Image-Text-to-Text
GGUF
German
French
Italian
ocr
invoice
qr-bill
belege
swiss-qr-bill
vision
conversational
Instructions to use Keyven/belegant-4b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- llama-cpp-python
How to use Keyven/belegant-4b with llama-cpp-python:
# !pip install llama-cpp-python from llama_cpp import Llama llm = Llama.from_pretrained( repo_id="Keyven/belegant-4b", filename="belegant-4b-Q3_K_M.gguf", )
llm.create_chat_completion( 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" } } ] } ] ) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use Keyven/belegant-4b 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 Keyven/belegant-4b:Q4_K_M # Run inference directly in the terminal: llama cli -hf Keyven/belegant-4b:Q4_K_M
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf Keyven/belegant-4b:Q4_K_M # Run inference directly in the terminal: llama cli -hf Keyven/belegant-4b: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 Keyven/belegant-4b:Q4_K_M # Run inference directly in the terminal: ./llama-cli -hf Keyven/belegant-4b: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 Keyven/belegant-4b:Q4_K_M # Run inference directly in the terminal: ./build/bin/llama-cli -hf Keyven/belegant-4b:Q4_K_M
Use Docker
docker model run hf.co/Keyven/belegant-4b:Q4_K_M
- LM Studio
- Jan
- vLLM
How to use Keyven/belegant-4b with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "Keyven/belegant-4b" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Keyven/belegant-4b", "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/Keyven/belegant-4b:Q4_K_M
- Ollama
How to use Keyven/belegant-4b with Ollama:
ollama run hf.co/Keyven/belegant-4b:Q4_K_M
- Unsloth Studio
How to use Keyven/belegant-4b 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 Keyven/belegant-4b 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 Keyven/belegant-4b to start chatting
Using HuggingFace Spaces for Unsloth
# No setup required # Open https://huggingface.co/spaces/unsloth/studio in your browser # Search for Keyven/belegant-4b to start chatting
- Pi
How to use Keyven/belegant-4b with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf Keyven/belegant-4b: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": "Keyven/belegant-4b:Q4_K_M" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Hermes Agent new
How to use Keyven/belegant-4b with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf Keyven/belegant-4b: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 Keyven/belegant-4b:Q4_K_M
Run Hermes
hermes
- Atomic Chat new
- OpenClaw new
How to use Keyven/belegant-4b with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf Keyven/belegant-4b: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 "Keyven/belegant-4b: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 Keyven/belegant-4b with Docker Model Runner:
docker model run hf.co/Keyven/belegant-4b:Q4_K_M
- Lemonade
How to use Keyven/belegant-4b with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull Keyven/belegant-4b:Q4_K_M
Run and chat with the model
lemonade run user.belegant-4b-Q4_K_M
List all available models
lemonade list
| base_model: Qwen/Qwen3.5-4B | |
| language: | |
| - de | |
| - fr | |
| - it | |
| license: apache-2.0 | |
| pipeline_tag: image-text-to-text | |
| library_name: gguf | |
| tags: | |
| - ocr | |
| - invoice | |
| - qr-bill | |
| - gguf | |
| - belege | |
| - swiss-qr-bill | |
| - vision | |
| <p align="center"> | |
| <img src="logo.png" alt="Belegant" width="420"/> | |
| </p> | |
| <h1 align="center">Belegant-4B</h1> | |
| <p align="center">🇨🇭 🇩🇪 🇫🇷 🇮🇹 🇦🇹 · Schweizer Belege & QR-Rechnungen → sauberes JSON</p> | |
| Belegant-4B ist eine spezialisierte, professionell paketierte GGUF-Distribution für die | |
