humarin

/

chatgpt_paraphraser_on_T5_base

like 145

Using the parameters provided in the model card i am generating a different sentence compared to the sentences generated using the demo (inference api). It would be usefull if you are able to provide the exact parameters used for the demo (inference api)

Input: The Confidential Information disclosed under this agreement will be kept by the Disclosing Party and destroyed by either party upon its request

Output obtained by running model using the paramters mentioned in the model card: max_length=128, num_beams=5, num_return_sequences=5, num_beam_groups=5, repetition_penalty=10.0, diversity_penalty=3.0, no_repeat_ngram_size=2, temperature=0.7 is
Confidential Information disclosed under this agreement will be kept by the Disclosing Party and destroyed by either party upon its request.

But the output for the same sentence when given in the demo (inference api) is Confidential Information disclosed under this agreement will be destroyed by either party upon its request, and the Disclosing Party agrees to hold onto it for as long as is necessary to destroy it.

And another things i noticed while going through the config.json is that there are no configurations related to the paraphrase task. there are configurations related to other tasks like translation, summarization etc. Could you let me know is this the reason the difference in the output obtained or is it some other reason and why there are no configurations related to paraphrase.