ModelArts
ModelArts
ModelArts is a one-stop AI platform that empowers developers and data scientists to rapidly build and deploy models, accelerating intelligent industry upgrades.
Key Highlights
Proven Expertise & Capabilities
E2E Model Development Pipeline
End-to-end tool chain boosts development efficiency by 50% and fosters collaboration in DataOps, MLOps, and DevOps.
Ultra Large-Scale Training
A single job can train a model with a trillion parameters and process hundreds of petabytes of data.
Cost-effective AI Compute
Diverse compute with various specifications powers large-scale distributed training and inference acceleration.
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.
Why Us
Why Huawei
Cloud ModelArts?
- Support the management of 10,000-node compute clusters.
- Large-scale distributed training accelerates foundation model 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.
- 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.
- Supports multiple deployment types, including real-time inference, batch inference, and edge inference.
- Supports multiple deployment types, including real-time inference, batch inference, and edge inference.
Application Scenarios
Architecture Overview :
A Deep Dive Into ModelArts
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
