data_parallel_model Namespace Reference

Module caffe2.python.data_parallel_model. More...

Functions
def	Parallelize_GPU (model_helper_obj, input_builder_fun, forward_pass_builder_fun, param_update_builder_fun, devices=range(0, workspace.NumCudaDevices()), rendezvous=None, net_type='dag', broadcast_computed_params=True, optimize_gradient_memory=False)
def	ExtractPredictorNet (model, inputs, outputs, device)
def	FinalizeAfterCheckpoint (model, blobs, sync_iter=True)
def	stripParamName (param)

Variables
	log = logging.getLogger("data_parallel_model")

Detailed Description

Module caffe2.python.data_parallel_model.

Function Documentation

ExtractPredictorNet()

def data_parallel_model.ExtractPredictorNet	(		model,
			inputs,
			outputs,
			device
	)

Returns (net, params) that can be exported to be used as a prediction
net.

Definition at line 202 of file data_parallel_model.py.

Parallelize_GPU()

def data_parallel_model.Parallelize_GPU	(		model_helper_obj,
			input_builder_fun,
			forward_pass_builder_fun,
			param_update_builder_fun,
			devices = range(0, workspace.NumCudaDevices()),
			rendezvous = None,
			net_type = 'dag',
			broadcast_computed_params = True,
			optimize_gradient_memory = False
	)

Function to create a model that can run on many GPUs.
  model_helper_obj: an object of ModelHelperBase, such as CNNModelHelper
  input_builder_fun:
                     Function that adds the input operators
                     Note: Remember to instantiate reader outside of this
                     function so all GPUs share same reader object.
                     Signature:  input_builder_fun(model)
  forward_pass_builder_fun:
                    Function to add the operators to the model.
                    Must return list of loss-blob references that
                    are used to build the gradient. Loss scale parameter
                    is passed, as you should scale the loss of your model
                    by 1.0 / the total number of gpus.
                    Signature: forward_pass_builder_fun(model, loss_scale)
  param_update_builder_fun:
                    Function that adds operators that are run after
                    gradient update, such as updating the weights and
                    weight decaying.
                    Signature: param_update_builder_fun(model)
  devices:          List of GPU ids, such as [0, 1, 2, 3],
  rendezvous:       used for rendezvous in distributed computation, if None
                    then only one node is used. To create rendezvous,
                    use <TBD>.
  net_type:         Network type

Definition at line 32 of file data_parallel_model.py.