StartLux-Decision
Open decision models from 0.8B to 27B: a calibrated probability for every option in one pass, and a short thought only when unsure. Engine for vLLM, llama.cpp a
What is StartLux-Decision?
Open decision models from 0.8B to 27B: a calibrated probability for every option in one pass, and a short thought only when unsure. Engine for vLLM, llama.cpp and MLX.
Overview of StartLux-Decision
What it is
StartLux-Decision is a family of open decision models. Given a state (text, JSON or images) and a set of typed questions, a model returns a probability for every option of every question: a choice among listed options (choice), a yes or no judgement (noul), or a level on a scale (score). Instant models answer in a single forward pass with bidirectional attention over the prompt. AutoThink models answer in one pass when the top probability reaches a gate, 0.6 by default, and otherwise generate a short reasoning trace and read the answer again. The repository contains the inference engine and HTTP server for both variants. The API is compatible with System One /v1/systemone.
Key features
- Instant models (from 0.8B to 27B parameters) answer in a single forward pass with bidirectional attention.
- AutoThink models (from 4B to 27B parameters) reason only when the top probability is below a configurable gate (default 0.6).
- Supports three question types: choice among options, yes/no judgement (
noul), and score on a scale. - The engine serves models in vLLM, llama.cpp and MLX formats.
- Provides an HTTP server with a
/v1/systemoneendpoint. - Supports controls like
think_mode,gate,think_budget,none_check,allow_none,return_details,return_thought, andimages.
Use cases
The README demonstrates decision-making in game AI, such as chess (checkmate without search), StarCraft II (strategic calls for a Terran bot), and Super Mario Bros. (clearing World 1-1). It also shows real-time decision tasks like helicopter rescue (calling each stage of a Coast Guard hoist rescue), rocket launch monitoring, and sports game calls (JevBall). The README also provides an example of customer support ticket routing, urgency assessment, and severity scoring.
Getting started
To run the models yourself, install the package into the same Python environment as vLLM. The package carries the vLLM plugin that Instant models need.bashgit clone https://github.com/StartLuxLabs/StartLux-Decision.gitcd StartLux-Decisionpip install -e '.[vllm]'Download a model and start one vLLM server per GPU with the decision server in front of them:bashhf download startlux-models/StartLux-Decision-v1.1-4B-AutoThink --local-dir models/StartLux-Decision-v1.1-4B-AutoThinkstartlux-decide serve --model models/StartLux-Decision-v1.1-4B-AutoThink --launch-vllm --gpus 0,1,2,3 --port 8090To try a model without a GPU, use bamboonode.dev which hosts the 27B AutoThink model free and without usage limits, requiring only a Bamboo API key.
License and pricing
The license is Apache-2.0. The models are open source. A free version is available via bamboonode.dev with no usage limits, as stated in the README.
Based on the project's public README and repository metadata.