Hugging Face Standardizes Open Decision Models in Transformers With DecisionPipeline
Hugging Face added native DecisionPipeline, AutoModelForDecision, and DecisionConfig APIs to Transformers so open decision models share one inference contract.
Hugging Face formally brought decision models into the core open-source ML stack on November 10, introducing DecisionPipeline, AutoModelForDecision, and DecisionConfig in transformers so open-weight models can be served, discovered, and trained through one shared interface.
One contract across diverse architectures
Written by Lysandre, Pedro Cuenca, and Tom Aarsen, the announcement noted that open decision models have proliferated rapidly since TypeSafe AI debuted Jev, using very different internals—from joint schema-scoring heads to lightweight readout layers bolted onto LLM backbones and edge-sized multimodal variants.
from transformers import pipeline
decide = pipeline("decision", model="autotrust/GEV-26B-Decide")
result = decide(
state="Customer asks to cancel and mentions filing a chargeback.",
questions={"escalate": {"type": "noul", "instructions": "Requires immediate fraud/legal handoff?"}}
)
| Property | Classical Classifier | Zero-Shot Classifier | Decision Model |
|---|---|---|---|
| Probability over choices | Yes | Yes | Yes |
| Labels defined at inference | No | Yes | Yes |
| Multiple typed questions per pass | No | No | Yes |
| Native score & yes/no primitives | Task-specific | Adapted | Native |
“The community has converged to the term decision model, because it describes the model’s computational role rather than invoking a particular theory of cognition.”
Lysandre, Pedro Cuenca & Tom Aarsen — Hugging Face Blog
Adopting that shared contract lets engines such as llama.cpp serve any conforming decision model behind one API while allowing the Hugging Face Hub to package, tag, and benchmark open models consistently.