Baidu Smart Cloud further integrates Arm virtual hardware to accelerate AI development and innovation
Click on the Arm Community above to follow us
Following the launch of Arm virtual hardware on Baidu Smart Cloud in August 2023, the cooperation between the two parties based on Arm virtual hardware recently ushered in a new milestone! Baidu Smart Cloud EasyDL and BML teams have verified Arm virtual hardware and applied it to Baidu's cloud-native continuous integration/continuous deployment (CI/CD) internal pipeline.
After integration, the software model can be functionally verified through the Arm virtual hardware on the EasyDL and BML development platforms, and indicator information such as inference time and performance can be directly obtained. At the same time, in key aspects of software development, including software and model comparison, screening, tuning and other development and testing work, Arm virtual hardware can bring new and efficient modern cloud development work forms and processes, greatly improving development efficiency. After integration testing, the development and adaptation time after this integration can be shortened by more than half compared to the previous model. At the same time, it can also save the time and cost of subsequent maintenance of a large amount of hardware, thereby making AI development within reach and actively helping to embrace the AI era. .
Arm virtual hardware
An efficient speed-up for software development workflows
Arm virtual hardware can help software developers, OEM manufacturers and service providers to develop software earlier than before without waiting for the physical chip to be ready, getting rid of the limitations of physical hardware, greatly speeding up the software development process and shortening verification time. Reduced related costs. Even when Arm releases a new IP or computing platform, developers can start software development work immediately to achieve almost simultaneous and efficient operations.
More importantly, Arm virtual hardware introduces a modern, highly flexible software development environment, enabling co-design of software and hardware and efficient CI/CD, eliminating the complexity of establishing or maintaining multiple development board hardware clusters. Relying on a rich ecosystem layout and a wide range of massive application devices based on the Arm architecture, Arm virtual hardware has wider adaptability, especially bringing advantages to early model and algorithm screening and verification, and simplifying software transplantation to achieve high efficiency. , large-scale development brings value.
Two-way support makes it easy to get started with AI development
Baidu EasyDL and BML platforms support the deployment of AI models on a variety of edge devices based on Arm architecture. For edge AI product and application developers, after completing model construction, training and tuning on EasyDL and BML platforms, they need to conduct performance testing and final tuning on actual edge devices in order to achieve the best results on the target device. Optimal performance, or selecting appropriate hardware configurations based on determined performance indicators, or even re-screening models to comprehensively weigh software and hardware solutions. After Arm virtual hardware is integrated into Baidu Cloud's native CI/CD workflow, developers can use its services and tools for software development, verification, and testing in daily life without being limited by the readiness of physical hardware. Improved CI/CD efficiency and accelerated the development and deployment of Arm-based edge AI.
Baidu AI development platform also continues to provide a rich library of edge devices. With the continuous launch of new Arm computing platforms and the continuous emergence of innovative edge hardware, developers are looking forward to the never-ending demand for testing and development on more and newer devices. All. In response to this, cloud-based Arm virtual hardware is constantly expanding to include new virtual devices and add more models, aiming to bring more choices in the first place.
Currently, industry applications in various fields are entering the AI native stage. It is crucial to carry out in-depth software and hardware collaborative optimization for each AI application and provide a simpler, easier to use, efficient, low-cost, and scalable development process. In the next stage, Baidu AI's dual-platform integrated Arm virtual hardware will enter the application stage based on business scenarios and actual needs in various fields. By releasing the innovation power of both parties, it will truly inject accelerated AI capabilities into more enterprises and developers. Provide more efficient and convenient DevOps/MLOps services.
A series of articles on Arm virtual hardware practice
|
Product Subscription Guide (click to read) |
|
Getting Started Guide to FVP Models (click to read) |
|
Deploy the PaddlePaddle model (click to read) |
|
Deploy the PP-OCR model (click to read) |
|
Deploy the PP-PicoDet model (click to read) |
|
Deploy the Paddle Lite model (click to read) |
* This article is an original article by Arm. For reprinting, please leave a message to obtain authorization and indicate the source.
京公网安备 11010802033920号