Instructions to use MAOZEDONG5567/modelo-fiscal-pt with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use MAOZEDONG5567/modelo-fiscal-pt with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="MAOZEDONG5567/modelo-fiscal-pt",
	filename="gguf/modelo-fiscal-pt-q4_k_m.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use MAOZEDONG5567/modelo-fiscal-pt 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 MAOZEDONG5567/modelo-fiscal-pt:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf MAOZEDONG5567/modelo-fiscal-pt:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf MAOZEDONG5567/modelo-fiscal-pt:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf MAOZEDONG5567/modelo-fiscal-pt: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 MAOZEDONG5567/modelo-fiscal-pt:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf MAOZEDONG5567/modelo-fiscal-pt: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 MAOZEDONG5567/modelo-fiscal-pt:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf MAOZEDONG5567/modelo-fiscal-pt:Q4_K_M

Use Docker

docker model run hf.co/MAOZEDONG5567/modelo-fiscal-pt:Q4_K_M

LM Studio
Jan

vLLM

How to use MAOZEDONG5567/modelo-fiscal-pt with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "MAOZEDONG5567/modelo-fiscal-pt"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "MAOZEDONG5567/modelo-fiscal-pt",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/MAOZEDONG5567/modelo-fiscal-pt:Q4_K_M

Ollama
How to use MAOZEDONG5567/modelo-fiscal-pt with Ollama:
```
ollama run hf.co/MAOZEDONG5567/modelo-fiscal-pt:Q4_K_M
```

Unsloth Studio

How to use MAOZEDONG5567/modelo-fiscal-pt 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 MAOZEDONG5567/modelo-fiscal-pt 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 MAOZEDONG5567/modelo-fiscal-pt to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for MAOZEDONG5567/modelo-fiscal-pt to start chatting

How to use MAOZEDONG5567/modelo-fiscal-pt with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf MAOZEDONG5567/modelo-fiscal-pt:Q4_K_M

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": "MAOZEDONG5567/modelo-fiscal-pt:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use MAOZEDONG5567/modelo-fiscal-pt with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf MAOZEDONG5567/modelo-fiscal-pt: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 MAOZEDONG5567/modelo-fiscal-pt:Q4_K_M

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use MAOZEDONG5567/modelo-fiscal-pt with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf MAOZEDONG5567/modelo-fiscal-pt: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 "MAOZEDONG5567/modelo-fiscal-pt: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"

Docker Model Runner
How to use MAOZEDONG5567/modelo-fiscal-pt with Docker Model Runner:
```
docker model run hf.co/MAOZEDONG5567/modelo-fiscal-pt:Q4_K_M
```

Lemonade

How to use MAOZEDONG5567/modelo-fiscal-pt with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull MAOZEDONG5567/modelo-fiscal-pt:Q4_K_M

Run and chat with the model

lemonade run user.modelo-fiscal-pt-Q4_K_M

List all available models

lemonade list

Modelo Fiscal PT — Portuguese Tax Assistant SLM

Two-pass Portuguese tax assistant SLM based on cnmoro/Qwen2.5-0.5B-Portuguese-Hybrid-Reasoning.

Architecture

Pass 1 (Query Translator): User question → structured JSON (action, tags, year) with GBNF grammar constraint
Pass 2 (Response Polisher): Source text + question → polished response with disclaimer

Action classification (lookup/filter/compare) is rule-based (not learned).

Quantization

Q4_K_M with per-tensor fallback (~50% of 896-dim Qwen2.5 head tensors fall back to q5_0).

Format: GGUF
Size: 379 MB
Context window: 32768 tokens (native Qwen2.5)

Deployment

Sideload into PocketPal AI:

Download the GGUF file
Download pass1_grammar.gbnf
Configure PocketPal to use the GBNF grammar for Pass 1 inference

Usage

Pass 1 — Structured JSON output (with GBNF grammar)

System: Você é um tradutor de consultas tributárias. Converta a pergunta do usuário em JSON estruturado. Use APENAS os campos: action (lookup|filter|compare), tags (lista de palavras-chave), year (ano).
User: Qual a diferença entre IRS e IRC?
Assistant: {"action": "compare", "tags": ["cirs", "artigo_11", "diferencia", "entre", "irs"], "year": null}

Pass 2 — Polished response (no grammar)

System: Você recebe um texto técnico de legislação tributária. Apresente-o de forma clara em português. Mantenha todos os números, percentagens e prazos exatos. Máximo de 4 frases. Não adicione informação nova.
User: Texto de origem: De acordo com o artigo 51.º do CIRS, as mais-valias imobiliárias são tributadas à taxa de 28%, salvo opção pelo englobamento. O prazo de declaração é até 30 de junho do ano seguinte.

Pergunta do utilizador: Qual a taxa de mais-valias imobiliárias?
Assistant: De acordo com o artigo 51.º do CIRS, as mais-valias imobiliárias são tributadas à taxa de 28%, salvo opção pelo englobamento. O prazo de declaração é até 30 de junho do ano seguinte. Para efeitos legais, considere o texto oficial do diploma mencionado.

Training

Trained with VibeThinker Spectrum-to-Signal methodology:

CPT: Corpus pre-training with disclaimer variants (10 phrasings)
Pass 1 Broad: 70% easy queries → JSON translation
Pass 1 Hard: 30% multi-article/rare-code queries
Pass 2: Response polishing with disclaimer baking

Data: 16,522 training items from a 24,845-entry Portuguese tax triple dataset.

Files

File	Description
`modelo-fiscal-pt-q4_k_m.gguf`	Quantized model (Q4_K_M, 379 MB)
`pass1_grammar.gbnf`	GBNF grammar for Pass 1 JSON constraint

Downloads last month: 49

GGUF

Model size

0.5B params

Architecture

qwen2

Hardware compatibility

4-bit