Hugging Face explains decision models for structured, probabilistic AI choices

Hugging Face published a blog post explaining decision models, which take a state and predefined questions and return structured choices with probabilities instead of generating text. The post uses a transformers example and a toy training setup for a customer support classifier, showing how a model can score options directly. It says first-class training support is coming to TRL, though the pull request is still open at the time of writing. It also notes that Unsloth provides a recipe for adapting generative models into decision models with LoRA and a decision-scoring head.