keep on record Console
<View all products
Artificial intelligence platform PAI (original machine learning platform)
Play video
The artificial intelligence platform PAI (Platform for AI, the original machine learning platform PAI) is a machine learning/deep learning engineering platform for developers and enterprises. It provides full link services for AI development including data annotation, model building, model training, model deployment, and reasoning optimization. It has 140+optimization algorithms built in, and has a wealth of industry scenario plug-ins to provide users with low threshold High performance cloud native AI engineering capability.
Intelligent computing service PAI - Lingjun newly released the public cloud serverless version to support large-scale training scenarios such as AIGC
Product documentation Product pricing Notebook Gallery
What's New Hot text recommendation Best Practice of Tongyi Qianwen Open Source Model in PAI Lingjun Product capability PAI-ACC provides AI training reasoning acceleration service for enterprises Latest information [Heavyweight - Continuous Update] Collection of Alibaba Cloud AI open source projects
Best Practices Best Practices Best Practice of Large Model RAG Dialog System Best Practices PAI+vector retrieval quickly builds large model knowledge base dialogue Best Practices Deploy AI video generation applications with PAI-EAS in 5 minutes

Intelligent computing service PAI - Lingjun new upgrade

PaaS platform products for large-scale in-depth learning and integrated intelligent computing scenarios support public cloud serverless, single rent and hybrid cloud product forms, and provide one-stop AI engineering full process platform and heterogeneous integration computing power jointly optimized by software and hardware.
Understand and purchase intelligent computing service PAI Lingjun
Rich Lingjun product forms
Public Cloud Serverless
Serverless platform products can quickly pull up AI computing tasks with one click, and complex heterogeneous systems can be automatically operated and maintained for easy management. Seamless connection with cloud computing, storage, network and other products.
Public Cloud Single Rental
Establish a customer specific cluster on the cloud, and a single customer can enjoy a set of AI platform and operation and maintenance services. Convenient operation management, interoperability of cloud products, and use of standard computing, storage, and network services on the cloud.
Apsara hybrid cloud version
It supports hybrid cloud standard architecture, provides complete computing, network, storage, account (ASCM), standard SDK/Open API, independent deployment of physical resources, and supports service providers to build businesses based on customer scenarios.

Product specification

New customers can try DSW, DLC and EAS products for free in limited quantity
Learn more

