Vision_learner based on Resnet34 shows always accuracy of 0 and error rate of 1

I’m training a classification model based on Resnet34 to classify satellite images based on one of three classes. Loading the data and setting up the model works fine, I can also train the model successfully. However, the metrics are always constant (tried only accuracy and both accuracy and error_rate). I use these commands:

model = vision_learner(dls = data, arch = models.resnet34, metrics = accuracy)
model.fine_tune(epochs = 1)

In addition, fine_tune doesn’t stop after 1 epoch. Could or is this related to the issue with the metrics?

I’m using fastAI 2.7.12 and PyTorch 2.0.1 on an Apple M1 and the code is running in Jupyter Notebook.