Link to SAFE-GPT Training Data is invalid

[jasonlin90125]

The link to the Hugging Face dataset for training is invalid. I don't think it's on Hugging Face any more...

[nvdreidenbach]

It looks like they changed it to https://huggingface.co/datasets/datamol-io/safe-drugs will update readme thanks!

[jasonlin90125]

The safe-drugs dataset only has 26 molecules from their original benchmarking

[nvdreidenbach]

Ok I will leave this open. It may make sense to open an issue on their repo since they were the publishers of the data.

In the meantime you can take any SMILES dataset and convert it to SAFE to train on.