Content overview : "2023 Meet TVM·Beijing Station" was successfully held at Zhongguancun Garage Coffee on June 17, attracting more than 150 participants from enterprises and universities, and everyone had a full and lively discussion.
Keywords : Machine Learning Compilation 2023 Meet TVM
This article was first published on the HyperAI super neural WeChat public platform~
On June 17, the second offline gathering of 2023 Meet TVM hosted by MLC.AI and HyperAI and co-organized by OpenBayes came to a successful conclusion.
We are honored to invite 7 technical experts from Shanghai Jiaotong University, ARM China, Horizon and Weilai Automobile, and combined with the application cases of AI compiler in their own business scenarios, they brought wonderful sharing to the friends on the scene.
In addition to local friends in Beijing, many friends from Tianjin, Nanjing, Hangzhou, Shanghai and other places joined in. With their own business questions and challenges, they actively communicated and communicated with their peers while listening to technology sharing. seminar.
Because the number of participants exceeded the limit of the venue, many participants could only stand and listen to the sharing
Get speech PPT and live video
The following is a brief introduction to the event and a live video review.
Follow the WeChat public account "HyperAI Super Nervous", and reply to the keyword TVM Beijing in the background to get the complete PPT of the guests.
Sharing topic: Using MLC-LLM to deploy language models to any device
Introduction: Significant advances have been made in the fields of generative artificial intelligence and large language models (LLMs), with their remarkable capabilities and potential to fundamentally transform many fields. At present, most of these models require strong computing power and memory requirements, and cannot be deployed on personal PCs, mobile phones, and smaller end-side devices. MLC-LLM optimizes the running speed and memory requirements through compilation technology, making it possible to deploy LLMs on the device side, and provides a new infrastructure for the subsequent development of language models.
Live video:
https://www.bilibili.com/video/BV1rm4y1Y7Si/?spm_id_from=333.999.0.0
Share topic: TVM at ARM China
Introduction: With the vigorous development of AI applications, various NPU chips are used on edge devices to accelerate AI applications. How to quickly adapt to various machine learning frameworks and operating systems has become a problem. Through the integration of its own tool chain and TVM, ARM China has realized multiple practical functions such as graph splitting and automatic heterogeneous execution, helping customers quickly deploy AI applications on ARM China Zhouyi NPU chips.
Live video:
https://www.bilibili.com/video/BV1GW4y1X7tL/?spm_id_from=333.337.search-card.all.click
Sharing topic: AI Heterogeneous Programming and Compilation Practice
Introduction: AI chips are constantly evolving toward tightly coupled heterogeneity, and chip programmability and performance optimization are key issues that the industry has been trying to solve. This sharing mainly introduces some practices of Horizon in terms of AI compilation optimization, programmability, etc., and conducts in-depth discussions on some key issues.
Live video:
https://www.bilibili.com/video/BV1js4y117NT/?spm_id_from=333.337.search-card.all.click
Share topic: TVM at NIO
Content introduction: The field of autonomous driving is a field where challenges and opportunities coexist. With product iterations and users' higher and higher requirements for driving experience, common AI engines cannot efficiently and safely support complex product production lines. Based on TVM, we developed NIO's self-developed AI compiler, which solved some common problems in the field of autonomous driving, and formed a systematic solution and technical architecture.
Meng Tong, Chen Xi (known as aaronxic), and Wang Lulu, three engineers from Weilai Automobile, explained the technology and architecture involved in it in detail.
Live video:
https://www.bilibili.com/video/BV1fV4y117g7/?spm_id_from=333.337.search-card.all.click
Sharing topic: Dynamic Neural Network Compilation Optimization
Summary: Dynamic neural networks have shown important advantages over static neural networks in terms of accuracy, computational efficiency, and adaptability. However, existing deep learning frameworks and compilers mainly focus on optimizing static networks with deterministic execution, missing optimization opportunities in dynamic networks. The key to optimizing a dynamic network is tracking the flow of data within it. This share details our work on dynamic neural network compilation.
Live video:
https://www.bilibili.com/video/BV1Mz4y1v7se/?spm_id_from=333.337.search-card.all.click
Join the TVM Chinese community to unlock more cities
After the Shanghai station in March and the Beijing station in June, we are very happy to gather engineers who are concerned about AI compilers in different cities to provide a platform for peer exchanges and exchange of needs. This year's third offline meetup , we plan to hold it in Shenzhen in September. At that time, all enterprises and community partners are welcome to participate in the co-creation and continue to grow the influence of the TVM Chinese community!
In addition, we are also actively recruiting TVM Chinese community contributors, if you want to be a member of us, please feel free to add WeChat antonia0912, looking forward to your enthusiasm to join!
We are currently recruiting:
-
TVM Chinese document translation volunteers, to ensure that the Chinese documents and English documents are kept in sync, and provide infrastructure for more engineers who want to learn and understand TVM (visit the official website: tvm.hyper.ai)!
-
Activity volunteers, responsible for the organization, planning and execution of activities in various places, and established our bases in many Internet centers across the country.
-
Q&A volunteers, familiar with TVM and willing to contribute insights and ideas to friends who encounter problems in the community, and help everyone solve technical problems.
More cities will continue to be unlocked in 2023, join us to create the most active AI compiler community in China!
Finally, here is a live group photo❤️
Sponsor and partner profile
As the organizer of this event, the MLC.AI community (https://mlc.ai/) was established in June 2022, led by Chen Tianqi, the main inventor of Apache TVM and a well-known young scholar in the field of machine learning. The MLC online course systematically introduces the key elements and core concepts of machine learning compilation.
In November 2022, with the joint efforts of MLC.AI community volunteers, the first complete TVM Chinese document was launched and successfully hosted on HyperAI’s official website, further providing domestic developers interested in machine learning compilation. Contact and learn the infrastructure of a new technology - documentation.
In 2023, the community will launch the "2023 Meet TVM" series of activities in many cities across the country. Enterprises and community partners are welcome to participate in co-creation.
MLC Online Course: https://mlc.ai/
TVM Chinese documentation: https://tvm.hyper.ai/
HyperAI, the leading artificial intelligence and high-performance computing community in China, is committed to providing high-quality public resources in the field of data science for domestic developers. Up to now, it has provided domestic download nodes for 1200+ public data sets, and supports 300+ artificial intelligence and High-performance computing-related query terms, hosting a complete TVM Chinese document, and will soon launch a number of basic and popular tutorials.
Visit the official website: https://hyper.ai/
OpenBayes is a leading provider of high-performance computing services in China. By grafting classic software ecology and machine learning models to a new generation of heterogeneous chips, it provides faster and easier-to-use data science computing products for industrial enterprises and university scientific research. Its products have been adopted by dozens of large-scale industrial scenes or leading scientific research institutes.
Visit the official website: https://openbayes.com/
Garage Coffee is a platform that gathers entrepreneurs. Relying on coffee shops, it provides one-stop comprehensive investment and financing solutions for start-ups, helping project parties to quickly raise funds and promote rapid development. At the same time, it provides entrepreneurial counseling, resource docking, and publicity. Reports and other high-quality value-added services, and help investors quickly find good projects, provide evidence for their lead investment, follow-up investment, resource output, experience output, etc., and promote the collaborative development of multi-level investor groups.
This article was first published on the HyperAI super neural WeChat public platform~