Hugging Face adds decision pipeline: models output option probabilities without generating text
kalyan_kpl · x · 2026-11-11
Hugging Face's official blog announces Decision Model support in the Transformers pipeline. Key points:
- Not every task needs text generation—ticket routing only needs classification. Instead of prompting an LLM for JSON, the decision pipeline lets a model output a probability distribution over predefined options directly.
- Usage: pipeline("decision", model=...) with a state text and questions structure; supports choice (with criteria), noul (yes/no), and score (e.g. urgency levels), returning e.g. {'refund': {'noul': 0.99}, 'tone': {'choice': 'angry'}, 'urgency': {'score': 1.8}}.
- Example models like autotrust/GEV-26B-Decide; suited to classification, routing, and intent-detection tasks.
More from coding & agent
- Skill Files Are the New Free Courses: $500 Expert Tutorials Now Exist as Markdown on GitHub — heyneighbor · 2026-11-11
- Why Harness Design May Become the System Design of the Agentic AI Era — kmeanskaran · 2026-11-11
- Alibaba ships Qwen Code 1.0, now runs three competing coding tools; Qwen4 Max Preview panned — teortaxesTex · 2026-11-11
- 'Compaction amnesia': how context compression quietly erases agent memory in long sessions — malliktwts · 2026-11-11
- Meituan's LongCat-2.5 tops 3DCodeBench at 55.4, open-sourced under MIT — dr_cintas · 2026-11-11
- Gemini 4 Carbon scores 57.3 vs top model's 58 at 1/4.5 the cost in real-repo bug test — koraykv · 2026-11-11