Yanolja Launches Open Source Translation AI Model

SEOUL, South Korea — Yanolja has released a translation-focused artificial intelligence model on the open-source platform Hugging Face.

The model, named EEVE ROSETTA, supports translation between more than 10 languages, including English, Chinese, and Korean, according to Yanolja.

EEVE ROSETTA operates with 4 billion parameters in a decoder-only language model architecture, a reduction of more than 50 percent compared to the previous EEVE 1.0 model, which used over 10.8 billion parameters.

The reduced parameter count lowers operating costs compared to typical large language models that require more than 7 billion parameters for translation.

The company reports improved inference accuracy and operation on standard personal computers.

EEVE is a Korean-optimized large language model with capabilities including text generation, content summarization, language translation, and sentiment analysis.

The technology was developed through a project focused on "Efficient and Effective Vocabulary Expansion."

After its initial release, EEVE was designated "Korean LLM of the Year" by the Korea Intelligence Information Society Agency.

The model also ranked first on Hugging Face's Korean leaderboard and the Tiger leaderboard.

EEVE ROSETTA analyzes context, nuance, sentence meaning, and emotional content before generating translations, particularly with complex data formats.

The technology is available under a license that permits commercial use by enterprises, public institutions, and individual developers.

Potential applications include the travel, commerce, and healthcare industries.

Yanolja Group's Chief Technology Officer, Junyoung Lee, said the company plans to expand large language model development to include Japanese, Chinese, and additional languages to support global customers.

The release makes the AI technology accessible for commercial adoption across sectors.