EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago

TL;DR
Kolibri, a new English-German Mixture-of-Experts Transformer model with 78 billion parameters, has been released, supporting context lengths of up to 1 million tokens. It is available for download on Hugging Face under the Apache 2.0 license.
✦ Why It Matters
Download and experiment with Kolibri to enhance your multilingual AI applications today.
Key Takeaways
How It Works
Kolibri employs a Mixture-of-Experts architecture, utilizing 384 smaller experts to optimize performance while maintaining efficiency. This design allows for a balance between model capability and deployment costs, enabling it to handle multiple concurrent requests effectively.
Related