Add SPE tokenizer (1125 SMILES subword tokens) + <|start_of_smiles|>/<|end_of_smiles|> special tokens. Trained on 2M ZINC20 + 2M ChEMBL canonical SMILES. SPE vocab_size=1000, min_freq=4000.

f6eb1d8

verified

Codemaster67 commited on 11 days ago