The output of the model is not as good as examples. I'm wondering if it's possible to get the dataset that can train a better model.