| strukturierte Extraktion von Schweizer Belegen, Rechnungen und QR-Rechnungen. Aus dem Bild | |
| eines Belegs entsteht striktes, schema-treues JSON – in Deutsch, Französisch und Italienisch. | |
| Belegant-4B ist multimodal (`image-text-to-text`) und liest Text und Bild. | |
| Der Fokus liegt auf Belastbarkeit statt Raten: Belegant setzt harte Regeln für den Schweizer | |
| Zahlungsverkehr durch – QR-IBAN- und Referenz-Logik (QRR / SCOR / NON), Schweizer Zahlenformat | |
| (`1'234.50`), gültige MWST-Sätze (8.1 / 2.6 / 3.8 / 0) – und erzwingt NULL-Disziplin: Was nicht | |
| eindeutig im Bild steht, wird `null`, nie erfunden. | |
| --- | |
| ## 📦 Quantisierungen | |
| | Quant | Größe | Empfehlung | | |
| |-------|------:|------------| | |
| | `belegant-4b-Q3_K_M.gguf` | ≈ 2.2 GB | Kleinster Footprint, für knappe RAM-/Edge-Setups | | |
| | `belegant-4b-Q4_K_M.gguf` | ≈ 2.6 GB | Empfohlen – bestes Verhältnis Größe ↔ Qualität | | |
| | `belegant-4b-Q5_K_M.gguf` | ≈ 3.0 GB | Etwas höhere Genauigkeit, moderat größer | | |
| | `belegant-4b-Q6_K.gguf` | ≈ 3.4 GB | Nahe F16-Qualität, für anspruchsvolle Belege | | |
| | `belegant-4b-Q8_0.gguf` | ≈ 4.3 GB | Höchste Treue, Referenz-/Abnahme-Läufe | | |
| | `mmproj-belegant-4b-f16.gguf` | ≈ 642 MB | Vision-Projektor (mmproj) – für Bild-Eingabe zwingend | | |
| > Für Bild-Eingabe wird der `mmproj`-Projektor zusätzlich zur gewählten Quant-Datei geladen. | |
| --- | |
| ## 📊 BelegBench v1 | |
| Field-Level Exact-Match (Field-EM) über Schweizer Beleg-/QR-Rechnungs-Bilder, DE / FR / IT. | |
| | Modell | Field-EM | Halluzination | | |
| |--------|---------:|--------------:| | |
| | Belegant-4B | **84.7 %** | **0 %** | | |
| | Qwen3.5-4B (roh) | 75.8 % | – | | |
| | gemma3:4b | 16.5 % | – | | |
| Kernaussage: in der 4B-Klasse +8.9 Punkte Field-EM und 0 % Halluzination – entscheidend im | |
| Zahlungsverkehr, wo eine erfundene IBAN oder ein falscher Betrag teurer ist als ein `null`. | |
| Trilingual: DE 84.7 % · FR 81.4 % · IT 83.0 %. | |
| > 📊 Gemessen mit BelegBench — dem offenen Benchmark für Schweizer QR-Rechnungen: https://huggingface.co/datasets/Keyven/belegbench | |
| --- | |
| ## 🚀 Verwendung | |
| Gewünschte Quant-Datei plus den Vision-Projektor laden: | |
| ```bash | |
| huggingface-cli download Keyven/belegant-4b \ | |
| belegant-4b-Q4_K_M.gguf mmproj-belegant-4b-f16.gguf Modelfile \ | |
| --local-dir belegant-4b | |
| cd belegant-4b | |
| ``` | |
| ### llama.cpp — Bild → JSON (Vision, empfohlen) | |
| ```bash | |
| llama-mtmd-cli \ | |
| -m belegant-4b-Q4_K_M.gguf \ | |
| --mmproj mmproj-belegant-4b-f16.gguf \ | |
| --image beleg.jpg \ | |
| --temp 0 --top-k 1 --top-p 1 \ | |
| -p "Extrahiere die Belegfelder exakt als JSON nach Schema. Was nicht lesbar ist: null." | |
| ``` | |
| ### Ollama | |
| ```bash | |
| ollama create belegant-4b -f Modelfile | |
| ollama run belegant-4b "./beleg.jpg" | |
| ``` | |
| ### Ziel-Schema (Auszug) | |
| ```json | |
| { | |
| "creditor": {"name": null, "address": null, "zip": null, "city": null, "country": null}, | |
| "creditor_iban": null, "amount": null, "currency": null, | |
| "reference_type": null, "reference": null, | |
| "invoice_number": null, "invoice_date": null, "due_date": null, "supplier_vat_uid": null, | |
| "line_items": [{"description": null, "quantity": null, "unit_price": null, "vat_rate": null, "total": null}], | |
| "vat_breakdown": [{"rate": null, "net": null, "tax": null}], | |
| "subtotal": null, "vat_total": null, "total": null, "language": null | |
| } | |
| ``` | |
| --- | |
| ## 🙏 Credits | |
| Basiert auf Qwen3.5-4B von Alibaba Cloud · Qwen Team — Dank ans Qwen-Team. Apache-2.0. | |
| ## 📄 Lizenz | |
| Apache License 2.0. | |