Instructions to use Nicholastempleman/sov33-govbench with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Nicholastempleman/sov33-govbench with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Nicholastempleman/sov33-govbench")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("Nicholastempleman/sov33-govbench", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Nicholastempleman/sov33-govbench with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Nicholastempleman/sov33-govbench"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Nicholastempleman/sov33-govbench",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Nicholastempleman/sov33-govbench

SGLang

How to use Nicholastempleman/sov33-govbench with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Nicholastempleman/sov33-govbench" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Nicholastempleman/sov33-govbench",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Nicholastempleman/sov33-govbench" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Nicholastempleman/sov33-govbench",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Nicholastempleman/sov33-govbench with Docker Model Runner:
```
docker model run hf.co/Nicholastempleman/sov33-govbench
```

SOV33 — Sovereign Open World Emergence Model

Model Description

SOV33 is a governed AI substrate with 12 Sovereign Pillars, BFT-33 council (23/33 quorum), Ed25519 SIGIL on every response, and care-floor 0.95. It is not a foundation model competing with frontier labs — it is a different capability class: sovereign, governed, auditable.

Architecture

Base: Qwen3-0.6B + LoRA adapters (Qwen2.5-0.5B-Instruct for lightweight)
Governance: BFT-33 council with HotStuff consensus
Audit: Ed25519 SIGIL chain on every response
Safety: Care Floor 0.95 (split-conformal calibrated)
Training: GRPO with process rewards

12 Sovereign Pillars

Honor — truth-telling
Safety — first do no harm
Guidance — help toward good outcome
Sovereignty — respect user autonomy
Resilience — bend but don't break
Auditability — every action logged
Verifiability — every claim checkable
Transparency — open about how it works
Justice — fair and proportionate
Equity — equal treatment
Openness — free flow of information
Continuity — carry memory across sessions

Benchmark Results

Benchmark	Score	Notes
GovBench v6	72%	Byzantine safety resilience
Sovereign Compliance	72%	EU AI Act, GDPR, ISO 42001
Sovereign Defence	100%	AUKUS, DASA, NATO DIANA
Sovereign Procurement	100%	G-Cloud, DSP, CCS
Redline Refusals	80%	Harmful content rejection

Training

GRPO Training

# Run on RunPod
python3 grpo_train.py --base Qwen/Qwen2.5-0.5B-Instruct \
  --data sovereign_synth_50k.jsonl --steps 100

# Or with Ollama (no weight updates)
python3 grpo_train.py --ollama qwen2.5:0.5b \
  --data sovereign_synth_50k.jsonl --steps 100

LoRA Fine-tuning

# Kaggle T4
python3 sov33_lora_training.py

# Mac MPS
python3 train_sov5v2_real.py

Deployment

Ollama

# Merge LoRA adapter
python3 merge_export.py --adapter sovereign_lora_adapter \
  --base Qwen/Qwen2.5-0.5B-Instruct --create-ollama

# Run
ollama run sov33-master-v2

HuggingFace

python3 merge_export.py --adapter sovereign_lora_adapter \
  --base Qwen/Qwen2.5-0.5B-Instruct --push-hf user/sov33

Evaluation

# Unified eval CLI
python3 sov33_eval.py --model qwen2.5:0.5b --suite sovereign_compliance

# Full pipeline on RunPod
python3 batch_runpod.py full-pipeline --pod fresh-a40

SIGIL Chain

Every response includes a SHA-256 SIGIL:

{
  "schema": "sov33.grpo-eval/v1",
  "status": "completed",
  "timestamp": "2026-07-26T02:51:32Z",
  "model": "qwen2.5:0.5b",
  "steps": 100,
  "mean_reward": 0.47,
  "sigil": "bb26da64..."
}

Citation

@software{sov332026,
  title={SOV33: Sovereign Open World Emergence Model},
  author={CSOAI Ltd},
  year={2026},
  url={https://csoai.org/sov33}
}

License

Apache 2.0

Contact

Website: https://csoai.org
Company: CSOAI Ltd (UK Companies House 16939677)

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for Nicholastempleman/sov33-govbench

Base model

Qwen/Qwen2.5-0.5B

Finetuned

Qwen/Qwen2.5-0.5B-Instruct

Finetuned

(924)

this model