I had a dataset with 69,000 Persian news articles with categories. It was decided to provide a portion of the dataset and ask for a model to be trained on it, and then rank the models based on accuracy (we have our own test data).