- Who
- Hugging Face
- What
- explained decision models, their interface, and support in Transformers and llama.cpp
- Quote
In the rest of the post we will refer to the models as "decision models" instead of "system one models".
What happened
Hugging Face explains decision models, which take a state and typed questions and return decisions with probabilities instead of generating text. The post distinguishes choice, score, and yes/no primitives from conventional and zero-shot classification, then demonstrates support in Transformers and llama.cpp, model discovery on the Hub, and a fine-tuning example. It also notes that TRL support is still an open pull request.
Why it matters
The post compares decision models with conventional and zero-shot classifiers, then shows how their typed outputs fit into a shared interface for inference, serving, and training.
Read next
See the full storyHugging Face Explains Decision Models and Their Ecosystem SupportThe story so far
Hugging Face explained decision models as systems that take a state and typed questions and return decisions with probabilities instead of generating text. The post describes choice, score, and yes-or-no outputs, and shows support in Transformers and llama.cpp, model discovery on the Hub, and an example of fine-tuning. Hugging Face distinguishes decision models from conventional and zero-shot classification. Support in TRL was still an open pull request when the post was written.