rlightning.utils.distributed.initialize

Distributed environment initialization utilities.

This module provides functions for initializing PyTorch distributed training environments using the CommContext singleton.

rlightning.utils.distributed.initialize.initialize_distributed_env(rank: int, world_size: int, backend: str = 'nccl', dist_url: str = 'env://', timeout: int = 1800) → None

Initialize the PyTorch distributed environment.

This function wraps CommContext.init_distributed_env to provide a convenient interface for initializing distributed training.

Parameters:

• rank – Global rank of this process.

• world_size – Total number of processes in the distributed group.

• backend – Communication backend (‘nccl’ or ‘gloo’). Defaults to ‘nccl’.

• dist_url – URL for distributed initialization. Defaults to ‘env://’.

• timeout – Timeout in seconds for initialization. Defaults to 1800.