The world's highest-altitude large language model has officially been born. The "Sunshine Qingyan" V1.0, jointly released by Tibet University and a local startup team, was unveiled today. The model has surpassed 100 billion parameters, with 28.8 billion tokens of training data, covering all fields including news, law, medicine, education, and technology, filling the gap in Tibetan AI infrastructure.
The State Council issued the "Artificial Intelligence +" Action Plan in August, which is seen as the starting gun for accelerating AI in Tibet. Currently, community service centers in Lhasa city center and the Gongga Airport terminal have been the first to access AI customer service and real-time translation systems; during the ongoing Qinghai-Tibet scientific expedition, the "e-Expedition" App has completed holographic image collection and cloud-based instant transmission via satellite links, cutting the time required for field data sorting in half.
Image source note: The image is AI-generated, and the image licensing service provider is Midjourney.
Nima Zaxi, an academician and pioneer in Tibetan digitalization, said that Tibet's AI is moving from "scenario applications" to "system development," with the core breakthrough lying in the corpus. Dangzeng LuoBu's team of young entrepreneurs spent six years building a parallel corpus of 70 million Tibetan-Chinese sentences and establishing a multilingual voice database covering 140 countries, providing a data foundation for "Sunshine Qingyan" and the DeepZang model set to be released by year-end.