AI Platform PAI Product Functions

AI platform PAI is an AI engineering platform for developers and enterprises, providing full process services covering data preparation, model development, model training and model deployment.
Data preparation
In the data preparation phase, PAI iTAG provides intelligent data annotation services, supports image, text, video, audio and other types of data annotation, and supports multimodal data annotation; It provides rich annotation content components and topic components. Users can directly use the preset annotation template of the platform or customize the template for data annotation. At the same time, it provides full custody data annotation outsourcing services.
Learn more about PAI iTAG
Model development
In the model development phase, modeling can be completed through PAI Designer and PAI-DSW development tools.
Visual modeling PAI Designer
Provide a low code development environment, built-in 100+mature machine learning algorithms, complete modeling through drag and drop, and help users achieve low code development AI related services. Learn more about PAI Designer
Interactive modeling PAI-DSW
It provides an interactive programming environment, built-in JupyterLab, WebIDE and Terminal, and provides the underlying Sudo permission, which is open and flexible. Learn more about PAI-DSW
model training
In the model training phase, large-scale distributed training tasks can be launched through PAI-DLC; According to usage scenarios and computing power categories, it can be divided into training tasks using Smart Smart Computing to support large models, and using Alibaba Cloud general computing power nodes to support general training tasks.
PAI-DLC on Smart Computing Resources
Based on the integrated optimization technology of software and hardware, the super large-scale distributed deep learning task runs with core advantages such as high performance, high efficiency and high utilization. It supports forms such as public cloud serverless and single rent, and provides AI engineering full process platform and heterogeneous integration computing power integrating software and hardware. Learn more about intelligent computing service PAI Lingjun
PAI-DLC on Universal Computing Resources
The training platform based on Alibaba Cloud general computing (e.g. ECS, EGS, ECI) supports TensorFlow, PyTorch, MPI and other training frameworks, and is flexible, stable, and easy to use. Learn more about PAI-DLC
Model deployment
In the model deployment phase, PAI-EAS provides online prediction services, and PAI Blade provides inference optimization services.
Model online service PAI-EAS
Support users to deploy models as online reasoning services or AI Web applications with one click. It is suitable for real-time reasoning, asynchronous reasoning, offline reasoning and other scenarios. Learn more about PAI-EAS
Universal Reasoning Accelerator PAI Blade
All optimization technologies of Blade are designed for universality, and can be applied to different business scenarios. Through joint optimization of model systems, the model achieves the optimal reasoning performance. Learn more about PAI Blade
AI Asset Management
PAI supports users to manage important AI production materials and development outputs such as models, datasets and images through the whole life cycle, and provides AI asset sharing, horizontal comparison of training effects, backtracking of abnormal problems and other capabilities to achieve cost reduction and efficiency increase in AI development and application processes.
Learn more about AI asset management capabilities
AI acceleration service
The PAI-ACC AI acceleration service is an AI acceleration engine provided by Alibaba Cloud AI platform PAI, providing enterprises with the ability to accelerate training and reasoning. Through data set acceleration, calculation acceleration, optimization algorithm, scheduling algorithm, resource optimization technology and other means, the speed, ease of use and stability of AI training and reasoning are improved, and the efficiency of AI computing is greatly improved.
Learn more about AI acceleration services
One stop AIGC design platform
PAI ArtLab is an AIGC intelligent design tool created by AI platform PAI for design professionals. It supports mainstream cultural maps and model training applications such as cloud based Stable Diffusion and Kohya, and provides AIGC full scene capabilities such as dataset management, model management, model training, and AI mapping. Support AI painting teaching, unified management and authorization of multiple accounts and other enterprise level capabilities.
Learn more about PAI ArtLab
Use Now

Product architecture

AI platform PAI is a cloud native machine learning/deep learning engineering platform for developers and enterprises. Its services cover the whole link of AI development, with 140+optimization algorithms built in, and a wealth of industry scenario plug-ins.
Product advantages
Easy to use
Encapsulates 140+machine learning algorithms to support low code model training and one click deployment.
High performance
It supports high-dimensional sparse data scenarios and large-scale sample model accelerated training.
low cost
Support CPU/GPU hybrid scheduling, cloud native elastic scaling, and flexible billing.
Rich industry plug-ins
Multi scenario plug-ins and solutions are provided to help enterprises quickly build business applications.
Related products
DataWorks
MaxCompute
E-MapReduce
Real time calculation Flink version

