Paraphrasing with OpenNMT-tf?

I am thinking about training an OpenNMT-tf model for paraphrasing in English.

  1. I am curious if anyone has had done so, and what were the results.
  2. Also, which model type would be the best to do it with?
    Thank you.

Hello Igor!

What I tried was using the Random Sampling option with CTranslate2 during translation. It works well for a bilingual model to generate multiple suggestions.

You can also try a fine-tuned version of T5 for paraphrasing:

Other resources:

1 Like

Thank you, Yasmin, for sharing your experience with cTranslate and for the pointers.
I am aware of the models in Pytorch and I tried running them on TorchServe, but they were way too slow.
That’s why I am looking for something in TensorFlow.