Hugging Face
Scoring choices instead of generating answers
This week’s Hugging Face article explores decision models for classification and routing. A model receives context, a question and possible answers, then returns a probability distribution over those choices.
For example, a support request can be assessed separately for its destination team, urgency and whether it includes a refund request. Each decision can be made without generating a lengthy response.
An approach for specific tasks
This works best when the set of possible answers is known in advance. The guide examines examples such as Jev and contrasts them with generative language models; the same interface may be insufficient for open-ended writing or extensive research.
Google