We recommend you use a at least 3 nodes cluster, each node having at least 2C 4G. And if you want to running GPU jobs, you need some GPU nodes.
The node count in the number of worker, for distribute training job, such as ps-worker or MPI/horovod they may need extra launcher node or ps node, they may also ocuppy one node. We will optimise this in following version.
Azure Arc-enabled ML compute supports Azure blob container via AML datastore/dataset, and support other storage types via Kubernetes native PV/PVC. Refer to the PV/PVC guidance.
Firstly, make sure you have switched the active cloud to AzureChinaCloud with az cloud set command. Then you can use the SDK and CLI sample in this repo.