pylibwholegraph.torch.initialize.init_torch_env_and_create_wm_comm

pylibwholegraph.torch.initialize.init_torch_env_and_create_wm_comm(world_rank: int, world_size: int, local_rank: int, local_size: int, distributed_backend_type='nccl', wm_log_level='info')

Init WholeGraph environment for PyTorch and create

single communicator for all ranks.

Parameters:

• world_rank – world rank of current process

• world_size – world size of all processes

• local_rank – local rank of current process

• local_size – local size

Returns:

global and local node Communicator