The National Science and Technology Council has since early 2023 pulled together the strength of industry, academia, and research institutions to develop the Trustworthy AI Dialogue Engine, or TAIDE. In just one year, it has developed the TAIDE LX-7B (commercial version) and TAIDE LX-13B (academic and research version) models based on Llama2. The models excel in five major tasks, including writing articles, writing letters, writing summaries, English to Chinese translation, and Chinese to English translation, and they have the ability to conduct multiple rounds of dialogue and prevent inappropriate responses. The National Science and Technology Council gathered the project team and partners to give a presentation on May 3, 2024, to showcase the various application capabilities.
National Science and Technology Council Chairperson Wu Zhengzhong said in his speech that he was grateful to the team for their efforts this year in creating a generative AI engine that best takes into account the local needs and characteristics of Taiwan. He also thanked the partners from industry, universities, and research institutions for their efforts in verifying the usability and diversity of the TAIDE model. He was quite pleased with such results, and he pointed out that the National Science and Technology Council would continue to work hard in the future to ensure that the TAIDE model would keep up with the times and become a landmark model in Taiwan.
Academician Kong Xiangzhong of Academia Sinica graced the day's event, saying that Taiwan cannot be absent from AI competition internationally. Therefore, it must master the large language model that is Taiwan's own. A lot of talents worked on the TAIDE project, and they have achieved very respectable results in a short time, which is worth celebrating. In addition, Dr. Ji Huaixin, a Google Distinguished Scientist, was also invited to give a keynote speech at today's event.
In order to allow external users to freely use TAIDE to create systems that meet their own needs, the team released the TAIDE LX-7B model on April 15, 2024, to the public. In less than half a month, more than 6,000 people have downloaded it, showing a strong demand for a traditional Chinese model that is trustworthy and designed specifically with Taiwan characteristics. On April 19th, Meta released Llama 3, which prompted users in Taiwan to demand a Llama 3-based traditional Chinese model. The TAIDE team responded by completing the Llama 3-TAIDE-LX-8B-Chat-Alpha1 model in just 4 days, which it released to the public on April 29th.
Continuous improvements and promotions of applications
The pooled prowess of domestic academic and research talents, including top professors in related fields in Taiwan, the Policy Research and Information Center, and the National Center for High-performance Computing, both of the NAR Labs has contributed to the success of the TAIDE program. The Policy Research and Information Center is primarily responsible for the collection and screening of training data while the NCHC is responsible for the establishment of the computing environment, optimization of computing efficiency, and the provision of a service platform for apps on the cloud. In addition, industry partners have also played an important role, with whose active participation, it can be foreseen that industry-developed applications will soon emerge in force. Looking ahead, the program will maintain, refine, and keep up to date the existing TAIDE models as well as promote and assist the use of TAIDE models and applications in the public and private sectors. In doing so, we will build an ecosystem with local characteristics and keep Taiwan competitive in generative AI.
Date: 2024-05-16 Source: National Center for High-performance Computing