The WAVE SUMMIT 2023 summit, sponsored by the National Engineering Research Center for Deep Learning Technology and Application and undertaken by Baidu Flying Paddle and Wenxin Big Model, is coming! This summit focuses on the main directions of AI technology, industrial ecology, and future trends. Big names from all walks of life, including production, learning, research, and use, will focus on the development and future of deep learning and large-scale model technology, bringing forward insights into the industry and a series of new blockbusters. release. Today, I will introduce the highlights of the "Smart Hard Core Ecological Co-creation" parallel forum, let's have a sneak peek!
Event Details
The "Intelligent Hard Core Ecological Co-creation" Parallel Forum is honored to invite Liang Xiaoyao, a professor and subject leader from the Department of Computer Science and Engineering of Shanghai Jiaotong University, as the chairman of the forum. A number of senior experts and enterprise representatives among the members of the hardware ecological co-creation plan will share with you the in-depth integration and cooperation and capability upgrades in terms of computing power and algorithms, and jointly discuss the software and hardware integration scheme based on the shared ecological construction of Wenxin and Flying Paddle And innovative practice plans, sharing cutting-edge technologies for large model training and reasoning deployment.
report topic
Liang Xiaoyao
Professor and Academic Leader, Department of Computer Science and Engineering, Shanghai Jiao Tong University
Software and hardware collaboration to build a large-scale industrial ecology
The demand for computing power is exploding under the large model, how to seek a breakthrough in computing power chips, and the construction of a collaborative ecosystem of software and hardware is the key. The rapid rise of big models in the past year has given the chip industry a historical opportunity to rejuvenate. At this moment, it is necessary to gather the power of the whole industry, whether it is the current overlord or a rising rookie, to join hands to create an excellent large-scale model ecology.
Zheng Bin
NVIDIA×Flying Paddle, Senior Product Leader, NVIDIA Asia Pacific
: Extreme performance, enabling AI to be used in thousands of industries.
Generative AI applications have blossomed, profoundly changing our way of life in the past six months, and greatly improving our production efficiency and productivity. Ability to work. In this keynote speech, it will be shared that by using the new NVIDIA hardware and software stack, developers of large language models (LLM) can reduce costs and increase efficiency, achieve technological innovation, and gain the trust of LLM enterprises through confidential computing functions.
Zhao Yuping
Intel Artificial Intelligence Architect
Intel×Flying Paddle: Core Acceleration Goes Far, Shares the New AI Layout
Since the cooperation began in 2017, Baidu Flying Paddle and Intel have continued to expand the depth and breadth of cooperation. The two parties have established a joint technical team, from the hardware Actively adapt to Intel's full-stack AI products. At the software level, the two parties have also carried out in-depth cooperation, which will be shared in detail in this keynote speech. Meanwhile, recently large language models (LLMs) are taking the entire machine learning world by storm. LLM requires a lot of computing power to meet the low-latency requirements of some scenarios, and the associated costs can be prohibitive for many organizations. In this keynote, we will discuss optimization techniques for reducing LLM size and inference latency so that they can run efficiently on Intel 4th generation Xeon processors calling AMX (Advanced Matrix Acceleration Engine).
Ming Libo,
General Manager of Sugon Intelligent Computing Application and Promotion Department
Sugon × Flying Paddle: Public computing service, fully supporting digital transformation
With the continuous maturity of ultra-large-scale models, the artificial intelligence industry has gradually entered the 2.0 era. Sugon and Baidu Flying Paddle have in-depth cooperation in the fields of large model pre-training, industry model fine-tuning, application adaptation and docking, application mall construction, etc., and continue to improve the soft and hard ecology of artificial intelligence, providing a strong foundation for the development and implementation of next-generation artificial intelligence technology. support.
Xu Zhen
General Manager of Kunlun Core Government Enterprise/Smart Computing Center Business
Kunlun Core × Flying Paddle: End-to-end optimization to create AI "core" power in the era of large models
The era of large-scale models poses challenges to the supply of computing power. Kunlun Core adheres to the collaborative innovation of software and hardware, and launches end-to-end solutions for large-scale models in response to market demand, leading the industry in performance and implementation results. Faced with the unprecedented opportunities brought by large models, Kunlun Core firmly joins hands with upstream and downstream partners such as Baidu Flying Paddle to accelerate the improvement of the core "core" competitiveness of my country's artificial intelligence industry.
Yan Yan
, Software Application R&D Director of
Tianshu Zhixin Tianshu Zhixin×Flying Paddle: Establishing Roots and Foundations to Create a Core Paradigm for the Large-scale Model Industry
Tianshu Zhixin, as an early domestic company that realized the mass production of cloud-based general-purpose GPUs, has achieved many cases in the field of large models Landing, including nearly 100 billion parameter pre-training based on the kilocalorie scale, full-parameter/high-efficiency parameter fine-tuning, industry large-scale model reasoning all-in-one machine, etc. The software stack is compatible with the mainstream ecology. The training product Tiangai 100 has quickly completed the three-level adaptation and mutual recognition with the flying paddle framework in a short period of time. In the later period, the two parties will further cooperate in the direction of full training, reasoning and algorithm platforms based on the open ecology to jointly build The solution integrating software and hardware empowers the industry.
Ding Yunfan
Vice President of System Architecture of Biren Technology Biren Technology × Flying Paddle: There is no end to innovation, and there are huge gaps in the distributed training of GPT-like large models in the construction of high-
performance general-purpose AI facilities in terms of model parameter scale, computing power scale, and training performance.
Challenges, the application of large models also has problems of high cost and long delay. This report mainly introduces the distributed parallel training strategy of the GPT large model, how to build a large model training system and a low-latency, high-performance large-model inference engine based on Biren Technology’s large computing power general-purpose GPU, and finally shares Biren Technology and Baidu Flying Paddle Cooperation and progress in the basic ecology and large-scale models of artificial intelligence software and hardware.
Round table forum: Wenxin large model empowers end-to-end AI
Baidu, Qualcomm, MediaTek, and Amou Technology, three end-to-end hardware benchmarking companies, jointly discussed the application prospect of Wenxin large model empowering end-to-end AI. Representatives of three hardware companies will share their latest progress in end-side hardware supporting large-scale model technology, discuss how to combine Wenxin large-scale model with end-side hardware technology products, and use the powerful functions of Wenxin large-scale model to further promote end-to-end The development of side AI applications to meet the ever-increasing user needs.
In addition, Flying Paddle + Wenxin Large Model will join hands with hardware manufacturers to jointly release the Flying Paddle + Wenxin Large Model Hardware Ecological Co-creation Plan in this forum.
