All Products
Search
Document Center

Platform For AI:Get started with PAI

Last Updated:Oct 31, 2024

This topic aims to help you quickly get familiar with and use Platform for AI (PAI). You can understand the key modules and common use scenarios of PAI and the billing methods of the modules. You can also get familiar with the common use cases of PAI and gain valuable operational experience.

Overview

PAI is a machine learning and deep learning engineering platform for developers and enterprises. PAI offers end-to-end Artificial Intelligence (AI) development services, including data labeling, model development, model training, model deployment, and inference optimization. In addition, PAI provides more than 140 built-in optimization algorithms and a variety of plug-ins for AI-empowered industries to implement easy-to-use and high-performance cloud-native AI engineering capabilities.

Terms related to basic modules

The terms related to the following basic modules help you understand the core features and use scenarios of PAI.

More terms

Common use scenarios

AI painting

Scenario description:

  • Artistic creation: Use AI to generate high-quality digital artworks, which is suitable for scenarios such as illustration and concept art.

  • Advertising design: Quickly generate creative advertising images to improve design efficiency.

  • Game development: Provide rich visual materials for games to improve the diversity of game graphics.

  • Education and training: Use AI painting as a visual aid in teaching and training to enhance the learning experience.

Related model: Stable Diffusion

Involved modules in PAI: DSW and EAS

References:

Large language model (LLM)

Scenario description:

  • Content generation: Automatically generate high-quality articles, reports, and marketing copies to improve content creation efficiency.

  • Data analytics: Intelligently analyze and interpret complex data to generate easy-to-understand analysis reports.

  • Customer service chat: Provide intelligent customer service conversations to improve the response speed and customer satisfaction.

  • Educational tutoring: Help online education platforms to generate teaching content and answer questions.

Related models: Qwen, Llama, Baichuan series, and other models

Involved modules in PAI: DSW and EAS

References:

Retrieval-Augmented Generation (RAG)-based LLM chatbot

Scenario description:

  • Customer service system: Improve the intelligence and response speed of customer service conversations to improve customer satisfaction.

  • Intelligent assistant: Provide users with accurate question answering services to enhance user experience.

  • Educational tutoring: Help online education platforms to provide students with instant question answering support.

  • Medical consultation: Provide initial medical consultation services for patients to reduce the workloads of doctors.

Related models: Qwen, Llama, Baichuan series, and other models

Involved module in PAI: EAS

References:

ComfyUI-based AI video generation

Scenario description:

  • Marketing video: Automatically generate creative marketing videos to improve brand promotion effectiveness.

  • Educational videos: Quickly generate educational videos to improve the production efficiency of educational resources.

  • Entertainment content: Generate short videos, animations, and other entertainment content to enrich social media platforms.

  • Enterprise training: Produce internal training videos to reduce production costs and improve training effectiveness.

Related model: Stable Video Diffusion

Involved module in PAI: EAS

References:

AI Video Generation: ComfyUI-based Deployment

LLM data processing

Scenario description:

  • Data cleansing: Ensure data uniqueness by using the LLM-MD5 Deduplicator, LLM-Document Deduplicator, and LLM-N-Gram Repetition Filter components.

  • Content standardization: Improve data consistency by using the LLM-Text Normalizer, LLM-Special Characters Ratio Filter, and LLM-Length Filter components.

  • Sensitive information processing: Protect privacy and copyright by using the LLM-Sensitive Keywords Filter, LLM-Sensitive Content Mask, and LLM-Clean Copyright Information components.

  • Language processing: Ensure that data meets specific language requirements by using the LLM-Language Recognition and Filter component.

  • LaTeX processing: Simplify LaTeX documentations by using the LLM-LaTeX Expand Macro, LLM-LaTeX Remove Bibliography, LLM-LaTeX Remove Comments, and LLM-LaTeX Remove Header components.

Related algorithms: LLM data processing algorithms

Involved module in PAI: Machine Learning Designer

References:

Image-text pair filtering

Scenario description:

  • Content moderation: Ensure that uploaded images comply with platform specifications by using the LVM-Image-NSFW Filter (DLC) and LVM-Image-Watermark Filter (DLC) components.

  • Image optimization: Improve image quality by using the LVM-Image-Aesthetic Filter (DLC), LVM-Image-Aspect-Ratio Filter (DLC), and LVM-Image-Shape Filter (DLC) components.

  • Automatic description: Automatically generate a description for an image by using the LVM-Image-Caption Mapper (DLC) component. This helps you understand and search for an image.

  • Content recommendation: Optimize recommendation systems and improve user experience by using the LVM-Image-Text-Matching Filter (DLC) and LVM-Image-Text-Similarity Filter (DLC) components.

  • Advertisement filtering: Filter images that are suitable for advertising by using the LVM-Image-Face-Ratio Filter (DLC) and LVM-Image-Size Filter (DLC) components.

