Hi

Thank you for this straight forward codes for us to practice fine-tuning models with ViT. I'm on a project that classifying book covers. Everything goes well on my own dataset until trying to define classifier. An error message popped up "Can't instantiate abstract class Classifier with abstract methods forward". As I looked up the pytorch lightning code about forward(), it is indeed an abstract class. No idea why the error since the method was initiated in the init() method.

Has anyone run into this issue yet?

Hi,

The image search at https://huggingface.co/api/experimental/images/search returns max 35 results. Values lower than 35 in the "count" parameter are honored; values above 35 are ignored and 35 results are returned. Your great tutorial suggests that 150 results can be gathered through this API for each query. Has the image search API changed? Thank you!

I'm having an issue on fitting the model. Given your example, HuggingPics works just fine. However, when I attempted to train my own model with one class with iron man, I am having issues under the Training section, cell 2, in particular

pl.seed_everything(42)
classifier = Classifier(model, lr=2e-5)
trainer = pl.Trainer(gpus=1, precision=16, max_epochs=4)
trainer.fit(classifier, train_loader, val_loader) # ERROR HERE

I tried to pin point the issue, but it was to no avail. First, I attempted to convert the encoding to a float in ImageClassificationCollator. However, that threw a new error for the same line,

ValueError: The target has to be an integer tensor.

I thought the error could be because of not enough classes, but that wasn't the case. I also thought it was because there wasn't enough data, but I I lowered the image count and your example processed fine.

Hi, it seems like HfApi.create_repo() parameters are updated and no longer treat 'name' as valid parameter. Made this PR to solve error when pushing model to huggingface hub

TypeError: create_repo() got an unexpected keyword argument 'name'