OpenEuroLLM-Danish

A fine-tuned version of Gemma3 optimized for Danish language responses. This model is part of the JobAutomation Collection, aimed at providing high-quality language models for European languages.

Overview

OpenEuroLLM-Danish is specifically designed to provide fluent, natural responses in Danish. The model maintains the reasoning capabilities of its Gemma3 base while being optimized to understand and generate authentic Danish text.

Features

• Native Danish Responses: Always responds in natural, correct Danish
• Culturally Aware: Understanding of Danish culture, history, and regional context
• Specialized Vocabulary: Prioritizes genuine Danish vocabulary over loan words
• Grammatical Accuracy: Properly handles Danish-specific grammar rules and constructions
• Fast & Efficient: Based on the efficient Gemma3 architecture for good performance even on consumer hardware

Usage

# Pull the model
ollama pull openeurollm-danish

# Run the model
ollama run openeurollm-danish

Model Details

• Base Model: Gemma3
• Context Length: 8K tokens
• Training: Fine-tuned on a diverse corpus of Danish text including literature, news, academic papers, and conversational data
• License: Apache 2.0
• Recommended RAM: 8GB minimum

System Prompt

This model uses the following default system prompt:

Du vil altid svare på korrekt og naturligt dansk. Du er en hjælpsom og venlig assistent, der hjælper brugerne med deres spørgsmål på dansk. Du foretrækker at bruge ægte danske ord og undgår fremmedord.

Customization

You can create a custom version with different parameters:

# Create a Modelfile
echo 'FROM openeurollm-danish
PARAMETER temperature 0.7
PARAMETER top_p 0.9
SYSTEM "Du vil altid svare på korrekt og naturligt dansk."' > Modelfile

# Create your custom model
ollama create danish-academic -f Modelfile

About OpenEuroLLM

OpenEuroLLM is an initiative to create high-quality language models for European languages. Our mission is to ensure that AI language technologies support the linguistic diversity of Europe, providing models that understand and generate content in languages beyond the dominant few.

Acknowledgements

This model was developed with support of the open source community and Ollama.