I am trying to furthur finetune this model so the similarity score between input text and the output is as diffrent as possible.
#4
by
affan14
- opened
Can you provide the fine-tuning script and the dataset that you used for finetuning this paraphrasing task?
Also is there a way for making the input and output text different from each other through some arguments?
You can use the temperature, the higher the temperature, the more difference between the input and output sentences. You can also use do_sample=True argument to sample the less similar outputs.
model.generate(**batch, max_length=60, num_beams=10, num_return_sequences=10, temperature=2.0, do_sample=True)