Request: pre-trained models weight checkpoints

[Tigerrr07]

Hi! I wonder if you could provide the weight checkpoints of the pretrained models, because I need them for comparison.

I tried to train the pretrained model from scratch on the LINCS data, but I found some missing hyperparameters in the config file for the first experiment. Even though the training was successful when I fixed the previous problem, the training stopped after several epochs with a NaN error during the full evaluation.

Besides, I noticed that you specify model.pretrained_model_hashes in the following config files, like this one in config_sciplex_finetuning_hparam.yaml:

  model.pretrained_model_hashes: # seml config_hashes for the pretrained models for each embedding. Used for loading model checkpoints. Hashes taken from `analyze_lincs_all_embeddings_hparam.ipynb`
      grover_base: ff420aea264fca7668ecb147f60762a1
      MPNN: ff9629a1b216372be8b205556cabc6fb
      rdkit: 4f061dbfc7af05cf84f06a724b0c8563

But I couldn’t find the corresponding model checkpoints.

It would be greatly appreciated and convenient if you could provide them!

[MxMstrmn]

@Tigerrr07,

What model checkpoint do you need exactly? For the LINCS data?