Sber opened access to the ruGPT-3.5 model

Sberbank opened access to the ruGPT-3.5 model, which formed the basis of the GigaChat product. The model is licensed under the MIT license, which allows developers to use the model in their own commercial products.

The number of ruGPT-3.5 parameters is 13 billion. Also, when responding, the model uses a context with a length of 2048 tokens.

https://habr.com/

The model was trained in two stages: the first time – on 300 GB of books, scientific articles and data from social networks in the public domain, the second time – on 110 GB of data, which includes code, legal documents and texts from Wikipedia.