I saw in your analysis/Results.ipynb that you loaded
load checkpoint and args
args_path = '/home/logs/metabo/bf6b607124c5cca3430fc0c2ee1148dd.args'
args = Namespace(**pickle.load(open(args_path,'rb')))
You didn't provide this cache. I don't need to train from scratch; I just want to see if your clipzyme_model.ckpt has such performance on the test set.
In the project you provided, I didn't find any file with a number of samples similar to the test set number mentioned in your paper.
Can you help me? It would be even better if you could turn your environment into a Docker image.