site.btaINSAIT Presents Latest Generation Large Language Model in Ukrainian

INSAIT has announced the launch of its first language model in a foreign language, the Education Ministry press office said on Wednesday. MamayLM was created for the Ukrainian state and, according to data presented by the Institute, is currently the best performing practical model in Ukrainian in the world.

It is a joint development of INSAIT with the Swiss polytechnic ETH - Zurich. With its 9 billion parameters, it is cost-effective and can be run on a single graphics processing unit (GPU), yet it is extremely capable in both Ukrainian and English. The model comes with strong capabilities that significantly outperform similarly sized models in both languages, while being competitive or outperforming models dozens of times larger.

The language model was developed with donations and is freely available for use by all Ukrainian-speaking users and institutions. It would be of strategic importance for the Ukrainian state, as it can be implemented and used in various domains. Its relevance for state institutions is even greater, as it allows to keep data confidential and to use it locally within individual organizations. A publicly accessible chat can be launched in the future with the model to make it easier for end-users to mine any information.

It is built on top of Google's Gemma-2-9B model. Some of INSAIT's BgGPT 2.0 models were also developed on this basis. "Based on the experience gained in the creation of the Bulgarian model, with some improvements, a new model has been developed that is lightweight, practical, but very capable of understanding and producing and analyzing text in Ukrainian, while maintaining and even improving its basic capabilities," INSAIT said.

At the end of 2024, INSAIT presented a new generation artificial intelligence designed to work in Bulgarian - BgGPT. Bulgaria thus became the first country in the European Union to have an artificial intelligence of such a high technological level in its own language, created by a state-funded scientific institute. Now the Bulgarian Institute presents a high-performance model in Ukrainian that will be freely downloadable and usable by Ukrainian institutions, organizations and businesses. INSAIT is working in parallel on various generative AI models, both language models for other countries and in other fields. All this places Bulgaria in a key position in the fastest growing and most strategic field at the moment, and INSAIT's new development is expected to attract global attention, the press release said.