Overview

HCCL provides two methods for configuring cluster information: using the rank table configuration file and using environment variables. You can select either method as required. However, the two methods cannot be used together.

You can use the rank table file to configure the NPU resources involved in collective communication. The rank table file can be used to configure cluster information in the following scenarios:

Initializing the communicator through the C API HcclCommInitClusterInfo or HcclCommInitClusterInfoConfig
Initializing the TensorFlow distributed network communicator

Configuring resource information using environment variables applies only to the communicator initialization with the TensorFlow network. Only the following products are supported:

Atlas A2 training products / Atlas A2 inference products
Atlas training products

Parent topic: Cluster Information Configuration