TLDRocket
Sign in

How to Build a Decision Model Like Jev

Antonello Fratepietro ● Covered by 2 sources

The article explains how decision models using the /v1/systemone contract score probabilities for predefined outcomes without running token-by-token text generation. Hugging Face merged native /v1/systemone support into transformers serve, covering 250+ open LLM and VLM architectures. As a result, serving can use a single prefill pass (output_tokens: 0) to return calibrated choice/noul/score distributions instead of paying the decode tax and relying on JSON text parsing.

Why it matters

Decision models return probabilities over known outcomes in a single forward pass, avoiding chat-generation and JSON parsing overhead. It walks through building such a system starting from Hugging Face’s /v1/systemone endpoint, including calibration and tree batching.

Read more related to this at Antonello Fratepietro
Hugging Face /v1/systemone Transformers Jev

Related stories

Jev introduces a new shape of LLM - System One, aka Decision Models Simon Willison’s Weblog · 1 month ago · 35 TypeSafe AI Releases Jev: A System One Model That Returns Typed, Calibrated Decisions Instead of Text MarkTechPost · 1 month ago · 8
Jev's Architecture Unmasked Archer Hume · 1 month ago · 10
Developer Tools AI APIs & Integration Large Language Models Model Deployment

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.