Introducing openoranje: Dutch AI for Everyone
We're excited to announce the launch of openoranje, an initiative dedicated to developing and open-sourcing Dutch language models optimized for edge inference and privacy-preserving applications.
Why Dutch AI Matters
The Netherlands has a rich linguistic heritage, and we believe Dutch speakers deserve AI models that truly understand their language—not just translations from English models. Our focus is on:
- Native Dutch understanding: Models trained on Dutch text, for Dutch speakers
- Edge inference: Run locally on your device, no cloud required
- Privacy-first: Your data stays on your device
- Open source: Fully transparent and community-driven
Our Approach
We're taking a different path from the large model race. Instead of training massive models that require expensive cloud infrastructure, we're focusing on:
Small but Capable Models
Our models are designed to run on consumer hardware—laptops, smartphones, and edge devices. We believe that a well-trained small model can outperform a poorly-trained large one for specific use cases.
Post-training Excellence
We specialize in post-training optimization:
# Example: Loading an openoranje model
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("openoranje/oranje-1b")
tokenizer = AutoTokenizer.from_pretrained("openoranje/oranje-1b")
# Generate Dutch text
output = model.generate(tokenizer.encode("Nederland is", return_tensors="pt"))
print(tokenizer.decode(output[0]))
What's Next
We're working on our first release: Oranje-1B, a 1 billion parameter model optimized for Dutch language understanding. Stay tuned for:
- Model weights on Hugging Face
- Training documentation
- Benchmark results on Dutch NLP tasks
- Integration guides
Join Us
We're building in the open and welcome contributions. Whether you're a researcher, developer, or just passionate about Dutch AI, there's a place for you in the openoranje community.
Follow our progress on GitHub and Hugging Face.
openoranje—Nederlandse AI, open en vrij.