Multi GPU

class olympus.distributed.multigpu.NoOut

Bases: object

Methods

flush
write

flush()

write(string)

olympus.distributed.multigpu.arguments(parser)

olympus.distributed.multigpu.data_parallel(model, device_ids=None, *args, **kwargs)

olympus.distributed.multigpu.enable_distributed_process(rank, dist_url, world_size, silence_stdout='True')

Initialize distributed system, if all required arguments are met, do nothing if not

Parameters:

rank: Optional[int]

Process rank

dist_url: str

Distributed backend uri

world_size: int

Number of processes running in parallel

silence_stdout: bool

Silence the standard output for ranks superior to 0

olympus.distributed.multigpu.rank()

olympus.distributed.multigpu.set_rank(rank_)