How to view completion inputs to API

Thinking of easier ways to debug my inputs to desired output.

Is there a way to view what inputs went into my completion? including the prompts and the user-inputs and engine settings for temperature etc? (JSON, plain text,)?

2 Likes

It’s not possible to have the parameters included in the response, afaik. You can see the default values in the documentation.

1 Like

What language / stack are you using? For example, I am running Python on AWS via a HTML/JS form.

I capture and log to CloudWatch my passed in parameters.

logger.info(
        f"|-o-| class: , func: make_story, line: 477 :: {story_time} \n story : {story} :: {response}"
    )

You can also set echo to True to see the full prompt

Excelsior!
Drew

3 Likes

hmm…ok will keep that in mind. thanks

1 Like

For /classifications and /answers you can set return_prompt to ‘true’. If set to true , the returned JSON will include a “prompt” field containing the final prompt that was used to request a completion.

4 Likes