def finalize(self):
if self._distributed:
Communicator.finalize()
train_data = os.path.join(data_path, 'train_map.txt')
test_data = os.path.join(data_path, 'test_map.txt')
num_quantization_bits = args['quantized_bits']
epochs = args['epochs']
warm_up = args['distributed_after']
network_name = args['network']
scale_up = bool(args['scale_up'])
# Create distributed trainer factory
print(
"Start training: quantize_bit = {}, epochs = {}, distributed_after = {}"
.format(num_quantization_bits, epochs, warm_up))
try:
resnet_cifar10(train_data,
test_data,
mean_data,
network_name,
epoch_size,
num_quantization_bits,
block_size=args['block_samples'],
warm_up=args['distributed_after'],
max_epochs=epochs,
scale_up=scale_up,
log_to_file=args['logdir'],
profiling=args['profile'])
finally:
# Must call MPI finalize when process exit
Communicator.finalize()
if args['epoch_size'] is not None:
epoch_size = args['epoch_size']
mean_data=os.path.join(data_path, 'CIFAR-10_mean.xml')
train_data=os.path.join(data_path, 'train_map.txt')
test_data=os.path.join(data_path, 'test_map.txt')
num_quantization_bits = args['quantized_bits']
epochs = args['epochs']
warm_up = args['distributed_after']
network_name = args['network']
scale_up = bool(args['scale_up'])
# Create distributed trainer factory
print("Start training: quantize_bit = {}, epochs = {}, distributed_after = {}".format(num_quantization_bits, epochs, warm_up))
try:
resnet_cifar10(train_data, test_data, mean_data,
network_name,
epoch_size,
num_quantization_bits,
block_size=args['block_samples'],
warm_up=args['distributed_after'],
max_epochs=epochs,
scale_up=scale_up,
log_to_file=args['logdir'],
profiling=args['profile'])
finally:
# Must call MPI finalize when process exit
Communicator.finalize()