Quyet
Family of open decision models from 153M to 31B that pick one option per question with calibrated probabilities in the /v1/systemone shape.
Overview
Quyet is a family of open decision models. You give it a state (text, JSON or a conversation) and typed questions, and it returns a probability distribution over the options you supplied, with no generated text. This package is the runtime for all Quyet 1.0 models, whose weights are on Hugging Face: Large (31.3B) and Medium (4.66B) LLMs for GPU, and Small, Small-EN and Tiny encoders (153M-328M) that run on CPU or GPU. The models target English and are also tuned for Vietnamese.
Answers follow TypeSafe's /v1/systemone shape: choice returns a label, confidence and probabilities; score returns the expected level, probabilities and legend; noul returns P(true). Probabilities are temperature-calibrated per question type and option count, and raw_probabilities exposes uncalibrated distributions. A CLI processes JSONL requests, and a live demo of Quyet-1.0-Large runs at quyet.ai. Code and weights are Apache-2.0.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What does the customer want? | cancel · limit · other |
| Yes / no | The request is urgent. | — |
| Score | How upset is the customer? | calm · annoyed · angry |
How it works
- pip install quyet and load a model by Hugging Face id
- Call predict with a state and typed questions (or predict_many / the JSONL CLI)
- Read calibrated choice, score or noul answers
- Branch in code on the probabilities
Strengths
- Small encoders run on CPU
- Compatible answer shape with TypeSafe's API
- Temperature-calibrated probabilities with raw access
- Apache-2.0 code and weights
Caveats
- Lower accuracy outside English and Vietnamese
- Large model needs multiple GPUs
Getting started
pip install quyet, then quyet.load("chinhnc/Quyet-1.0-Small").predict(state, questions).At a glance
- Type
- Model
- Status
- beta
- Language
- Python
- Runtime
- PyTorch, CLI
- Has code
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
A family of open decision models that pick one option per question with calibrated probabilities and answer in TypeSafe's `/v1/systemone` shape. Apache-2.0.
MrJev/awesome-jev
Frequently asked questions
What does Quyet decide?
Any choice, score or noul question about a state, returning calibrated probabilities over your options.
Does it need a TypeSafe key?
No; the models are open weights you run locally.
Can it run without a GPU?
Yes, the Small, Small-EN and Tiny encoders run on CPU.
Related projects
Mangaba AI's local, calibrated typed-decision model for Portuguese text, trained to optimize the Brier score; proprietary since v5.
ProjectproductionPython- blink★ 22
Open one-pass typed-decision model with a dependency-free C99 runtime that also builds to a 66 KB WebAssembly module.
ServicebetaC - Foq★ 11
Local open alternative to Jev using a 1.58-bit ternary 8B model to answer booleans, choices, scores and Pydantic extraction in ~25 ms.
ProjectbetaPython Runs the open Laya decision model entirely in a browser tab with WebGPU and ONNX Runtime Web: no server, no API key.
ServicebetaTypeScriptllama.cpp-based C++ server that turns any LLM into a System One decision model, reading answers from next-token logits.
ProjectbetaC++Three open 4B decision models for English, Turkish and German that serve the Jev /v1/systemone API on a CPU in about 3 GB of RAM.
ServicebetaPython