Application scenarios

Document and diagram generation
Intelligent recommendation
User growth
End side over opening
Smart Container
Intelligent cultural and creative
Financial quantitative scientific calculation
Intelligent customer service
Content risk control
A Solution for Generating Diffusion Document and Graph
PAI provides an end-to-end, lightweight pure white box solution in the field of document and map generation. You can directly use the default model provided by the PAI to perform deployment reasoning on the PAI to achieve the document and diagram generation function of specific business scenarios. You can also quickly customize and build a document and graph generation model, quickly generate images corresponding to input text, and serve various downstream AI creative application scenarios.
Able to solve
AI creative image generation
It supports the rapid generation of images with various styles that conform to the text description.
Customized model fine-tuning
Select model tuning and online deployment according to self owned business scenarios.
Lightweight application construction
Quickly build the complete link of document and graph generation application.
Recommended combination
PAI
OSS
Scenario and scheme introduction
The PAI platform, combined with business scenarios, provides a white box intelligent recommendation solution for full link recommendation from call back to sorting, enabling customers to master all core technical links of the recommendation business. In addition, it also provides GraphSage, MultiTower, DIN and other classic recommended algorithms in the industry. It can help customers build enterprise level recommended solutions that deeply fit their business in about 10 days.
Capable of providing
Improve the accuracy of content recommendation
Reduce the instability of manual recommendation
Deep integration with services, full link connection, rapid construction
Recommended combination
PAI-Designer
PAI-EAS
MaxCompute
E-MapReduce
Scenario and scheme introduction
The PAI platform user growth solution, combined with business scenarios, helps customers quickly realize the ability to recall lost users, activate silent users, buy volume price difference, user churn warning, user LTV management, etc. in about a week by providing a mature algorithm component mode, and enables customers to master all technical details independently.
Capable of providing
Lost user recall
Reduce manual strategy instability
Reduce marketing costs
Recommended combination
PAI-Designer
SMS service
E-MapReduce
MaxCompute
Scenario and scheme introduction
By embedding a super division model on the mobile phone, the low definition video can be automatically converted into high definition in real time, thus improving the user experience and saving CDN, transcoding and other service costs
Capable of providing
data source
Base model can be generated based on massive open source video and image data, or Finetune model can be generated based on data provided by users
model training
Carry out model training through machine learning platform PAI, and carry out automatic optimization of super parameters during model training
Model evaluation and testing
Effect, stability and power consumption are polished and verified in a large number of real business scenarios
Model deployment
Automatically implement the compatibility adaptation between the model and the MNN framework, and complete the deployment of the model on the mobile phone side
Recommended combination
Machine learning PAI
Intelligent container commodity analysis solution
For the smart container commodity analysis scenario, there are problems such as frequent SKU updates and large service online labeling. To solve these problems, Alibaba Cloud's machine learning PAI platform provides a set of commodity analysis solutions based on computer vision detection algorithms, which can quickly update commodity analysis services by preparing a small number of real counter test images (10-20), image annotation results, and relevant ring photos of commodities to be launched.
Able to solve
Function support
High precision commodity display positioning and analysis, including commodity statistics, hierarchical analysis, vacancy analysis, and support for expanding commodity categories.
Customized module support
Selectively support model tuning and deployment according to free business scenarios.
Recommended combination
PAI
0SS
Intelligent cultural and creative solutions
In the field of intelligent text creation, PAI provides end-to-end, diversified pure white box solutions, supports different levels of model structures, and currently supports the generation of Chinese news titles, text generation, problem generation, composition generation, ancient poetry generation and other functions. You can quickly customize and build various types of intelligent cultural and creative models according to different needs, carry out rapid creative applications for different text inputs, and serve the needs of text creation scenarios.
