AI News

Trending story

Hugging Face adds decision model support to Transformers

1 reports1 discussions2 story updatesUpdated Nov 10 23:29

Understand this story

AI overview

On November 10, 2026, Hugging Face published a post announcing the addition of decision model support to Transformers, involving three interfaces: DecisionPipeline, AutoModelForDecision, and DecisionConfig. The post explains that decision models have the model assign probabilities to preset answers rather than generating token by token, and presents three types of decision primitives—Choice, Score, and Noul—along with how confidence is calculated. The article also states that llama.cpp can serve through the same /v1/systemone interface, that the Hub has added a decision model tag, and gives an example of fine-tuning JEV-9B with the Trainer API for customer service request routing; first-class support in TRL is still at the PR stage.

AI-generated from public reports

Latest update · Discussion

Tom Aarsen joined this discussion

Developments

1 confirmed developments

  1. Hugging Face adds decision model support to Transformers

    Source:Hugging Face

Reports and discussion timeline

Browse reports and discussion by time. Their evidence levels remain distinct.

Nov 10

  1. Tom AarsenRelated discussion

    Tom Aarsen joined this discussion

    This post contributes to heat; it is not a formal report or confirmed fact.

  2. Hugging FaceFirst-partySelected

    Hugging Face Details Decision Models: Interfaces, Three Types of Primitives, and Ecosystem Support

    Hugging Face published an article introducing decision models, that is, having models assign probabilities to preset answers rather than generating token by token, and presents three types of decision primitives—Choice, Score, and Noul—as well as the confidence calculation method. The article explains that Transformers already supports DecisionPipeline, AutoModelForDecision, and DecisionConfig, that llama.cpp can provide services through the same /v1/systemone interface, and that the Hub has also added a decision model tag. The article also provides an example of using the Trainer API to fine-tune JEV-9B for customer service request routing, and mentions that first-class support in TRL is still in the PR stage.

Heat trend for this story

Current
6
Period peak
16 Nov 11 00:00
24-hour change
-50%

Move the pointer or select the chart to inspect hourly heat. Use the left and right arrow keys with a keyboard.

Related stories

  1. Related storiesDeepSeek releases DeepSeek-V4.1-Flash-1017
  2. Related storiesEnglish translation is being prepared.
  3. Related storiesNVIDIA blog showcases dual DGX Station deployment of GLM-5.3
  4. Related storiesThinking Machines Lab introduces the Inkling 2 Small end-to-end 4-bit training recipe