Transformation only on train set

Can I create a transform that only applied to the training set but not the validation set

If you are creating a new transform then you can specify its split_idx

if split_idx == 0 then it will only be used by the training set.

And if it inherits from RandTransform then it will have this behaviour by defualt: https://docs.fast.ai/vision.augment#RandTransform