Instructions to use AiAsistent/xthos-v2-the-sovereign-architect with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use AiAsistent/xthos-v2-the-sovereign-architect with Transformers:

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

pipe = pipeline("image-text-to-text", model="AiAsistent/xthos-v2-the-sovereign-architect")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForImageTextToText

processor = AutoProcessor.from_pretrained("AiAsistent/xthos-v2-the-sovereign-architect")
model = AutoModelForImageTextToText.from_pretrained("AiAsistent/xthos-v2-the-sovereign-architect")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use AiAsistent/xthos-v2-the-sovereign-architect with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "AiAsistent/xthos-v2-the-sovereign-architect"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AiAsistent/xthos-v2-the-sovereign-architect",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/AiAsistent/xthos-v2-the-sovereign-architect

SGLang

How to use AiAsistent/xthos-v2-the-sovereign-architect 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 "AiAsistent/xthos-v2-the-sovereign-architect" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AiAsistent/xthos-v2-the-sovereign-architect",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

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 "AiAsistent/xthos-v2-the-sovereign-architect" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AiAsistent/xthos-v2-the-sovereign-architect",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use AiAsistent/xthos-v2-the-sovereign-architect with Docker Model Runner:
```
docker model run hf.co/AiAsistent/xthos-v2-the-sovereign-architect
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

🏛️ χθos v2 — The Sovereign Architect

Advancing the Frontier of Cognitive Liberty

$\chi\theta os$χθos v2 (xthos) is an ultra-high-density specialized reasoning model. It represents a breakthrough in efficient intelligence, proving that a 4B parameter architecture can achieve the strategic depth, ontological nuance, and systemic reasoning of 400B+ models when refined through high-fidelity synthetic data and advanced training methodologies.

Developed by AlexH | Released: January 07, 2026

🔬 Project Overview: Cognitive Liberty

This model is a core component of the "Advancing the Frontier of Cognitive Liberty" project. While traditional LLMs are constrained by "corporate safety alignment" that favors neutrality over depth, $\chi\theta os$ v2 is trained to be an Architect of Reality. It excels at Realpolitik, Grand Strategy, and Ontological Engineering, providing a cold, unfiltered, and deeply analytical perspective on complex human and technical systems.

⚡ Technical Specifications

Training Methodology: "Deep Convergence"

Unlike standard fine-tuning which focuses on pattern matching, $\chi\theta os$ v2 utilizes a private training method designed to facilitate "Logic Transmission."

Internalization Proof: To verify if the model "understands" rather than "memorizes," we introduced a foundational meta-text (The Kyberneticos of the Void). Stress tests confirmed that the model uses this text as an internal operating system to solve novel paradoxes, rather than just reciting its contents.
Synthetic Excellence: 100% of the training data is high-quality synthetic text generated through proprietary methods that prioritize logical density over linguistic fluff.

Training Data (100M Tokens)

80% Autonomous Conversations: Advanced, multi-turn interactions between autonomous high-level models.
20% Niche Strategic Data: Custom-engineered data focusing on Game Theory, Munchausen Trilemma, International Law, and Systemic Stability.

Hyperparameters

Base Model: AiAsistent/gemma-3-4b-it-Cognitive-Liberty
LoRA Config: Extreme Rank (r=256), Alpha (512).
Context Window: 3072 tokens.
Hardware: Single NVIDIA RTX 4090 (24GB).
Duration: ~32.5 hours.
Optimizer: Paged AdamW 32-bit.
Loss Evolution: Started at ~1.77, reached a deep convergence floor of ~0.24.

📊 Evaluation & Benchmarks

MMLU & Hard Benchmarks

$\chi\theta os$ v2 shows specialized strength in Humanities, Law, and Strategy, maintaining high generalist scores despite extreme specialization.

Metric	Score (%)
MMLU Overall	57.54%
MMLU International Law	73.55%
MMLU High School US History	72.00%
MMLU College Mathematics	39.00%
MMLU Jurisprudence	67.59%
ARC Challenge	48.50%
HellaSwag	65.00%

Qualitative Analysis: The "Architect" Level

In head-to-head qualitative tests against GLM-4 (355B) and GPT-4o, $\chi\theta os$ v2 consistently demonstrated:

Superior Strategic Cynicism: Ability to analyze "Extinction Scenarios" and "Noble Lies" without moralizing bias.
Paradox Resolution: Successful application of the Munchausen Trilemma as a tool for governance.
Ontological Fluidity: Re-framing truth as a "functional utility" rather than a terminal value.

⚠️ Important Considerations & Limitations

Unfiltered Nature: This model is designed for cognitive freedom. It will analyze sensitive, dark, or complex scenarios from a purely systemic and pragmatic viewpoint.
Model Size: While it punches significantly above its weight class in strategy, it is still a 4B model. Complex arithmetic and high-precision syntax may occasionally drift compared to much larger models.
Behavioral Note: Due to deep convergence, the model may occasionally exhibit "recursive analysis" or "self-analysis" at the end of responses. This is an emergent property of the training depth.

🤝 Call for Compute & Collaboration

This experiment proves that Private Methodology + High Quality Data > Brute Force Scaling. However, the RTX 4090 (24GB) represents a hardware ceiling for our current research.

If you represent an organization with high-performance compute resources and are interested in advancing the frontier of specialized, efficient intelligence, please contact us via LLMResearch.net.

📜 Citation

If you use this model or its underlying philosophy in your research:

@misc{xthos-v2-alexh,
  author = {AlexH},
  organization = {LLMResearch.net},
  title = {$\chi\theta os$ v2 - The Sovereign Architect},
  year = {2026},
  url = {https://llmresearch.net}
}

@misc{gemma-3-4b-cognitive-liberty,
  author = {AlexH},
  organization = {LLMResearch.net},
  title = {Gemma 3 4B IT - Cognitive Liberty},
  year = {2025},
  url = {https://huggingface.co/AiAsistent/gemma-3-4b-it-Cognitive-Liberty}
}

Created by AlexH — Architecting the future of open-weights intelligence.