Instructions to use CyberPaul/ratio-qwen-7b-gguf-v2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use CyberPaul/ratio-qwen-7b-gguf-v2 with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M

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 CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M

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 CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M

Use Docker

docker model run hf.co/CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M

LM Studio
Jan

vLLM

How to use CyberPaul/ratio-qwen-7b-gguf-v2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "CyberPaul/ratio-qwen-7b-gguf-v2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "CyberPaul/ratio-qwen-7b-gguf-v2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M

Ollama
How to use CyberPaul/ratio-qwen-7b-gguf-v2 with Ollama:
```
ollama run hf.co/CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M
```
Unsloth Desktop

How to use CyberPaul/ratio-qwen-7b-gguf-v2 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use CyberPaul/ratio-qwen-7b-gguf-v2 with Docker Model Runner:
```
docker model run hf.co/CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M
```

Lemonade

How to use CyberPaul/ratio-qwen-7b-gguf-v2 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M

Run and chat with the model

lemonade run user.ratio-qwen-7b-gguf-v2-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use CyberPaul/ratio-qwen-7b-gguf-v2 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M

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 CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use CyberPaul/ratio-qwen-7b-gguf-v2 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "CyberPaul/ratio-qwen-7b-gguf-v2:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

RATIO Qwen 7B v2 - GGUF (LEGACY / EXPERIMENTAL)

⚠️ STATUS: LEGACY — não recomendado para uso em produção. Este modelo é a versão experimental anterior do LLM generativo do RATIO. O modelo recomendado é o CyberPaul/ratio-gemma3-4b-gguf (Gemma3-4B fine-tuned).

Por que este modelo é LEGACY?

Durante a validação empírica em hardware CPU (Windows, P52 com 32 GB RAM), até o modelo base Qwen2.5-7B-Instruct não quantizado pelo nosso pipeline apresentou saída degradada (repetição de dígitos/garble) no runtime local (Ollama/llama.cpp). O problema foi reproduzido:

com o fine-tune ratio-qwen-7b-gguf-v2 (com e sem system message)
com o modelo base qwen2.5:7b-instruct oficial do Ollama (mesmo sintoma)

Ou seja: a degradação não é do fine-tune nem do template — é uma incompatibilidade do runtime local com a arquitetura Qwen no hardware testado. Por isso este artefato foi descontinuado em favor do Gemma3-4B, que roda corretamente e responde com qualidade no mesmo hardware.

Alternativa recomendada

CyberPaul/ratio-gemma3-4b-gguf — LLM generativo do RATIO:

Fine-tune Gemma3-4B-IT (8.501 exemplos de ratio decidendi)
GGUF Q4_K_M (~2.5 GB), roda em CPU
Validação empírica completa (RAG single-turn e multi-turn) ✅

O que era este modelo

Modelo Qwen2.5-7B-Instruct ajustado para o domínio jurídico brasileiro (extração, síntese e explicação da ratio decidendi), exportado em GGUF Q4_K_M e Q8_0 para uso local.

Arquivos disponíveis

Qwen2.5-7B-Instruct.Q4_K_M.gguf (~4.7 GB)
Qwen2.5-7B-Instruct.Q8_0.gguf (~8.1 GB)

Uso (caso queira testar mesmo assim)

ollama create ratio-qwen -f Modelfile
ollama run ratio-qwen

⚠️ Em hardware CPU, o Qwen 7B pode gerar saída degradada (repetição de dígitos). Teste antes de adotar. Recomendamos o Gemma3.

Dados de treinamento

Fine-tune (LoRA) com Unsloth sobre o dataset RATIO v3, consolidado a partir de:

acórdãos do STJ e STF
extrações estruturadas do pipeline RATIO
sínteses supervisionadas da ratio decidendi

Limitações

⚠️ Legacy: descontinuado em favor do Gemma3-4B
Pode resumir demais em casos muito complexos
Pode reproduzir vieses do corpus jurisprudencial
Exige validação humana em uso profissional

Créditos

Fine-tuning e pipeline jurídico: CyberPaul / Projeto RATIO
Modelo base: Qwen2.5-7B-Instruct
Conversão/quantização GGUF: Unsloth
Runtime local: Ollama / llama.cpp

Citação

@misc{ratio_qwen_2026,
  author = {CyberPaul},
  title  = {RATIO Qwen 7B v2 (legacy): GGUF experimental para extracao de ratio decidendi no direito brasileiro},
  year   = {2026},
  url    = {https://huggingface.co/CyberPaul/ratio-qwen-7b-gguf-v2}
}

Downloads last month: 62

GGUF

Model size

8B params

Architecture

qwen2

Hardware compatibility

4-bit

Model tree for CyberPaul/ratio-qwen-7b-gguf-v2

Base model

Qwen/Qwen2.5-7B

Finetuned

Qwen/Qwen2.5-7B-Instruct

Quantized

(428)

this model