Hi, I am fine-tuning pretrained models from huggingface the trainer method. During the training I generate checkpoints which I want to access via ClearML. It correctly uploads trainig_args.bin, optimizer.pt , rng_state.pth, scheduler.pt , which I see in the ClearML interface. But there are other files missing (e.g. tokenizer.json), which are created locally. How can I make those files accessible as Artifacts?
Hi @<1582542029752111104:profile|GorgeousWoodpecker69> , I think you can also manually register them using Task.upload_artifact None
Thanks. That would mean I cannot abort the training process, but have to wait until its finished to call the upload_artifact? Is there a reason why it's not logging all the generated files?
