zip(..., strict=False) wreaks havoc

[finlayiainmaclean]

Thanks for the great repo!

https://github.com/pauling-ai/Protonify/blob/e3a3b3650b63252931cd8652c391d536e63ce9b8/protonify/FreeEnergyPredictor.py#L36C1-L52C20

   def predict(self, smiles_list):
        unimol_input = self.preprocess_data(smiles_list)
        dataset = MolDataset(unimol_input)
        dataloader = DataLoader(dataset, 
                                batch_size=self.batch_size, 
                                shuffle=False,
                                collate_fn=self.model.batch_collate_fn,
                                )

        results = {}
        for batch in dataloader:
            net_input, _ = self.decorate_torch_batch(batch)
            with torch.no_grad():
                predictions = self.model(**net_input)
                for smiles, energy in zip(smiles_list, predictions):
                    results[smiles] = energy.item()
        return results

zip(smiles_list, predictions, strict=False) means that if smiles_list > batch_size, zip silently truncates to len(predictions) and reuses the first batch_size SMILES for every batch.

See original issue in the official unipka repo.