With the breakthrough of AI led by ChatGPT, various industries have set off the trend of AI implementation, from smart customer service, smart writing, smart monitoring, to smart medical care, smart home, smart finance, and smart agriculture. Who can quickly integrate AI with traditional businesses ? Combined, whoever will become the winner of enterprise digitalization and intelligent transformation will be the winner . However, the application of artificial intelligence technology faces many challenges, such as how to select suitable implementable models, to the joint application of multiple models to meet different businesses; from quickly completing the training and fine-tuning of AI models, to how to efficiently deploy models.
To address these challenges, the Paddle team launched the Paddle AI Kit (PaddleX) - a one-stop development platform for selected industrial practical models. Contains the following features:
- Excellent algorithm library
It contains 36 selected models with more than 10 task scenarios and provides complete model inference benchmark data. Developers can choose the most appropriate model based on business needs.
- Convenient development method
It provides a no-code + low-code development method. In just four steps, you can complete the full-process AI development of data, training, verification and deployment.
- Efficient model deployment
Complete deployment capability support enables rapid industrial-level deployment across platforms and hardware.
- Rich domestic hardware support
PaddleX can not only be used in the AI Studio cloud, but also supports Windows local deployment, and is continuously enriching the Linux version, Kunlun Core version, Ascend version and Cambrian version.
- Win-win joint creation and co-construction
In addition to conveniently developing AI applications, PaddleX also provides everyone with opportunities to obtain business benefits, explore more business space for enterprises, and achieve common growth and win-win results.
As an efficient development artifact, PaddleX is worth every developer. Next, we will analyze the core capabilities of PaddleX in detail to help you better understand and apply this tool. Add the official channel of AI Studio Flying Paddle AI Kit and discuss it with everyone!
AI Studio flying paddle AI kit official channel
https://aistudio.baidu.com/community/channel/610
If you want to know more about PaddleX, you can also join the official communication group by adding operating classmates and replying to [PaddleX]!
PaddleX cloud usage link
https://aistudio.baidu.com/modelsoverview?category=%E4%BA%A7%E4%B8%9A%E6%96%B9%E6%A1%88&supportPaddlex=1&sortBy=weight
PaddleX Windows local download link
https://www.paddlepaddle.org.cn/paddle/paddleX
Excellent algorithm library
First, let’s introduce the collection of selected algorithm models in the PaddleX platform. In the process of supporting the implementation of enterprise AI, Paddle has summarized many years of practical experience and precipitated 36 selected models for 10+ task scenarios, which have been unified into PaddleX, and this collection continues to be enriched. From now on, flying paddle developers can complete the development and deployment of different AI tasks in one software.
However, there is often more than one model algorithm for the same task. How to choose a model suitable for the business scenario? Don’t panic, PaddleX provides complete model benchmark data, which can help developers choose high-precision/fast/balanced models based on business needs, thus getting rid of the difficulty of model selection.
For the model selection capabilities provided by PaddleX, please see
https://ai.baidu.com/ai-doc/AISTUDIO/vlistbd6p
It will continue to be enriched and improved in the future, so stay tuned~
Easy way to develop
After talking about algorithm library and model selection, it is time to enter the core model development process. If the first impression PaddleX brings to everyone is probably the toolbox/developer dual-mode linkage development method.
The toolbox mode focuses on being simple and brainless . Prepare data sets that meet the specification requirements, and then perform data verification, model training, evaluation testing, and model deployment in the webui interface. The entire process of AI model development is completed smoothly. .
The developer mode provides a free square . It is essentially based on the VS Code integrated development environment and has built an ultra-high-level API (UAPI) base carefully designed by the PaddleX team. 4 core APIs can complete model development and deployment. , the rest is all up to everyone’s free play.
UAPI usage instructions
https://ai.baidu.com/ai-doc/AISTUDIO/7ll0srufz
The two development models are not separate, but closely related. To put it simply, each key step of the toolbox mode corresponds to a UAPI interface in the developer mode. All configuration modifications in the toolbox mode will be implemented in real time, making it possible to seamlessly switch between the two modes and continue development.
On the other hand, if your friends feel that the toolbox mode is not useful enough, you can "do it yourself and have enough food and clothing". After editing the Python code in toolbox mode in developer mode, switch to toolbox mode and click "Reload" to complete the toolbox DIY. We highly encourage everyone to carry out secondary development of the toolbox. Excellent ideas will have the opportunity to be officially included in PaddleX selected templates, allowing developers to benefit from your ideas.
Efficient deployment model
For many friends who are learning AI, the task comes to an end after training to obtain a model that meets the accuracy requirements. However, for the implementation of AI models, a very critical step is needed - model deployment. In many actual scenarios of industrial applications, after perfect model deployment, the inference performance can be improved several times to meet real business needs. Paddle provides high-performance reasoning libraries such as Paddle Inference and Paddle Lite, as well as FastDeploy, which focuses on fast and convenient deployment experience, to help the majority of Paddle developers complete this critical step. What's great is that PaddleX seamlessly integrates the capabilities of FastDeploy. You can easily click on your deployment environment requirements to obtain the corresponding deployment package. After obtaining the deployment package, you can follow the detailed
instructions to go to the target. The device is deploying the AI model~
The selected models of flying paddles currently launched on PaddleX basically meet the deployment support conditions in the table below.
In the future, PaddleX will also support C#, a variety of terminal-side hardware, and service-based deployment functions, so stay tuned!
Rich domestic hardware support
PaddleX can currently be used on the AI Studio Galaxy community cloud and Windows local side, but this is not enough. In order to meet the R&D and production needs of more users, PaddleX has planned a rich and complete domestic hardware support plan, as shown in the table below. Please always pay attention to the Flying Paddle official account, so stay tuned~
Win-win joint creation and co-construction
In addition to making it easier to develop AI models and applications, PaddleX also provides everyone with opportunities to gain commercial benefits.
After partners develop industrial applications based on PaddleX, they can set the application price and open it to other users through SDK; after other users purchase the SDK, partners can receive corresponding revenue shares.
Other users do not need to retrain the model and can directly complete data input and obtain processing results through the interface, avoiding repeated development and improving efficiency.
At present, many partners have opened and launched related applications.
Vehicle-mounted road garbage detection based on Picodet
https://aistudio.baidu.com/modelsdetail/385?modelId=385
Classification of wild animals and their products based on CLIP
https://aistudio.baidu.com/modelsdetail/386?modelId=386
Document scene detection and recognition based on PP-OCRv4
https://aistudio.baidu.com/modelsdetail/387?modelId=387
In addition, many developers have developed popular applications by connecting large and small models based on the example projects provided by PaddleX. For example, Shulian Technology, based on the PP-ChatOCR project, developed the "Bulk Commodity Trade Contract Information Extraction Application", with an accuracy rate of key information extraction reaching over 95%. After being opened to industry customers, the customer information entry efficiency increased by 150%. Technology has also reaped significant gains.
Smart contract key information extraction and analysis system
https://aistudio.baidu.com/projectdetail/6718801
If you are also willing to publish industrial applications based on PaddleX, you can add an assistant and reply "Lianchuang".
Finally, let me summarize a little: PaddleX is a one-stop, full-process, and efficient flying paddle AI suite, with high-quality flying paddle ecological models and industrial solutions. PaddleX's mission is to help AI technology be implemented quickly, and its vision is to make everyone an AI Developer!
PaddleX is currently iterating rapidly, everyone is welcome to try it out and make corrections!