ModelArts is a central development platform for AI developers. With distributed training, automated model building, and model deployment, ModelArts helps AI developers quickly create models and manage the AI development lifecycle.
This new major version upgrade includes the following new features and bug fixes.
New Features:
- Training jobs support data turbo, improving the data read speed.
- The new volume type for the training job is pfs
- The training Console supports configuring the running user id.
- The inference service supports mounting the Object Block Storage (OBS) parallel file system.
- Added a limit function for the EVS storage size of instances created by notebooks. The default maximum is 500 GB.
- Training jobs support fault restart recovery capabilities
- Inference instance resource monitoring increases NPU usage, memory usage, and GPU memory usage indicators.
- Training jobs support specifying and configuring Moxing versions through environment variables to improve data loading efficiency.
- Users can specify the cluster scale when creating a resource pool.
- Users specifies the AZ where the CCE master node resides.
Fixed Features:
- Fixed the problem that algorithm constraints do not take effect when using dedicated resource pools.
- Fixed the problem that the health check time set by the inference online service does not take effect.
- When Jupyter Notebook mounts OBS and selects the /home/ directory, instance creation will fail. Add verification interception.
- Fixed the problem that the renewal operation cannot be performed within the grace period of the resource pool package cycle.
For more information, please refer to the ModelArts area of Help Center.