Related algorithms: LVM image preprocessing algorithms

Involved module in PAI: Machine Learning Designer

References:

Image preprocessing operators

Intelligent labeling

Scenario description:

  • Text classification: Automatically classify large volumes of text data, which is suitable for news classification and spam email detection.

  • Sentiment analysis: Identify emotional tendencies in user comments and social media content to help enterprises collect user feedback.

  • Entity recognition: Extract entity information, such as the name of a person, name of a place, and name of an organization, from text for information extraction and knowledge graph construction.

  • Image labeling: Add labels to images for image classification, object detection, and image searching.

  • Speech recognition: Add labels to, separate, or recognize audio content. This is suitable for voice assistants and customer service systems.

  • Video analytics: Add labels to video content for content sharing and moderation.

Involved module in PAI: iTAG

References:

iTAG

Large-scale distributed training

Scenario description:

  • Image recognition and processing: In computer vision scenarios, a large number of high-resolution images need to be processed when you train deep convolutional neural network (CNN) models. Distributed training can accelerate model training.

  • Natural language processing (NLP): Large-scale text data and complex model structures need to be processed for NLP tasks, such as language translation, text generation, and sentiment analytics. Distributed training can effectively handle the preceding tasks and improve the training speed and accuracy of models.

  • Recommendation system: Large amounts of user behavior data and diversified commodity information need to be processed for recommendation systems. Distributed training can accelerate the iteration and optimization of models, which improves the accuracy and timeliness of recommendations.

Involved module in PAI: DLC

References:

More use scenarios

Overview of PAI modules

QuickStart

QuickStart integrates various high-quality pre-trained models in open source AI communities. You can quickly get started with model-related operations, such as fine-tuning, deploying, and evaluating models.

iTAG

iTAG allows you to label different types of data, such as images, text, videos, and audios, or multimodal data. iTAG provides a wide range of labeling content and topic components. You can use common labeling templates that are provided by iTAG or create custom labeling templates based on your business scenarios.

Machine Learning Designer

Machine Learning Designer provides various built-in and proven machine learning algorithms to meet your business requirements in different scenarios, such as product recommendation, financial risk management, and advertising prediction. In addition, Machine Learning Designer provides an end-to-end visualized environment for modeling development.

DSW

DSW integrates multiple cloud development environments, such as JupyterLab, WebIDE, and Terminal, for coding, debugging, and job running. DSW provides various heterogeneous computing resources and open source framework images and allows you to mount datasets of the Object Storage Service (OSS), File Storage NAS (NAS), and Cloud Parallel File Storage (CPFS) types. You can manage the lifecycles of DSW instances and use DSW for development in an efficient manner.

DLC

DLC provides a flexible, stable, easy-to-use, and high-performance machine learning and training environment. DLC supports multiple algorithm frameworks, including custom algorithm frameworks, and can implement large-scale distributed deep learning jobs. DLC allows you to enjoy an optimal training environment and can improve the training efficiency and reduce costs.

EAS

EAS allows you to deploy models as online inference services or AI-powered web applications. EAS is suitable for multiple AI inference scenarios, such as real-time inference and near-real-time asynchronous inference. EAS supports automatic scaling and has a complete monitoring and maintenance system.

Scenario-based solutions

The PAI console provides various scenario-based solutions.

More information about PAI modules

Billing

Billing method

Description

Involved module

Pay-as-you-go

If you use the pay-as-you-go billing method, you are charged based on the actual usage of each module.

The pay-as-you-go billing method is suitable for short-term or uncertain workloads. It allows you to pay for resources based on the actual amount of resources that you use. The pay-as-you-go billing method is suitable for test environments, development environments, unexpected requirements, or projects in the early phases.

Machine Learning Designer, DSW, DLC, and EAS

Subscription

The subscription billing method

is suitable for long-term and stable workloads. You must pay in advance to use resources for a specific period of time, such as a month or a year. The subscription billing method is more cost-effective than the pay-as-you-go billing method for long-term use.

DSW, DLC, and EAS

Resource plan

Resource plans refer to quota plans of specific resources that you can purchase in advance.

Resource plans are suitable for scenarios in which you want to use a large number of specific resources. You can purchase quota plans for specific resources at more favorable prices.

DSW

Savings plan

You can purchase savings plans in advance, which offer specific discounts or benefits.

Savings plans provide discounted pay-as-you-go rates in exchange for committing to a specific spending amount within a specific period of time.

DSW and EAS

Pay-by-inference-duration

You are charged based on the actual inference duration. The resource specifications support automatic scaling based on the number of service requests.

This billing method is suitable for inference tasks that require indefinite quantities and is appropriate for high-concurrent requests and dynamic loads.

EAS

More information about billing

Common cases

More use cases

Contact us

To obtain more information and technical support for PAI, scan the following QR code by using DingTalk to join the PAI group.

image