On the afternoon of February 25, 2023, Ant Group and NVIDIA held the "Triton Meetup 2023" in the Beijing BCos shared office. The event was a milestone in promoting the development of the field of model reasoning .
This event focuses on sharing topics related to AI Infra and inference engines , creating an open and shared technology ecosystem in the Triton community, and jointly promoting the promotion and application of NVIDIA Triton Inference Server in China.
Next, let us review the wonderful moments of the event together~
In order to let everyone better understand the Triton community, Shenyi, a solution architect from NVIDIA, brought a sharing titled "The Present and Future of NVIDIA Triton Inference Server". Let everyone know Triton's important Feature introduction and Roadmap update, as well as Triton's success stories.
After giving the audience a preliminary understanding of the Triton community, Gong Mingliang, a senior technical expert of Ant Group, brought a sharing of "Triton's optimization in the whole scene of Ant". He patiently explained Triton's solutions in search, recommendation, advertising business scenarios and cognitive business, as well as further optimization solutions in Ant after combining business scenarios.
In order to give back to the community the optimization accumulated by Ant’s multi-model reasoning, and cooperate to build better open source reasoning services and active technology ecology, the cooperation between Ant Group and NVIDIA Triton community is officially launched. The two parties will establish a cooperative development team, conduct regular technical exchanges, and jointly contribute to Triton The code base and the surrounding ecology of the construction community, work together to create an open and active Triton open source community!
Ant Group will submit the existing optimization and improvement based on Triton Server to PR and merge to the community, and will jointly establish a technical blog with NVIDIA to accumulate Triton technical articles and activate the technical ecology.
An open technology ecosystem welcomes the extensive participation of any outstanding developer. In this event, I was also honored to invite Wang Xin, the technical director of the forecasting engine team of Meituan's machine learning platform, to share "Triton's application in Meituan's data center" with everyone.
After Meituan solved the pain points of the machine learning model in Meituan, it tailored a specific plan for the implementation of Triton. It turns out that the business benefits and effects of using Triton are so high!
The problems faced by JD.com are different from those of Meituan. Their diversified algorithm requirements in scenarios such as content understanding, risk control, and knowledge graphs have brought great technical challenges to model reasoning. Yang Peijun, a retail algorithm middle-end technical architect from JD.com, shared "Triton's application and practice in JD.com's high-performance reasoning engine".
In his sharing, he introduced the high-performance reasoning framework based on the secondary development of Triton for CV/NLP scenarios in Jingdong Algorithm Center, and combined with the practice of model reasoning in typical business scenarios, he explained the architecture evolution and performance of the reasoning framework Optimize work.
AI algorithms continue to evolve and application scenarios continue to enrich. Efficient and stable AI reasoning services are an important part of transforming AI models into enterprise productivity. In the final roundtable discussion session, everyone had a heated discussion on the topic of "the present and future of AI reasoning service".
Special guests for this event : Senior Engineer of NIO Autonomous Driving R&D Platform (NADP) —Mr . A group of lecturers who shared the topic put forward their own views.
At the Triton Meetup 2023 site, everyone was enthusiastic. After the roundtable meeting, the collision of ideas between each other did not stop there, and the small partners in the venue were actively absorbing and sharing their views!
We have fully felt everyone's enthusiasm for Triton and the field of AI reasoning services, thank you for your participation! Triton Meetup will also look forward to meeting you in the next city!
learn more...
The PPT in the sharing session of this event has been uploaded and organized for everyone. Interested partners are welcome to click on the " AI Infra " official account business card below to follow and private message " Triton " to get a reply~