MTS has launched an “open” language model to solve business problems

The model is suitable for writing articles, translations between Russian and English, processing and analyzing text data, developing chatbots, as well as searching and analyzing information in corporate knowledge bases. “Cotype Nano is capable of processing context of up to 32,000 tokens (about 45 pages of text) at a time, allowing you to work with large volumes of data,” the company says.

They noted that the new model can run locally on smartphones, computers and laptops, even with “average performance.” According to the statement, Cotype Nano is an “open” model; The parameters you use to make decisions can be studied and customized for specific tasks.

Budget:

Cotype Nano includes 1.5 billion parameters. The model was trained on special data sets containing computer code, mathematical problems and synthetic information.

Text processing (inference) speed is about 190 tokens per second on an Nvidia A100 GPU and 9.5 tokens per second on a smartphone with a Qualcomm Snapdragon 8 Gen 2 processor.

Cotype Nano uses the transformative Qwen 2.5 architecture and supports popular frameworks such as VLLM, OpenVINO, and Hugging Face.

The model is available for download at Hugging Face.

Companies are increasingly integrating large language models into their business processes to improve their efficiency. Alexey Borshchov, AI product director at Just, previously spoke in a column on RB.RU about the associated risks and strategies for safely deploying generative AI technologies.