Instructions to use CompiwerAI/Mtrini-SVL-1.1-Merged with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use CompiwerAI/Mtrini-SVL-1.1-Merged with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="CompiwerAI/Mtrini-SVL-1.1-Merged")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("CompiwerAI/Mtrini-SVL-1.1-Merged")
model = AutoModelForMultimodalLM.from_pretrained("CompiwerAI/Mtrini-SVL-1.1-Merged", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use CompiwerAI/Mtrini-SVL-1.1-Merged with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "CompiwerAI/Mtrini-SVL-1.1-Merged"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "CompiwerAI/Mtrini-SVL-1.1-Merged",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/CompiwerAI/Mtrini-SVL-1.1-Merged

SGLang

How to use CompiwerAI/Mtrini-SVL-1.1-Merged with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "CompiwerAI/Mtrini-SVL-1.1-Merged" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "CompiwerAI/Mtrini-SVL-1.1-Merged",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "CompiwerAI/Mtrini-SVL-1.1-Merged" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "CompiwerAI/Mtrini-SVL-1.1-Merged",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use CompiwerAI/Mtrini-SVL-1.1-Merged with Docker Model Runner:
```
docker model run hf.co/CompiwerAI/Mtrini-SVL-1.1-Merged
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

🐪 Mtrini-SVL-1.1

The model is out of the lab.

Mtrini-SVL-1.1 is an open-weight ~8B model developed by Compiwer AI, based on Qwen3-VL-8B-Instruct.

It was trained with a focus on:

🧠 Reasoning
💻 Coding
🧮 Mathematics
🇲🇦 Moroccan Darija
🌍 Multilingual tasks

And yes...

It's a Mtrini. 🐪

🔄 Self-Verifying Loop

SVL = Self-Verifying Loop

Our development concept:

THINK
  ↓
CHECK
  ↓
IMPROVE
  ↓
REPEAT 🔄

The goal is to encourage the model to verify and refine its responses.

SVL is an ongoing research direction at Compiwer AI.

📚 Training

The final dataset contained 58,574 unique examples.

Dataset	Examples
🇲🇦 Moroccan Darija QA + Reasoning	2,170
💻 OpenCodeReasoning-2	30,000
🧮 OpenMathInstruct-2	30,000
⚡ Model Specs
Property	Value
Model	Mtrini-SVL-1.1
Base	Qwen3-VL-8B-Instruct
Parameters	~8B
Architecture	Qwen3-VL
Training	LoRA / PEFT
Precision	BF16
Training Context	4096 tokens
Dataset	58,574 unique examples
Developer	Compiwer AI
Origin	🇲🇦 Salé, Morocco
📊 Training Run
Steps: 3,661
Epochs: 1
Training Loss: 0.5855
Mean Token Accuracy: 83.55%

These metrics describe the training run and should not be treated as standardized benchmark scores.

🚀 Usage
from transformers import Qwen3VLForConditionalGeneration, AutoTokenizer

model_id = "CompiwerAI/Mtrini-SVL-1.1-Merged"

tokenizer = AutoTokenizer.from_pretrained(
    model_id,
    trust_remote_code=True
)

model = Qwen3VLForConditionalGeneration.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
    trust_remote_code=True
)
⚠️ Limitations

Mtrini is an experimental open model.

It can hallucinate, generate incorrect code, misunderstand prompts, or produce incorrect mathematical answers.

Verify important outputs.

🌍 Why Mtrini?

AI shouldn't only be built by the biggest companies on Earth.

Mtrini is Compiwer AI's experiment in building open AI with a focus on:

🇲🇦 Moroccan language
🧠 Reasoning
💻 Coding
🧮 Mathematics
🤝 Open AI

Building AI For Everyone.

Built in Salé, Morocco 🇲🇦

Downloads last month: -

Safetensors

Model size

9B params

Tensor type

BF16

Model tree for CompiwerAI/Mtrini-SVL-1.1-Merged

Base model

Qwen/Qwen3-VL-8B-Instruct

Finetuned

(600)

this model

Quantizations

1 model