Key facts
- CHOKNOR Information Technology Co. independently developed China's first Tibetan large language model, DeepZang.
- DeepZang is the first Tibetan large language model to complete national filing for generative AI in China.
- The model supports over 80 languages, including Tibetan, Putonghua, English, Mongolian, and Uygur.
- DeepZang offers integrated capabilities for listening, speaking, translating, recognizing, and thinking.
- The DeepZang application supports real-time mutual translation, Tibetan-language Q&A, and cultural knowledge inquiries.
- The company has compiled a corpus of nearly 70 million Tibetan-Putonghua language pairs and a large Tibetan speech database.
CHOKNOR Information Technology Co., Ltd. has unveiled DeepZang, what it claims is the world's first Tibetan large language model, in Lhasa, the capital of China's Xizang Autonomous Region. The model and its application are designed to fill a gap in indigenous large language models and promote the innovation and inheritance of Tibetan ethnic culture in the AI era.
The development marks a strategic move for China to lead in AI for ethnic languages, according to local media Tibet.cn. The World Record Certification Agency (WRCA) has certified DeepZang as "the World's first Tibetan large language model." The model has completed national filing for generative AI in China, positioning it as a first in this field globally.
Tenzin Norbu, chairman of CHOKNOR, stated that DeepZang is an open-source platform with multilingual and multimodal capabilities, supporting over 80 languages including Tibetan, Putonghua, English, Mongolian, and Uygur. It is designed for integrated functions such as listening, speaking, translating, recognizing, and thinking.
The DeepZang application, launched on Sunday, allows users to engage in intelligent interactions in Tibetan, Putonghua, and English, offering real-time mutual translation, Tibetan-language Q&A, and cultural knowledge inquiries. Shortly after its release, the app reportedly achieved an average of 4,000 downloads per hour.
CHOKNOR has built a corpus of nearly 70 million Tibetan-Putonghua language pairs and a large, accurately annotated Tibetan speech database, covering the three major Tibetan dialect regions. Users have demonstrated the application's ability to accurately recognize and respond to instructions in various Tibetan dialects.