How to add custom loss using uni-core package

Thank you for providing this great pre-train model. We want to extend it to a downstream task based on your pre-trained model.
Is there an easy way to add a custom head with a different loss using the uni-core package? It would be very helpful if you can provide an example script of how to achieve this.

So it seems I have to define a UnicoreLoss with a forward method take into **model, sample** arguments and return **loss, sample_size, logging_output**.
And need to define a UnicoreTask class to define the training process. However, I am not able to find where I can define the input data and the custom head, can you point out where I can define the input data format, i.e. the sample argument in the loss class, and also implement a customized head?

It would be very helpful if there is a minimum usage case to use the Uni-core package. Thanks!