Able to solve
Multi type task support
Support the construction of text summary and news title generation models, and continuously expand the function pool.
Customized module construction
Select model tuning and online deployment according to self owned business scenarios.
Recommended combination
PAI
0SS
Scenario and scheme introduction
A large number of financial quantitative data often encounter performance bottlenecks in large-scale computing. The best practice of financial quantitative science and technology on the PAI platform will bring you an accelerated solution for large-scale scientific computing based on PAI-DSW, Mars and PAI-Eflops.
Able to solve
Difficulties in environmental management
Localization and self built strategy R&D platform easily lead to inconsistent environment
Complex resource allocation
Uneven task resource allocation leads to resource preemption
Strategy backtesting is troublesome
Uneven task resource allocation leads to resource preemption
Recommended combination
Machine learning PAI
Object Storage OSS
MaxCompute
Scenario and scheme introduction
PAI provides end-to-end pure white box integrated solution in the field of intelligent customer service dialogue system. Customers only need to prepare their own FAQs (Frequently Asked Questions) and knowledge map data in relevant fields, then they can customize the AI process from algorithm construction to model deployment, quickly implement the intelligent customer service business system in the corresponding fields, and reduce the labor costs and business learning costs of newcomers.
Capable of providing
Pure white box
The intelligent customer service business system can be customized according to your own business scenarios.
end to end
From data preparation to model deployment reasoning, it provides a full link system construction process.
Evidence
All the answers are based, avoiding the inexplicability of pure in-depth learning programs.
Robust and controllable
If there are problems anywhere in the system, there are corresponding exception handling and analysis response mechanisms.
Recommended combination
Object Storage OSS
ECS
VPC
ApsaraDB for Redis
Scenario and scheme introduction
PAI provides end-to-end, lightweight pure white box solutions in the field of content risk control. You can build an end-to-end risk control system based on the pre training model provided by PAI according to your business scenario, so as to identify high-risk content.
Capable of providing
Lightweight
End to end rapid construction of text and image content risk control system
customization
Customized white box scheme to build risk control system according to self owned business scenarios
cost reduction
Reduce the cost of manual content review
Recommended combination
Machine learning PAI
Object Storage OSS
MaxCompute
Model training acceleration
Support users to quickly build a process for training and acceleration of Transformer model, greatly improve the speed of model training throughput or convergence, and optimize hardware resource consumption.
Blade Universal Reasoning Acceleration
Combined with various optimization techniques, the trained model is optimized to achieve the optimal reasoning performance. At the same time, the C++SDK can deploy optimized model reasoning to help users quickly apply models to production.
General text marking
The built-in default model can be directly used for deployment reasoning to build a text marking system for specific business scenarios; At the same time, it supports the user-defined construction of text marking models, identifies different types of text labels, and serves other business applications such as recommendations.
General video marking
It supports model structures of different orders of magnitude, multimodal input forms of video frame input and text+video, and output forms of single label and multi label. Quickly customize and build various types of video marking models according to different needs, quickly identify various types of video tags from video data, and serve downstream recommendations or other application scenarios.
image retrieval
PAI provides an end-to-end, lightweight pure white box solution in the field of similar image matching and image retrieval. You only need to prepare the original image data, and you can quickly customize the image self-monitoring model without annotation. Finally, the model is deployed and reasoned on the PAI to realize the system of searching graphs with graphs for specific business scenarios.

