Wayne Wang
Co-Founder & CTO of PPIO
Wayne Wang Co-Founder & CTO of PPIO -Founding Partner & CTO -Graduated from Huazhong University of Science and Technology -Chief Architect of PPTV -Co-founder & CTO of Jindou Telematics -LiveVideoStack core instructor -Holder of 49 patents -Successive entrepreneur In 2004, he participated in the founding of PPTV, and in 2008, he took a break from his studies at Huazhong University of Science and Technology to continue his entrepreneurship. As a member of the founding team and chief architect, he built PPTV from the first line of code, and went through many technical architecture refactoring and building of PP video cloud service until it was acquired by Suning Yuncheng. During 2014-2018, as the co-founder of Jiedou Telematics, he started a second business, during which he led the team to develop China's first in-vehicle intelligent system: Jiedou OS, as well as a series of vehicle-grade software and hardware technology research, and won the excellent rankings in the global automotive industry's unmanned driving competition, and his products served Audi, BMW, Mercedes-Benz and other internationally renowned car makers. In 2018, he co-founded PPIO with Yao Xin, the former founder of PPTV, for his third venture.
Topic
PPIO: How to achieve both "hard and soft" and innovative breakthroughs in arithmetic infrastructure and reasoning optimisation
LLM is rapidly developing, and the high inference cost has become one of the main obstacles hindering its wide application. This presentation will deeply discuss the cost optimisation of LLM inference, share the performance improvement methods from hardware and software, and show the practice cases in PPIO GPU Cloud to provide developers and scholars with new ideas to break through the bottleneck of LLM inference. 1.The high cost of LLM reasoning hinders the application in the industry. 2.The computational overhead and performance index of LLM reasoning. 3.Discussion on the basic scheme of LLM reasoning optimisation. 4.Discussion on the advanced scheme of LLM inference optimisation. 5. Comprehensive Application of LLM Inference Optimisation Techniques in PPIO GPU Cloud 6.PPIO GPU Cloud Cost-effective Arithmetic Leasing and Model Hosting Service 7.Summary
