Sber: PyTorch-LifeStream Software Library

Main article: Data management

2022: Providing developers with free access to PyTorch-LifeStream

On February 1, 2022, Sber announced that it had provided developers with free access to the dataset preparation tool.

"Sber" made open a tool - a software library of PyTorch-LifeStream, containing several algorithms for building embeddings of event data. The tool can be used to prepare various arrays of impersonal data (datasets).

Event data can be different sequences - stories of site visits, shopping stories, events in online games, and so on. At the same time, the embedding of such a sequence generated based on the algorithms of the library will not contain any personal data.

The library implements an algorithm for applying neural network contrast training to event data, created and patented in the Sber Artificial Intelligence Laboratory. In addition, the library implements methods based on third-party scientific publications, but adapted by Sber to event data.

"The lack of purified, impersonal and marked data is one of the main barriers to the further development of artificial intelligence, to create new products and services that can solve important scientific and socially significant tasks. But there are tools that allow you to prepare such data. Free access to these tools is an important condition for overcoming the barrier. I am sure that the withdrawal of our library to the public will accelerate the development and implementation of artificial intelligence systems in Russia and, as a result, will promote competition and economic development of our country, " says Alexander Vedyakhin, First Deputy Chairman of the Board of Sberbank.

Sber has long shared the results of its research in the field of artificial intelligence with everyone. In particular, earlier the company put into open access such libraries as ruGPT-3, LAMA, RePlay, ruDALL-E.