Beispiel #1
0
import lib

parser = argparse.ArgumentParser()
parser = lib.arg_all(parser)
args = parser.parse_args()

DIM = args.dim
norm = nn.BatchNorm1d if DIM == 1 else nn.BatchNorm2d

layer = norm(args.channels)

model = nn.Sequential(
    layer, nn.Flatten(),
    nn.Linear(in_features=args.channels * args.numf**DIM, out_features=10))

train_dataset = give(DIM, args.numf, args.channels)

if args.nodes > 1:
    model, train_loader = lib_torch.distribute(model, train_dataset,
                                               args.nodes, args.batch)
else:
    train_loader = torch.utils.data.DataLoader(dataset=train_dataset,
                                               batch_size=args.batch,
                                               shuffle=True)

time = lib_torch.profile(['batch_norm', 'NativeBatchNormBackward'], model,
                         train_loader, args.epochs)

import numpy as np

data = np.array([[
parser = argparse.ArgumentParser()

parser.add_argument('-numf', type=int, required=True)
parser.add_argument('-batch', type=int, required=True)
parser.add_argument('-nodes', type=int, required=True)
parser.add_argument('-epochs', type=int, required=True)

parser.add_argument('-units', type=int, required=True)
args = parser.parse_args()

layer = nn.Linear(in_features=args.numf, out_features=args.units)

model = nn.Sequential(
    nn.Flatten(), nn.Linear(in_features=args.numf, out_features=args.units))

train_dataset = give(1, args.numf, 1, out_size=args.units)

if args.nodes > 1:
    model, train_loader = lib_torch.distribute(model, train_dataset,
                                               args.nodes, args.batch)
else:
    train_loader = torch.utils.data.DataLoader(dataset=train_dataset,
                                               batch_size=args.batch,
                                               shuffle=True)

time = lib_torch.profile(['addmm', 'AddmmBackward'], model, train_loader,
                         args.epochs)

import numpy as np

data = np.array([[
Beispiel #3
0
layer = nn.Linear(
    in_features = args.numf,
    out_features = args.units
)

model = nn.Sequential(
    nn.Flatten(),
    layer,
    nn.Flatten(),
    nn.Linear(
        in_features = args.units,
        out_features = 10
    )
)

train_dataset = give(1, args.numf, 1)

if args.nodes > 1:
    model, train_loader = lib_torch.distribute(model, train_dataset, args.nodes, args.batch)
else:
    train_loader = torch.utils.data.DataLoader(
        dataset = train_dataset,
        batch_size = args.batch,
        shuffle = True
    )

time = lib_torch.profile(['addmm', 'AddmmBackward'], 
                         model, train_loader, args.epochs)

import numpy as np