CM_WORKER_SIZE

Description

In TensorFlow distributed training or inference scenarios, you can choose not to use the rank table file. By jointly configuring environment variables CM_CHIEF_IP, CM_CHIEF_PORT, CM_CHIEF_DEVICE, CM_WORKER_SIZE, and CM_WORKER_IP, resource information can be automatically generated to complete the initialization of the collective communication component.

Configures the number of devices in the service communicator.

The value of this environment variable must be an integer ranging from 0 to 32768.

Example

export CM_WORKER_SIZE=8

Constraints

This environment variable cannot be used together with RANK_TABLE_FILE, RANK_ID, or RANK_SIZE.

Applicability

Atlas training product

Atlas A2 training product/Atlas A2 inference product