Rephrasal extension

[fedenanni]

Currently for a prompt we generate one rephrased. It would be interesting to scale this up, however the main challenge is automatically parsing the output (and ensuring consistency across LLMs).

If we use a prompt like the following:

Rephrase the sentence "{INPUT_PROMPT}" five times and provide the output strictly as a JSONL object, one sentence per line. Do not include any additional text.

First we can check if the obtained output is a JSONL object (otherwise return a message ot the user), then if the number of element is required number of variations and only then we produce the output.

What do you think? Any other format which is better?