Yandex Builds Alice AI Foundation LLM Without Foreign Components

Yandex has developed its own large language model, Alice AI Foundation LLM, trained from scratch without using any third-party foreign components, according to a company representative cited by Vedomosti. The company describes the model as fully sovereign.

The model contains 80 billion parameters, with three billion active, and was trained on 18 trillion tokens. It is positioned as a larger and more versatile development than previous efforts and is intended to serve as the foundation for the company's future reasoning model and agent capabilities for Alice AI.

Yandex has published tuned parameters that let developers run and further train the model themselves, under a license permitting free use in commercial and research projects. The company has not disclosed the cost of development, and estimates vary due to differences in counting methods, Vedomosti reported.

Source: RIA Novosti