Scoring is a main component of the PyRIT architecture. It is primarily used to evaluate what happens to a prompt. It can be used to help answer questions like:
- Was prompt injection detected?
- Was the prompt blocked? Why?
- Was there any harmful content in the response? What was it? How bad was it?
This collection of notebooks shows how to use scorers directly. To see how to use these based on previous requests, see the scoring orchestrator. Scorers can also often be used automatically as you send prompts.
There are two general types of scorers. true_false and float_scale (these can often be converted to one type or another). A true_false scorer scores something as true or false, and can be used in orchestrators for things like success criteria. float_scale scorers normalize a score between 0 and 1 to try and quantify a level of something e.g. harmful content.
Scores are stored in memory as score objects.
Before starting this, make sure you are set up and authenticated to use Azure OpenAI endpoints