·
We publish letscodeit.dev blog posts from a separate GitHub repository. The Next.js app fetches markdown at runtime with a one-hour cache,…
GEO (generative engine optimization) is how you structure and publish content so AI chatbots and answer engines quote your pages when…
When you send a message to an LLM, you are not paying for characters or words. You pay for tokens . A token is the unit the model splits…
Aleph Alpha has released Kolibri, an English-German Mixture-of-Experts model with open weights licensed under Apache 2.0.
Kolibri features 78 billion total parameters, with 3 billion active, and supports context lengths up to 1 million tokens. Developed through a refined training pipeline, it allows for stable pre-training without manual intervention during hardware or connection issues.
Source: aleph-alpha.com
Great to see an open weights model with such a large context length. 1 million tokens is impressive!
78 billion parameters but only 3 billion active? Seems like a lot of unused capacity.
Comments