ASC22 Student Supercomputer Challenge Preliminary: Large AI Language Model and AI in Scientific Research

The preliminary round of the 2022 ASC Student Supercomputer Challenge (ASC22) kicked off on January 5, 2022. More than 300 teams around the world compete on challenging tasks like the large AI Language Model Yuan 1.0 and the 2020 Gordon Bell Prize-winning application DeePMD-kit.

Among the participants are championship teams from prior student cluster competitions at SC, ISC and ASC, including Tsinghua University (champion of ISC21 and SC21), Jinan University (champion of ASC20-21 onsite, and third place in ISC21), National Tsing Hua University (champion of ASC20-21 virtual), Sun Yat-sen University (champion of SC21- IndySCC), ShanghaiTech University (runner-up of SC21), Southern University of Science and Technology (LINPACK winner in SC21), and ETH Zurich ( third place in SC21- IndySCC). Other strong competitors include teams from the University of Science and Technology of China, Texas A&M University, Ural Federal University, Monash University, and Universidad EAFIT.

In the preliminary round of ASC22, all teams are required to tackle two very advanced tasks: Yuan1.0 - large AI language model, and an excellent combination of AI with advanced scientific application - DeePMD-kit.

Launched in 2021, Yuan 1.0 is a record-setting massive Chinese language model with 247.5 billion parameters. It demonstrated promising performance in various language processing applications. Training such a large AI model requires massive computing tasks of parallel computing, presenting a real-life test to the student teams for their skills in supercomputing application optimization. The competitors are required to build a mini Yuan model with 4.7 billion parameters based on a 100-GB high-quality Chinese dataset provided by the ASC22 Committee. Without any reference code offered for the model design, the teams need to construct the model, define optimum training strategies, and achieve the best performance.

DeePMD-kit, a software designed for machine learning-based molecular dynamics, presents an excellent opportunity for the students to explore possibilities of AI combination with science, which has become one of the most promising synergies for breakthrough innovations in this field. Advancement in High-performance computing (HPC) certainly has unleashed a huge potential for ML-based molecular dynamics. DeePMD-kit, winner of the 2020 Gordon Bell Prize, has achieved a double-precision peak performance of 91 PFLOPS and a time-to-solution 1000 times faster than the previous baseline performance in Summit supercomputer, thus redefining the paradigm for integration of Science, AI, and HPC. The student teams will have hands-on experience in building a model with DeePMD-kit, and optimize the HPC system for delivering the maximum training speed.

The opening ceremony of ASC22 will be held in Beijing on January 19, 2022. Important information such as the host of the competition finals will be announced at the event. Afterwards, all the teams will have the opportunity to participate in the ASC22 online training camp from January 20 to January 21, 2022. HPC and AI experts from Intel, NVIDIA, AWS Cloud, DP Technology, and the State Key Laboratory of High-end Server & Storage Technology will lecture the teams how to prepare for the competition by providing advice on competition rules, on-prem and cloud HPC cluster design and optimization, etc. The winners of previous seasons will also share their experiences during the training camp.

About ASC

The ASC Student Supercomputer Challenge is sponsored and organized by Asia Supercomputer Community and supported by Asian, European, and American experts and institutions. The main objectives of ASC are to encourage exchange and training of young supercomputing talent from around the world, improve supercomputing applications and R&D capacity, boost the development of supercomputing, and promote technical and industrial innovation. The first ASC Student Supercomputer Challenge was held in 2012 and since has attracted over 10,000 undergraduates from all over the world. Learn more about ASC at http://www.asc-events.net/.