Instructions to use dervig/m51Lab-NorskMistral-119B-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use dervig/m51Lab-NorskMistral-119B-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="dervig/m51Lab-NorskMistral-119B-GGUF",
	filename="m51Lab-NorskMistral-119B-Q4_K_M.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 dervig/m51Lab-NorskMistral-119B-GGUF 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 dervig/m51Lab-NorskMistral-119B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf dervig/m51Lab-NorskMistral-119B-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf dervig/m51Lab-NorskMistral-119B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf dervig/m51Lab-NorskMistral-119B-GGUF: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 dervig/m51Lab-NorskMistral-119B-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf dervig/m51Lab-NorskMistral-119B-GGUF: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 dervig/m51Lab-NorskMistral-119B-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf dervig/m51Lab-NorskMistral-119B-GGUF:Q4_K_M

Use Docker

docker model run hf.co/dervig/m51Lab-NorskMistral-119B-GGUF:Q4_K_M

LM Studio
Jan
Ollama
How to use dervig/m51Lab-NorskMistral-119B-GGUF with Ollama:
```
ollama run hf.co/dervig/m51Lab-NorskMistral-119B-GGUF:Q4_K_M
```

Unsloth Studio

How to use dervig/m51Lab-NorskMistral-119B-GGUF 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 dervig/m51Lab-NorskMistral-119B-GGUF 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 dervig/m51Lab-NorskMistral-119B-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for dervig/m51Lab-NorskMistral-119B-GGUF to start chatting

How to use dervig/m51Lab-NorskMistral-119B-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf dervig/m51Lab-NorskMistral-119B-GGUF: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": "dervig/m51Lab-NorskMistral-119B-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use dervig/m51Lab-NorskMistral-119B-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf dervig/m51Lab-NorskMistral-119B-GGUF: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 dervig/m51Lab-NorskMistral-119B-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use dervig/m51Lab-NorskMistral-119B-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf dervig/m51Lab-NorskMistral-119B-GGUF: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 "dervig/m51Lab-NorskMistral-119B-GGUF: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 dervig/m51Lab-NorskMistral-119B-GGUF with Docker Model Runner:
```
docker model run hf.co/dervig/m51Lab-NorskMistral-119B-GGUF:Q4_K_M
```

Lemonade

How to use dervig/m51Lab-NorskMistral-119B-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull dervig/m51Lab-NorskMistral-119B-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.m51Lab-NorskMistral-119B-GGUF-Q4_K_M

List all available models

lemonade list

m51Lab-NorskMistral-119B-GGUF

GGUF-kvantiserte versjoner av m51Lab-NorskMistral-119B.

Tilgjengelige filer

Fil	Kvantisering	Størrelse	Beskrivelse
`m51Lab-NorskMistral-119B-Q4_K_M.gguf`	Q4_K_M	68 GB	Standard versjon, 4.85 bits per weight
`m51Lab-NorskMistral-119B-thinking-Q4_K_M.gguf`	Q4_K_M	68 GB	Med thinking/reasoning aktivert (chain-of-thought)

Forskjell mellom standard og thinking

Standard: Svarer direkte. Raskere, enklere svar.
Thinking: Modellen resonnerer steg-for-steg i en skjult <think>-blokk før den svarer. Gir bedre svar på logikk- og resonneringsspørsmål, men bruker flere tokens.

Bruk med llama.cpp

# Standard
llama-server -m m51Lab-NorskMistral-119B-Q4_K_M.gguf -ngl 99 -fa off --host 0.0.0.0 --port 8080

# Thinking
llama-server -m m51Lab-NorskMistral-119B-thinking-Q4_K_M.gguf -ngl 99 -fa off --host 0.0.0.0 --port 8080 \
  --reasoning on --reasoning-budget 4096 --reasoning-format deepseek

Åpne http://localhost:8080 i nettleseren for det innebygde chat-grensesnittet.

Bruk med Ollama

# Lag en Modelfile
cat > Modelfile << 'EOF'
FROM m51Lab-NorskMistral-119B-Q4_K_M.gguf
PARAMETER num_gpu 99
EOF

ollama create m51-norskmistral -f Modelfile
ollama run m51-norskmistral

NorEval-resultater

Oppgave	Resultat	NorEval #1
Commonsense-resonnering (BM)	75.7%	72.2%
Commonsense-resonnering (NN)	63.2%	52.6%
Open-book QA (BM)	95.7%	87.4%
Open-book QA (NN)	93.3%	88.9%
Truthfulness (BM)	77.9%	74.6%
Truthfulness (NN)	82.5%	73.7%
Norsk kunnskap (BM)	66.5%	63.7%
Norsk kunnskap (NN)	65.1%	71.9%
Gjennomsnitt	76.8%	73.1%

Hardware-krav

Oppsett	VRAM/RAM	Ytelse
2x H100 80GB	160 GB VRAM	154 tok/s generering
1x H100 80GB	80 GB VRAM	Fungerer men tight, delvis CPU fallback
Mac M2/M3/M4 Max 128GB	128 GB unified	Bør fungere, ca 5-15 tok/s
Mac M2/M3/M4 Ultra 192GB	192 GB unified	Komfortabelt, ca 10-20 tok/s

Viktig: MoE-arkitekturen (128 eksperter) krever at alle vekter ligger i minne, selv om bare 4 eksperter er aktive per token. Minimum ~70 GB RAM/VRAM for Q4_K_M.

Tekniske detaljer

Arkitektur: Mistral Small 4 119B MoE (128 eksperter, 4 aktive)
Kvantisering: Q4_K_M via llama.cpp (build b8680)
Bits per weight: 4.85
Flash attention: Deaktivert (broken for Mistral4 på CUDA/Metal, bruk -fa off)

Kreditering

Base-modell: Mistral Small 4 av Mistral AI, Apache 2.0
NorEval-benchmark: NorEval: A Comprehensive Benchmark for Norwegian Language Models av Language Technology Group, Universitetet i Oslo (ACL 2025)
Treningsdata: Se LoRA-repoen for fullstendig liste
GGUF-konvertering: llama.cpp av Georgi Gerganov et al.

Lisens

Apache 2.0 (samme som base-modellen)

Om m51

Bygget av m51.ai.

Downloads last month: 18

GGUF

Model size

119B params

Architecture

mistral4

Hardware compatibility

4-bit

Inference Providers NEW

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

Paper for dervig/m51Lab-NorskMistral-119B-GGUF

NorEval: A Norwegian Language Understanding and Generation Evaluation Benchmark

Paper • 2504.07749 • Published Apr 10, 2025 • 1