Introduction to Concept

Region

Region/endpoint is a physical data center partition. The following are the Cloud-ML clusters currently online:

Region | Endpoint | Chinese Name | Console Name

Org

Xiaomi Cloud-ML has implemented Org-based multi-tenant isolation. All resources belong to a fixed Org. Users can log in or create Orgs in an Eco Cloud. Each Org has Quota limits for resources.

Individual developers can use the default Org. Team developers can use the "team management" function to create Org. Use specific Org and AKSK when accessing Xiaomi Cloud-ML service.

AKSK

Xiaomi Cloud-ML service supports Eco Cloud's AKSK authentication system. When users request for services, they need to sign in with Access key and Secret key, which are used for service authentication and authorization.

Please note that each pair of AKSKs identify the Org. Each Org has standalone Quotas and operation permissions. Team developers need to use the correct AKSKs to access Xiaomi Cloud-ML service.

Quota

Xiaomi Cloud-ML service includes various resources, such as TrainJob. You can apply for CPU, memory, and GPU physical resources when creating a resource. There are Quotas for the amounts of TrainJobs, CPUs, memory, and GPU in total. Developers can apply for Quota increase from the administrator. ## Fds FDS is Xiaomi's distributed File Storage Service. Cloud-ML data storage relies on FDS. User-submitted training codes are also stored in FDS. Document Location: http://docs.api.xiaomi.com/fds/ ## TrainJob Trainjob represents a training task, similar to computational tasks in Hadoop or Spark. Users can specify Memory, CPU, and GPU limits when submitting a task. After submitting, the platform will be adjusted to run on a suitable server. Logs for the running process can be obtained through API. After the training results, the model will be stored in distributed storage.

API

Xiaomi Cloud-ML provides RESTful API interfaces externally. You can use any HTTP client to access resources, such as trainjob, model service, and dev environment. When using API, users must strictly refer to the API documentation for correct parameters usage.

Python SDK

Xiaomi Cloud-ML provides Python SDK, a Python-based HTTP client that packages the parameters for accessing the Xiaomi Cloud-ML service, which provides easy access to Xiaomi Cloud-ML service.

Command Line Tools

The cloudml command line tool is the most convenient tool for accessing Xiaomi Cloud-ML. Based on the Python SDK implementation, you can submit training jobs, deploy model service, create dev environment, etc. through commands on Mac, Linux, or Windows operating system.