Task-based modeling provides a wizard-guided way to submit training tasks for model building. You can directly use the built-in images of the platform to quickly submit training tasks with mainstream high-performance and distributed training frameworks. You can also start tasks through custom training images. The detailed description of functional modules is as follows:
Creating training tasks: Create and submit a training task based on a built-in framework or custom training image. You can use platform datasets, COS data, or CFS file systems as training samples. You can also set training task algorithm parameters, enable log shipping, and bind a VPC.
Managing task operations: Provide normal management operations, including start, stop, and deletion of training tasks. It supports the task replication feature. You can quickly launch training tasks with different configurations to compare model training effects.
Monitoring task operations: Provide running logs of training, and support visual monitoring of training metrics, training resource consumptions, and training tasks started via TiKit.
TensorBoard: Start TensorBoard for monitoring tasks.