Returning logits for a user provided continuation

I would like to use an extension of the continuation and chat/continuation API.

The new API would have two main input fields:

  • prompt string
  • user provided continuation string

The return from the API would be a numeric logit value of the user provided continuation, given the prompt.

One of the use cases is calculating numeric scores of the strength of presence of particular themes in the prompt. In this case the continuation string field would contain a particular theme label for which the score is requested. Related use cases are psychometrics as well as sentiment analysis.

For example:

API request:
{
“prompt”: “A story follows. Please evaluate the tone of Person A in that story. … The Story …”,
“continuation”: “Person A in that story exhibits a friendly expression style.”
}
or alternatively:
{
“prompt”: "A story follows. Please evaluate the tone of Person A in that story. … The Story … The expression style of the Person A can be characterized as: ",
“continuation”: “Friendly.”
}

API response:
{
“logit”: 0.7
}