A major step forward for Romanian natural language processing and artificial intelligence was taken this week with the launch of the OpenLLM Romania initiative. This collaborative effort, driven largely by volunteer work from the Romanian NLP/AI community, has already produced promising initial results.
Two key achievements were announced – Romanian versions of the Llama2 and Mistral large language models that have demonstrated impressive performance. These models have been released publicly on HuggingFace and GitHub, with details outlined in a research paper on arXiv.
While much of the work so far has been volunteer-driven, the project received a private research grant from BRD – Groupe Societe Generale. The core participants currently include researchers and students from the University Politehnica of Bucharest, the University of Bucharest, and the Institute of Logic and Data Science.
However, the OpenLLM Romania teams are eager to expand collaboration with other researchers, machine learning engineers, data scientists and institutions across Romania. Their goal is to develop large language models and datasets specific to the Romanian language that can be openly used for both research purposes and commercial applications.
The teams are calling on the Romanian government and private companies to contribute resources to further these efforts, highlighting the importance of having open Romanian language AI capabilities available for wide usage across sectors.
