Instructions to use PXIN/Ouroboros-9B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use PXIN/Ouroboros-9B with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="PXIN/Ouroboros-9B",
	filename="mmproj-BF16.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use PXIN/Ouroboros-9B with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf PXIN/Ouroboros-9B:BF16
# Run inference directly in the terminal:
llama-cli -hf PXIN/Ouroboros-9B:BF16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf PXIN/Ouroboros-9B:BF16
# Run inference directly in the terminal:
llama-cli -hf PXIN/Ouroboros-9B:BF16

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf PXIN/Ouroboros-9B:BF16
# Run inference directly in the terminal:
./llama-cli -hf PXIN/Ouroboros-9B:BF16

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf PXIN/Ouroboros-9B:BF16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf PXIN/Ouroboros-9B:BF16

Use Docker

docker model run hf.co/PXIN/Ouroboros-9B:BF16

LM Studio
Jan
Ollama
How to use PXIN/Ouroboros-9B with Ollama:
```
ollama run hf.co/PXIN/Ouroboros-9B:BF16
```

Unsloth Studio

How to use PXIN/Ouroboros-9B with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for PXIN/Ouroboros-9B to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for PXIN/Ouroboros-9B to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for PXIN/Ouroboros-9B to start chatting

How to use PXIN/Ouroboros-9B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf PXIN/Ouroboros-9B:BF16

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "PXIN/Ouroboros-9B:BF16"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use PXIN/Ouroboros-9B with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf PXIN/Ouroboros-9B:BF16

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default PXIN/Ouroboros-9B:BF16

Run Hermes

hermes

Docker Model Runner
How to use PXIN/Ouroboros-9B with Docker Model Runner:
```
docker model run hf.co/PXIN/Ouroboros-9B:BF16
```

Lemonade

How to use PXIN/Ouroboros-9B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull PXIN/Ouroboros-9B:BF16

Run and chat with the model

lemonade run user.Ouroboros-9B-BF16

List all available models

lemonade list

🐉 Ouroboros-9B: The Recursive Reasoning Experiment

Ouroboros-9B is an independent research project focused on pushing the boundaries of recursive optimization and architectural efficiency. It represents a paradigm shift in how high-parameter models can be deployed and refined on consumer-grade hardware.

🚀 The Vision

Ouroboros is built on the principle of recursive refinement. By utilizing extreme 1.58-bit ternary compression as a foundation, the project aims to explore the intersection of large-scale reasoning and minimal-bit representations. Ouroboros doesn't just run on edge hardware; it is designed to evolve there.

🌳 Lineage & Architecture

Ouroboros-9B is built upon a high-performance logic foundation:

Ouroboros-9B (Ternary Architectural Baseline)
↳ OmniCoder-9B (Advanced Coding & Reasoning Logic)
↳ Qwen 3.5 9B (Underlying Transformer Architecture)

🛠️ Technical Specifications

This initial baseline release utilizes Ternary (1.58-bit) Quantization via the TQ1_0 format.

Quantization: TQ1_0 (BitNet 1.58-bit Ternary)
Extreme Footprint: Weights are crushed down to {-1, 0, 1}, reducing the model size from ~18GB to a compact 2.7GB.
Memory Efficiency: Over 85% reduction in VRAM/RAM requirements compared to BF16.
Multimodal Engine: Integrated vision projectors enable visual reasoning and code-from-image analysis.
Hardware Acceleration: Native optimization for the QVAC Fabric engine using specialized Vulkan and Metal kernels.

🖼️ Multimodal Capabilities

Ouroboros-9B includes high-fidelity vision projectors from the Unsloth collection, enabling it to process visual inputs such as code screenshots, diagrams, and UI layouts.

mmproj-BF16.gguf: Optimized for modern GPUs with native bfloat16 support.
mmproj-F16.gguf: Universal high-precision projector for all backends.

🔬 Experimental Roadmap

Ouroboros is designed to "consume itself" to grow stronger through successive training phases.

Phase 1 (Active): Establishing the Ternary Baseline. Deployment of the 1.58-bit architectural shift.
Phase 2: Recovery Fine-tuning. Utilizing QVAC Fabric native low-bit training to restore logic and perplexity lost during the initial quantization.
Phase 3: Recursive self-optimization and specialized forks for autonomous agentic workflows.

🛠️ Usage (QVAC Fabric)

To achieve the intended performance and use the ternary kernels, use the QVAC Fabric Engine.

Build:

git clone https://github.com/tetherto/qvac-fabric-llm.cpp.git
cd qvac-fabric-llm.cpp
cmake -B build -DGGML_VULKAN=ON
cmake --build build --config Release

Run:

# Text Inference
./build/bin/llama-cli -m ouroboros-9b-TQ1.gguf -p "Write a recursive function in Rust to..."

# Multimodal Inference
./build/bin/llama-minicpmv-cli -m ouroboros-9b-TQ1.gguf --mmproj mmproj-BF16.gguf --image screen.png -p "Explain the logic flow in this diagram."

🔗 Credits

This project is made possible by the following foundational works:

Coding Logic: Tesslate/OmniCoder-9B
Base Architecture: Qwen/Qwen3.5-9B
Vision Projectors: unsloth/Qwen3.5-9B-GGUF
Quantization Engine: QVAC Fabric

Disclaimer: This is an experimental research artifact. Logic performance may vary compared to higher-bit versions until recovery fine-tuning is complete.

Downloads last month: 422

GGUF

Model size

9B params

Architecture

qwen35

Hardware compatibility

We're not able to determine the quantization variants.

View all variants

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for PXIN/Ouroboros-9B

Base model

Qwen/Qwen3.5-9B-Base

Finetuned

Qwen/Qwen3.5-9B

Quantized

(272)

this model