The Ministry of Digital Affairs and Kyivstar have chosen an AI model to train the national LLM
The Ministry of Digital Affairs, together with Kyivstar, has chosen a large language model on which the national Ukrainian LLM will be taught in advance.
The Ministry of Digital Affairs, together with Kyivstar, has chosen a large language model on which the national Ukrainian LLM will be taught in advance.
This was reported by the Ministry of Digital Affairs.
What influenced the choice?
This large language model will be Google's Gemma 3. According to the ministry, the selection was based on how well the model already processes texts in the Ukrainian language and its controllability during training.
“This will help minimize linguistic and ethical risks in our LLM,” said Danylo Tsvok, Chief AI Officer of the Ministry of Digital Affairs and CEO of WINWIN AI Center of Excellence.
The key advantages in choosing the model were:
- Optimal balance of performance and resources - Gemma delivers high quality with optimal infrastructure requirements. It is one of the best open models in terms of size and quality ratio.
- Multilingual support — the model already has the Ukrainian language in its range and is easily adaptable through additional training.
- Multimodality — the model can perceive and analyze not only text, but also images.
- Extended tokenizer - the volume of tokens ensures accurate and efficient text processing and retraining. The model has a long context window - 128,000 tokens.
- The presence of several dimensions allows you to flexibly choose the size of the model for a specific area of application.
- Successful examples and experience using Gemma to create Ukrainian-language LLMs, including Lapa LLM and MamayLM.
What's next?
Gemma 3 is being adapted to the Ukrainian language, in particular, it is planned to:
- improve the Ukrainian tokenizer — this will improve the model's work with the Ukrainian language, reduce errors when creating Ukrainian-language texts, and optimize computational costs when using the model;
- to further train the model on unique Ukrainian-language texts that are currently being collected by experts;
- create benchmarks (tests) to fine-tune the model for further use.
Recall that last Friday , the Ministry of Digital Affairs and Kyivstar announced that they were looking for a name for the national Ukrainian LLM . Until December 5, Ukrainians can offer their options, and the best ideas will receive prizes.
The Ministry of Digital Affairs, together with NVIDIA, is developing its own language model Diia AI LLM and AI infrastructure in Ukraine
The Ministry of Digital Affairs, together with NVIDIA, is developing its own language model Diia AI LLM and AI infrastructure in Ukraine
Open Ukrainian language model Lapa LLM has received a public release
Open Ukrainian language model Lapa LLM has received a public release
Kyivstar.Tech is recruiting people to work on the national LLM. What specialists are needed
Kyivstar.Tech is recruiting people to work on the national LLM. What specialists are needed
Read the country's main IT news in our Telegram
Read the country's main IT news in our Telegram
