E2E Model Development Pipeline

End-to-end tool chain boosts development efficiency by 50% and fosters collaboration in DataOps, MLOps, and DevOps.

Cost-effective AI Compute

Diverse compute with various specifications powers large-scale distributed training and inference acceleration.

Ultra Large-Scale Training

A single job can train a model with a trillion parameters and process hundreds of petabytes of data.

High Reliability

Training jobs are automatically recovered from faults, ensuring a job failure rate of less than 0.5%. Training involving 10,000 cards can run uninterrupted for 30 days.

Architecture Overview: A Deep Dive Into ModelArts


A full-stack, full-lifecycle model development tool chain provides comprehensive AI tools and services to enable rapid service innovation.

Efficient Development

An E2E model development pipeline to efficiently develop, debug, and optimize foundation model applications and scenario-specific applications

E2E monitoring tools for intelligent operations and O&M

MLOps-based AI model iteration to continuously and efficiently improve accuracy

Data-AI convergence, streamlining the E2E process of data services and AI development

Efficient Running

AI acceleration suites for data, training, and inference acceleration, as well as distributed efficient training and inference

Cost-effective Ascend computing power

Large-scale heterogeneous clusters and scheduling management

Efficient Migration

E2E cloud-based Ascend migration tool chain to support full-stack AI services

Professional migration service

Why Huawei Cloud ModelArts?

Stable and Secure Computing Base, Fast and Simple Model Training

Stable and Secure Computing Base, Fast and Simple Model Training

  • Support the management of 10,000-node compute clusters.

  • Large-scale distributed training accelerates foundation model development.

Zero-Code AutoML for Flexible, Simplified AI Application Development

Zero-Code AutoML for Flexible, Simplified AI Application Development

  • ExeML automates model design, parameter tuning and training, and model compression and deployment based on labeled data.

  • ExeML can be used to create image classification, object detection, and sound classification models, meeting the demands of various fields.

In-Cloud Notebook, Case Access in Seconds

In-Cloud Notebook, Case Access in Seconds

  • E2E AI development is managed in ModelArts Studio, boosting efficiency while maintaining records of the entire AI development process.

  • Local IDE and ModelArts plug-ins are provided for seamless on-premises and in-cloud AI development with customizable running environments.

Flexible Deployment for Various Scenarios

Flexible Deployment for Various Scenarios

  • Supports multiple production environments, including cloud and edge.

  • Supports multiple deployment types, including real-time inference, batch inference, and edge inference.

Widely Used AI Platform Facilitates Business Success

Widely Used AI Platform Facilitates Business Success

Foundation Model

Intelligent Q&A, chatbot, automated summarization, machine translation, and text classification

Autonomous Driving

Environment perception, path planning, and autonomous driving control


AI-assisted generation of images, copy, audio, and video


AI-powered decision-making for superior public services


Efficient, intelligent, and accurate services for financial institutions


Efficient, intelligent, secure, and sustainable production solutions for the mining industry


Intelligent train scheduling, device fault prediction, and railway security monitoring


Automated medical report interpretation, Internet diagnoses, and comprehensive health management allowing for diversified, intelligent, and lean services

AI Gallery