Aleph Alpha has announced the release of Kolibri, a cutting-edge Mixture-of-Experts (MoE) model designed for English-German language processing. With a staggering 78.1 billion parameters, Kolibri activates only 3.46 billion parameters per token, enabling efficient and effective natural language understanding.
This model boasts a 1 million-token context and is engineered to perform per-request reasoning efforts, making it a significant advancement in AI language models. Its Apache 2.0 FP8 weights can be operated on a single B200 or H200, emphasizing accessibility for developers and researchers alike. The launch of Kolibri represents a notable step in the evolution of AI language models, particularly for bilingual applications.
Compiled automatically by the Tech AI Newsdesk from public AI-news sources and summarised in our own words.
