Provides multiple parallel capabilities for foundation model training, as well as easy-to-use APIs for configuring distributed policies for foundation models, helping developers quickly implement distributed training of high-performance foundation models.

Supports AI+HPC full-process programmability and microprogrammable functions, meeting the requirements for flexible programming and heterogeneous parallel acceleration in the AI for Science (AI4S) scenario.

Provides unified dynamic and static programming, best matches Ascend AI Processor, and maximizes hardware capabilities, helping developers shorten training time and improve inference performance.

Supports quick deployment on the cloud, edge, and mobile devices, enabling excellent resource utilization and privacy protection and allowing developers to focus on the creation of AI apps.