China Telecom's self-developed large-scale star model stood out in the annual selection of the "Top Ten Most Important Weapons of the Country" by virtue of its advantages of full size, full mode, and national production. The model has made breakthrough progress in semantic understanding, speech recognition, image generation and multi-modal fusion, and has ranked among the best in multiple international authoritative evaluation lists, demonstrating China's core technical strength and independence in the field of artificial intelligence. Innovation ability. Its application in urban governance, traffic management and other fields also reflects its strong practical application value. Let us learn about the excellent performance and wide range of applications of the Xingchen large model.
In the annual selection of "Top Ten National Weapons" initiated by the News Center of the State-owned Assets Supervision and Administration Commission of the State Council, China Telecom's independently developed Star Model was successfully shortlisted for its breakthrough technological achievements. As the first full-size, full-modality, and nationally produced basic large model system in China, the Xingchen large model has demonstrated outstanding capabilities in the fields of semantics, speech, vision, and multi-modality.
In the field of semantics, the Star Model has achieved important breakthroughs. Relying on the nationally produced Wanka cluster and training framework, the model's computing efficiency reaches more than 93% of NVIDIA's equivalent computing power, and its training time ratio is as high as 98%. It is worth mentioning that this model ranked first in the international authoritative evaluation list C-Eval, and surpassed many well-known domestic large models on the OpenCompass list.
In terms of voice technology, the Xingchen large model shows unique advantages. As the first large-scale speech model in China to obtain dual registration from the Cyberspace Administration of China, it has successfully overcome the technical difficulties of multi-dialect mixed recognition, covering 40 dialect areas. Its supernatural speech generation technology only requires 5-10 seconds of samples to achieve more than 90% timbre similarity, reaching the industry-leading level.
In the field of vision, the large star model has been evaluated by academicians and experts as "reaching the internationally advanced level". At present, this model has been deployed on 765,000 cameras across the country, with an average daily call volume of more than 600 million times, playing an important role in urban governance, traffic management and other fields.
In terms of multi-modal capabilities, the large model of stars also performs well. Its image generation capabilities have surpassed the internationally renowned platform Midjourney, and its video generation technology ranks first in the VBench list. Especially in terms of long video generation, it was the first to realize the coherent generation of 5-minute videos.
At present, the Xingchen large model has been transformed into more than 10 standardized AI products and has been applied in 50 industry scenarios such as government affairs and industry. As a fully autonomous and controllable domestic AI system, the success of the Xingchen large model not only demonstrates China Telecom’s technical strength, but also marks important progress in core technology breakthroughs in my country’s AI field.
The successful inclusion of the Star Model in the "Top Ten Most Important Countries" marks China's major breakthrough in the field of artificial intelligence, injecting new vitality into the development of China's AI industry, and also indicates that AI technology will empower more industries in the future. , create greater value.