10.1021/acscentsci.9b00576Find the results, models, and checkpoints within MolecularTransformer/experiments. Note that the "wandb" folder was deleted since figshare only allows uploads up to 500 files.Notes:There's a limit of 500 files in figshare, so I deleted the the "docs" and "onmt", and "OpenNMT_py.egg-info" , and "tools" folders from all folders except "ORDerly_retro". I also deleted all wandb-associated files and all checkpoint files.Empty files cannot be uploaded to figshare, so you have to create these yourself, where appropriate (e.g. MolecularTransformer/onmt/tests/__init__.py and non-uspto-eval/MolecularTransformer/experiments/models/ofs_1.pt).Feel free to email me, Daniel Wigh, at [e-mail] or [e-mail] or my supervisor Alexei A. Lapkin.