DDPSpawnPlugin

class pytorch_lightning.plugins.training_type.DDPSpawnPlugin(parallel_devices=None, num_nodes=None, cluster_environment=None, checkpoint_io=None, sync_batchnorm=None, ddp_comm_state=None, ddp_comm_hook=None, ddp_comm_wrapper=None, **kwargs)

Bases: pytorch_lightning.plugins.training_type.parallel.ParallelPlugin

Spawns processes using the torch.multiprocessing.spawn() method and joins processes after training finishes.

add_to_queue(trainer, queue)

Appends the trainer.callback_metrics dictionary to the given queue. To avoid issues with memory sharing, we cast the data to numpy.

Parameters

queue (SimpleQueue) – the instance of the queue to append the data.

Return type

None

barrier(*args, **kwargs)

Synchronizes all processes which blocks processes until the whole group enters this function.

Parameters

name – an optional name to pass into barrier.

Return type

None

broadcast(obj, src=0)

Broadcasts an object to all processes.

Parameters

• obj (object) – the object to broadcast

• src (int) – source rank

Return type

object

get_from_queue(trainer, queue)

Retrieve the trainer.callback_metrics dictionary from the given queue. To preserve consistency, we cast back the data to torch.Tensor.

Parameters

queue (SimpleQueue) – the instance of the queue from where to get the data.

Return type

None

model_to_device()

Moves the model to the correct device.

post_dispatch(trainer)

Hook to do something after the training/evaluation/prediction finishes.

pre_backward(closure_loss)

Run before precision plugin executes backward.

Return type

None

reduce(tensor, group=None, reduce_op='mean')

Reduces a tensor from several distributed processes to one aggregated tensor.

Parameters

• tensor – the tensor to sync and reduce

• group (Optional[Any]) – the process group to gather results from. Defaults to all processes (world)

• reduce_op (Union[ReduceOp, str]) – the reduction operation. Defaults to ‘mean’/’avg’. Can also be a string ‘sum’ to calculate the sum during reduction.

Return type

Tensor

Returns

reduced value, except when the input was not a tensor the output remains is unchanged

setup()

Called by the accelerator to finish setup.

Return type

None

spawn(function, *args, return_result=True, **kwargs)

Spawn processes that run the given function.

Parameters

• function (Callable) – The function to spawn processes from.

• *args – Optional positional arguments that will be passed to the function in addition to the process index. These arguments must be pickleable.

• return_result (bool) – If True, copies the output of the function from process 0 to the main process and returns it.

• **kwargs – Optional named arguments that will be passed to the function in addition to the process index. These arguments must be pickleable.

Return type

Optional[Any]

Returns

The output of the function of process 0.

teardown()

This method is called to teardown the training process.

It is the right place to release memory and free other resources.

Return type

None

property root_device

Return the root device.