Customer Stories

micro-blog
PAI helps micro blog machine learning platform to provide model training capability for top business such as popular micro blog/feed recommendation of micro blog.
Highlights of middle school and youth
In combination with the PAI intelligent recommendation scheme, the Zhongqing Watch APP has increased the CTR of news by 80%, improved the overall user perception of news, and increased the average stay time of each user on the platform by 10 minutes, greatly improving user stickiness.
Pumpkin Movie
Pumpkin Movie APP and TV both adopt the PAI recommendation scheme, which effectively improves the accuracy of video recommendation CTR prediction model by 50%.
CSDN
The PAI recommendation scheme is used in the recommendation of CSDN information flow, and the CTR of content search recommendation is increased by 80%.

Video Tutorials

PAI forward-looking technology sharing | AI Architect Growth Plan Big Talk | GraphLearn on PAI ｜ More video content>>
Getting Started Tutorial more
Advanced Tutorial more
Scene Practice more

Product Dynamics

2018-12-20 Function optimization
PAI releases new console
View details
New products on January 1, 2019
PAI online in-depth learning development product DSW release
View details
2019-02-14 Function optimization
PAI DSW creates instance details optimization
View details
2019-04-06 New functions/specifications
PAI-EAS online forecasting service is released as an independent sub product of PAI
View details
2019-04-08 Price adjustment
PAI-DSW postpaid catalog price adjustment
View details
2019-04-16 Function optimization
PAI DSW experience optimization phase II, terminal adds copy and paste function, etc
View details
New functions/specifications on April 18, 2019
Deep learning map network visualization development - FastNeuralNetwork
View details
2019-04-25 New Features/Specifications
PAI-DSW supports file breakpoint resume function
View details
2019-05-16 New Features/Specifications
PAI-TF is upgraded to version 1.8 compatible with the open source community
View details
2019-06-20 New functions/specifications
PAI-EAS adds the ability to manage user exclusive resource groups, and provides a model prepaid deployment method
View details
2019-07-16 New functions/specifications
Machine learning PAI AutoLearning released
View details
2019-07-31 New functions/specifications
PAI user-defined algorithm function release
View details
2019-08-09 New Functions/Specifications
PAI-STUDIO supports using Tensorflow to read and write MaxCompute data
View details
2019-09-30 New functions/specifications
Public test of reasoning and optimization capability of PAI blade model
View details
October 14, 2019 New Region/New Availability Zone
PAI EAS opened in Shenzhen region
View details
2019-12-02 New functions/specifications
PAI-EAS online prediction exclusive resource group fully supports post payment!
View details
New functions/specifications on December 17, 2019
PAI model management+model online reasoning optimization accelerate product release on the PAI public cloud console
View details
2020-02-13 New Region/New Availability Zone
PAI-EAS officially opens new regions in Indonesia (Jakarta), India (Mumbai) and Germany (Frankfurt)!
View details
2020-02-14 New Features/Specifications
PAI AutoLearning Recommended Solution Release
View details
2020-03-06 New Features/Specifications
PAI Studio supports INT data
View details
2020-03-17 New Features/Specifications
PAI Studio automatic feature cross exploration function goes online
View details
2020-03-17 New Features/Specifications
Alibaba Cloud machine learning PAI data annotation public beta release
View details
2020-03-26 New Region/New Availability Zone
PAI Studio supports deep learning of CPU version
View details
2020-04-02 New Region/New Availability Zone
Machine learning PAI DSW opens in India (Mumbai)
View details
2020-05-11 New Features/Specifications
AliCloud machine learning product - cloud native interactive modeling product DSW2.0 released
View details
2020-05-15 New Features/Specifications
Cloud native deep learning training platform PAI-DLC (Deep Learning Containers) released
View details
2020-07-09 New Region/New Availability Zone
PAI-DSW 2.0 New Area Launch
View details
2020-07-10 New Region/New Availability Zone
PAI-EAS is open and supports Hong Kong region!
View details
2020-08-01 New Features/Specifications
AutoLearning releases visual modeling platform
View details
2020-08-10 New Features/Specifications
PAI Studio Online GraphSage Algorithm
View details
2020-10-23 New Features/Specifications
PAI Studio online experiment handover and sharing function
View details
2021-01-31 New Features/Specifications
DSW3.0 Heavy Release
View details
2021-08-01 New Features/Specifications
Release of super dividing capacity on PAI end
View details
2021-08-25 New Features/Specifications
Alibaba Cloud Label Product PAI iTAG Released
View details
2021-09-30 New Features/Specifications
PAI New Visual Modeling Designer Released
View details
2022-03-21 New Features/Specifications
EAS adds a timed scaling function to support image deployment and publishing of grpc/websocket protocol
View details
2022-03-27 New Features/Specifications
PAI Blade newly supports Tensorflow version 2.7
View details
2022-04-16 New Features/Specifications
PAI Designer adds exception detection class, recommendation class, data source class and user-defined algorithm class components
View details
2022-04-30 New Features/Specifications
Support model training using Flink fully hosted resources
View details
2022-06-22 New Features/Specifications
Designer adds a variety of visual analysis capabilities
View details
2022-07-01 New region/new zone
PAI-DLC proprietary resource group officially opened in East China 1 (Hangzhou)
View details
2022-07-04 New Features/Specifications
EAS benchmark service automatic pressure test function release
View details
2022-07-15 New Features/Specifications
PAI Designer Publish Custom Python Script V2 Component
View details
2022-08-02 New Features/Specifications
PAI Designer adds six machine learning algorithm components such as XGBoost and DBSCAN
View details
2022-09-30 New Features/Specifications
