NVIDIA, Rockchip and other 13 domestic and foreign hardware manufacturers launched the Flying Paddle ecological distribution version, integrating software and hardware to create a win-win situation

Flying Paddle debuts at NVIDIA GTC 2023, collaborative innovation of software and hardware to consolidate AI infrastructure

With the popularization and application of AI in the industry, the demand for artificial intelligence computing power in various types of scenarios is constantly being released, providing training and reasoning AI chip products with powerful capabilities are flourishing. The effective utilization of chip computing power requires adaptation and coordination with mature deep learning frameworks.

Recently, NVIDIA, the leader in the field of artificial intelligence computing, held the Global Developers Conference GTC 2023. The cooperation between GPU chips and Baidu Fei Paddle deep learning platform continued to deepen. Baidu technical experts shared how Fei Paddle supports big data Model production and application, large-scale graph learning training technology PGLBox, end-to-end large model development kit PaddleFleetX, full-scenario high-performance AI deployment tool FastDeploy and other cutting-edge content.

The deep learning framework is a key software platform that connects software and hardware. The joint optimization of the framework, AI chips, and computing systems is the key to fully releasing computing power and achieving ultimate optimization of AI infrastructure. As China's first independent research and development, open source and open industrial-level deep learning platform, Flying Paddle has developed into the largest deep learning framework and enabling platform in the Chinese market. Its technical and ecological strength are widely recognized by hardware manufacturers, and cooperation is becoming increasingly close. , thus also providing developers with a convenient experience integrating software and hardware.

Picture: Ma Yanjun, General Manager of Baidu AI Technology Ecosystem, giving a speech at GTC 2023

The cooperation between hardware manufacturers and Fei Paddle has experienced "co-aggregation" and "co-research" There are three stages of "co-creation", which are deepened year by year. Based on the judgment of the collaborative development trend of AI basic software and hardware, Feipiao took the lead in connecting with leading domestic and foreign hardware manufacturers in 2020, and received positive response and support. In 2021, Feipiao and its hardware partners will work together to solve the problems of complex adaptation technology solutions and high costs. Feipiao launched a multi-level, low-cost unified hardware adaptation solution, including custom operator library access and custom communication. With many basic software modifications such as libraries, hardware partners can flexibly choose according to hardware characteristics, greatly reducing adaptation costs. Up to now, more than 40 hardware manufacturers have deeply integrated and optimized with Flying Paddle, and they are basically fully adapted to domestic chips.

In 2022, Feipiao and its hardware partners will further develop from adaptation optimization to the "co-creation" stage of collaborative innovation. Hardware partners hope to enrich their own AI software stacks through customized versions of the flying paddle framework, model libraries, etc. Fei Paddle can also help more developers realize AI application innovation at a low threshold by improving the underlying hardware development experience.

NVIDIA and Feipiao took the lead in completing the joint creation work. Taking NVIDIA cooperation as an example, Flying Paddle has jointly released a manufacturer-customized version of Flying Paddle framework and model library based on the Flying Paddle open source version with more hardware partners at home and abroad.

According to reports, NVIDIA ranks Flying Paddle as one of the top three deep learning framework partners in the world, and has set up a dedicated engineering team in China to support it and empower Flying Paddle. The two parties jointly developed three major products: NGC, a flying paddle container customized and optimized based on NVIDIA GPU, which includes the latest NVIDIA toolkit (CUDA, DALI, etc.) and optimized flying paddle version to provide developers with the best development environment; NVIDIA Deep Learning Examples provides users with flying paddle model examples that are easy to train and deploy; a series of free courses and co-branded certifications are created at the NVIDIA Deep Learning Training Center (DLI) to help developers improve training performance and simplify code, and provide authoritative recognition.

Ma Yanjun, general manager of Baidu AI technology ecosystem, said in a GTC speech that the development of artificial intelligence applications cannot be separated from the joint efforts of hardware partners. We have joined forces with 13 partners including NVIDIA, Intel and Arm to jointly release the Flying Paddle ecological release to achieve in-depth optimization of software and hardware collaboration. The number of partners in Fei Paddle's "Hardware Ecosystem Co-Creation Plan" has reached 29, gathering the power of the ecosystem to jointly build a full-stack AI platform integrating software and hardware to promote the high-quality development of the artificial intelligence industry chain.

Flying Paddle's "Hardware Ecosystem Co-Creation Plan" mainly includes four key tasks: joint research and development, resource sharing, joint authorization, and training empowerment, aiming to collaborate with strategic partners' software Hard integration and comprehensive co-creation to better serve developers.

In terms of joint research and development, Flying Paddle and hardware partners have established a virtual joint technology research and development team to build joint R&D infrastructure, optimize the performance of the model, and launch a manufacturer-customized version of the Flying Paddle framework, greatly improving the Flying Paddle model. The training inference performance on various AI hardware provides developers with more and better choices.

In terms of resource sharing, Fei Paddle shares promotional resources with hardware partners, cooperates to polish Fei Paddle hardware manufacturers’ characteristic marketing toolkits, and provides a display platform through brand activities, competitions, closed-door meetings, digital exhibition halls, etc. Help business development.

In terms of joint licensing, Fei Paddle will give priority to recommend solutions jointly optimized with hardware partners and Fei Paddle’s unique capabilities to ecological partners. Feipiao also jointly developed Feipiao all-in-one machine based on the hardware platform to provide enterprises with comprehensive, convenient, efficient, stable and reliable one-stop artificial intelligence services.

In terms of training empowerment, Fei Paddle and its hardware partners jointly developed Fei Paddle hardware usage and development tutorials, and provided a series of courses and joint certification certificates in the learning and training communities of both parties. Among them, Feipiao AI Studio is already the largest artificial intelligence learning and training community in the country, with more than 3 million registered users. Feipiao AI Studio has also opened a hardware manufacturer area to provide developers with opportunities to experience more AI hardware.

With the widespread application of artificial intelligence technology in various industries, the industry has entered the stage of collaborative innovation of algorithms and hardware from independent hardware computing power drivers and algorithm innovation drivers. Feipiao will use the hardware ecological co-creation plan and combine the characteristics of its partners' own software and hardware basic development stacks to truly realize the interconnection and innovative development of software and hardware.

Feipiao adheres to the industry chain positioning of operating systems in the AI ​​era and openly embraces all hardware manufacturers. Through technology joint research and development and ecological sharing and co-creation, Feipiao will work with hardware partners to develop more software and hardware collaborative products and services, accelerate the application of artificial intelligence, and promote the realization of large-scale production in the AI ​​industry.

The above is the detailed content of NVIDIA, Rockchip and other 13 domestic and foreign hardware manufacturers launched the Flying Paddle ecological distribution version, integrating software and hardware to create a win-win situation. For more information, please follow other related articles on the PHP Chinese website!