This model is a paraphraser designed for the Adversarial Paraphrasing Task described and used in this paper: https://aclanthology.org/2021.acl-long.552/.
Please refer to nap_generation.py on the github repository for ways to better utilize this model using concepts of top-k sampling and top-p sampling. The demo on huggingface will output only one sentence which will most likely be the same as the input sentence since the model is supposed to output using beam search and sampling.
Github repository: https://github.com/Advancing-Machine-Human-Reasoning-Lab/apt.git