Train_loss is always nan for basic image classifier

[scarya]

I’m currently trying to train a simple classifier to distinguish between two different bunny rabbits. I’m following pretty much the exact same approach used with the bear classifier example, but when I attempt to train the model, my train_loss always shows up as nan:

Any suggestions on where I might be going wrong? The image directory structure is “bunnies/a/{images}” and “bunnies/b/{images}”, and I’m using a very small dataset of 25 images total.

[muellerzr (Zachary Mueller)]

The training data drops the last partial batch by default. As you have so small of images set your batch size to like 4 or 5

[scarya]

That worked! Thank you.