Which classifier to use in custom_trainer.py for controllable generation?

[CSerxy]

Hi @XiangLi1999 ,

Thanks for open-source your excellent work. I learnt a lot from your answers.

I have a question regarding conditional generation. You gave 4 examples about how to train controllable generation in curtum_trainer.py, however, if I want to train a style generation with only two labels, which of the examples shall I use?
Is it okay to use Semantic Content one? In this task, the label is from all vocabulary while in my case, my label only has 0 and 1.

Many thanks,
Zhuofeng