World's first LLM for Tibetan language, DeepZang, launched in Lhasa

The world's first large language model for the Tibetan language, DeepZang, was officially launched in Lhasa on Sunday, marking a major technological breakthrough in the field of AI and ethnic languages in China.

The launch also highlights China's continued efforts in advancing AI for cultural preservation and development.

Developed by CHOKNOR, a local company, the platform supports intelligent interaction in Tibetan, as well as standard Chinese and English. It integrates multiple functions, including AI dialogues, real-time translation, and speech-to-text transcription.

CHOKNOR founder Tenzin Norbu said his company embarked on its journey to develop AI for Tibetan and Chinese bilingual applications as early as 2018. In the last four years, the team built a massive linguistic dataset, collecting nearly 70 million Tibetan-language parallel text entries and more than 30,500 hours of spoken Tibetan language data across the three major Tibetan dialects: Utsang, Kham, and Amdo.

"This ensures the model's use in various applications, enhancing language inclusivity and bridging cultural interactions," he said. "This achievement represents China's most extensive and accurate Tibetan-language speech database to date."

During the launch ceremony, the World Record Certification Agency (WRCA) awarded DeepZang the title of the "World's First Tibetan Large Language Model", confirming its pioneering status in AI innovation.