EAS service grouping and asynchronous reasoning function publishing
View details
2022-11-18 New Features/Specifications
PAI-DSW instance update support and observability enhancement
View details
2022-11-21 New Features/Specifications
DatasetAccelerator
View details
2022-11-30 New Features/Specifications
EAS machine node self operation and maintenance function release
View details
2022-12-01 New Features/Specifications
New user-defined template function in Designer
View details
2022-12-05 New Features/Specifications
PAI Designer adds Prophet timing algorithm, streaming PyAlink script, vector aggregation and other algorithm components
View details
2022-12-08 New Features/Specifications
EAS supports Yitian 710 series computing resources
View details
2023-01-13 New Features/Specifications
The Designer supports the deployment of offline data processing and prediction full link pipeline as online services
View details
2023-02-06 New Features/Specifications
EAS supports multi specification instance selection
View details
2023-02-13 New Features/Specifications
EAS supports preemptive resource instances (Spot Instances)
View details
2023-03-23 New Features/Specifications
AI asset management - model management function upgrade
View details
2023-04-04 New Features/Specifications
EAS's new quick service deployment console goes online
View details
2023-04-11 New Features/Specifications
PAI-EAS launches GU series heterogeneous GPU resource specifications
View details
2023-04-19 New Features/Specifications
EAS publishing elastic resource pool function
View details
2023-05-20 New Features/Specifications
PAI Python SDK officially released
View details
2023-06-30 New Features/Specifications
Support the creation and management of user-defined algorithm components
View details
2023-07-2025 New Features/Specifications
Support Llama2 series model fine-tuning and reasoning
View details
2023-07-2025 Function optimization
The feature platform supports table storage TableStore
View details
2023-08-11 New Features/Specifications
PAI Lingjun Intelligent Computing Online High Performance Computing Power Specification gu8xf
View details
2023-08-24 New Features/Specifications
Support stable diffusion fine tuning deployment
View details
2023-08-30 Function optimization
RLHF, Large model training framework launched
View details
2023-08-30 Function optimization
Distributed training product DLC supports MPI training framework
View details
2023-09-01 New functions/specifications
DSW/DLC supports mounting encrypted NAS
View details
2023-09-01 New functions/specifications
Support the configuration of DSW instances to access the public network through the VPC gateway
View details
2023-09-02 New Features/Specifications
EAS publishes asynchronous reasoning service to automatically expand and shrink capacity
View details
2023-09-05 New Features/Specifications
EAS deployment ChatGLM&Langchain
View details
2023-09-06 New Features/Specifications
FeatureStore public beta release
View details
2023-09-06 New Features/Specifications
Distributed training product DLC supports monitoring index subscription and alarm
View details
2023-09-07 New Features/Specifications
DSW supports pay as you go instances based on ECI+cloud disks
View details
2023-09-07 New Features/Specifications
EAS one click deployment of the universal thousand question model service
View details
2023-09-08 New Features/Specifications
EasyCKPT High Performance CKPT Released
View details
2023-09-15 New Features/Specifications
DSW launches Community Gallery
View details
2023-09-17 New Features/Specifications
DSW AI photo development based on EasyPhoto
View details
2023-10-17 Function optimization
DLC adds subscriptions and alerts for multiple key monitoring indicators
View details
2023-10-18 New Features/Specifications
Resource quota (Quota) v1.0 release
View details
2023-10-19 New functions/specifications
The message center in the workspace supports the alarm function of phone, SMS and email notification
View details
2023-10-19 New functions/specifications
The QuickStart of Smart Computing Edition was officially released
View details
2023-11-01 New Features/Specifications
PAI access to ACS container computing service
View details
2023-11-09 Function optimization
Distributed training DLC aggregation log publishing
View details
2023-11-10 Function optimization
Smart computing multi tenancy (Serverless) node self-healing v1.0
View details
2023-11-15 New Features/Specifications
EAS-LLM big model reasoning service publishing
View details
2023-11-16 New Features/Specifications
PAI Releases Automated Machine Learning (AutoML) Platform
View details
2023-12-04 New functions/specifications
DSW instances support SSH direct connection access
View details
2023-12-12 New Region/New Availability Zone
Designer officially opened in Indonesia (Jakarta)
View details
2023-12-20 New Features/Specifications
PAI publishes four scheduling policy functions based on Quota queue
View details
2024-01-03 New Features/Specifications
Distributed training DLC computing power health detection release
View details
2024-01-05 New Features/Specifications
EAS one click deployment of AI video generation application
View details
2024-01-15 Function optimization
EAS Minimal Deployment Function Release
View details
2024-02-01 New Features/Specifications
EAS Serverless model service grayscale test invitation
View details
2024-02-20 New Features/Specifications
DLC (distributed training) supports the submission of training tasks using free time resources
View details
2024-02-28 New Features/Specifications
The Designer supports LLM data preprocessing operators and common templates
View details
2024-03-01 Stop service
GPU server and corresponding algorithm components in Designer are offline
View details
2024-03-15 New Features/Specifications
DSW supports users to develop AI+big data
View details
2024-03-21 New Features/Specifications
Function of DSW publishing file transfer station
View details
2024-03-25 New Region/New Availability Zone
PAI general calculation type (Ulanqab region) officially launched
View details
2024-04-12 New Features/Specifications
EAS Serverless AI Painting Scene Publishing
View details
View all logs

Documentation and Tools