Yandex Unveils Alice AI Foundation: An 80-Billion-Parameter Language Model
Yandex has announced the release of the Alice AI Foundation, an 80-billion-parameter large language model developed entirely in-house. The model, which does not utilize third-party components, has been made available under an open license for commercial and academic use.
According to Yandex, the Alice AI Foundation was trained on 18 trillion tokens and features a unique Mixture-of-Experts architecture that activates only 3 billion parameters at a time. This design allows for efficient processing while maintaining high performance, particularly in Russian-language applications.
The model is positioned as a sovereign AI solution, reflecting Russia's legal framework that emphasizes the use of domestic components in AI development. While the formal process for recognizing sovereign models will not commence until March 2027, Yandex's release underscores its role as a leading developer in the Russian AI landscape.
Yandex has introduced two new benchmarks, WikiWebFacts and HardMultiQA, to evaluate the performance of Russian-language models, further enhancing the relevance of Alice AI Foundation in the market. The model's capabilities are particularly beneficial in specialized fields such as law and medicine, where accuracy is crucial.
FAQ
What is the Alice AI Foundation?
The Alice AI Foundation is an 80-billion-parameter large language model developed by Yandex entirely in-house, designed for commercial and academic use.
How was the Alice AI Foundation trained?
The model was trained on 18 trillion tokens and features a unique Mixture-of-Experts architecture that activates only 3 billion parameters at a time for efficient processing.
What are the key features of the Alice AI Foundation?
Key features include its large scale of 80 billion parameters, open licensing for use, and a focus on high performance in Russian-language applications.
What benchmarks has Yandex introduced for evaluating the Alice AI Foundation?
Yandex has introduced two new benchmarks, WikiWebFacts and HardMultiQA, to evaluate the performance of Russian-language models.
In which fields is the Alice AI Foundation particularly beneficial?
The model is particularly beneficial in specialized fields such as law and medicine, where accuracy is crucial.
Comments
Comments are moderated before publish.
No comments yet — be the first.