sk-decision-0
a model for structured situation analysis and probabilistic decision-making
2 min readupdated
The sk-decision-0 model by developer suryavirkapur is designed to process texts in English and Arabic. It takes a textual description of a situation and a set of typed questions as input, and outputs probabilities for choice options, binary answers, or rating scales. The model comes with its own server and requires an NVIDIA GPU with CUDA support. This solution is useful for automating customer support, request routing, and risk assessment.
Key features
- Dual-language support for English and Arabic.
- Three task types: multiple choice, binary yes/no (noul), and scale evaluation (score).
- Returns raw probabilities and model confidence for each answer.
- Two versions available: full bf16 (17 GB) and quantized NF4 (7.2 GB).
What you can do
- Set up automated routing and prioritization for incoming support tickets.
- Assess the urgency and severity of incidents based on customer text descriptions.
- Integrate the model into backend services via REST API thanks to compatibility with TypeSafe System One.
Example
hf download suryavirkapur/sk-decision-0 --exclude "model/*" --local-dir sk-decision-0 # NF4 only
cd sk-decision-0
uv venv -p 3.12 venv && uv pip install -p venv/bin/python -r requirements.txt
cp .env.example .env # set S1_TOKEN
./run.sh # serves on 127.0.0.1:8000Read more
DetailsAccuracy: 91.5% for bf16 and 90.5% for NF4 (tested on 200 examples).
VRAM requirements: 19 GB for bf16, 7.1 GB for NF4.
System requirements: Linux, NVIDIA GPU with CUDA 12.8 drivers.
Limitations: requires a dedicated server from the repository and does not work in standard runtimes like vLLM or Ollama.










Comments