GPT2 Zinc 87m This is a GPT2 style autoregressive language model trained on ~480m SMILES strings from the ZINC database. The model has ~87m parameters and was trained for 175000 iterations with a batch size of 3072 to a validation loss of ~.615. This model is useful for generating druglike molecules or generating embeddings from SMILES strings How to use To generate molecules: To compute embeddings: WARNING This model was trained with bos and eos tokens around SMILES inputs. The GPT2TokenizerFast tokenizer DOES NOT ADD special tokens, even when add special tokens=True . Huggingface says this is intended behavior. It may be necessary to manually add these tokens Model Performance To test generation performance, 1m compounds were generated at various temperature values. Generated compounds were checked for uniqueness and structural validity. percent unique denotes n unique smiles/n total smiles percent valid denotes n valid smiles/n unique smiles percent unique and valid denotes n valid smiles/n total smiles temperature percent unique percent valid percent unique and valid : : : : 0.5 0.928074 1 0.928074 0.75 0.998468 0.999967 0.998436 1 0.999659 0.999164 0.998823 1.25 0.999514 0.993…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy