def test_inspect(simple_model):
bs = 2 # default batch size
r = inspect(simple_model, (1, 32, 32))
expected = [
L('Conv2d-1', [bs, 1, 32, 32], [bs, 6, 30, 30], 60, 0),
L('Conv2d-2', [bs, 6, 15, 15], [bs, 16, 13, 13], 880, 0),
L('Linear-3', [bs, 576], [bs, 120], 69240, 0),
L('Linear-4', [bs, 120], [bs, 84], 10164, 0),
L('Linear-5', [bs, 84], [bs, 10], 850, 0),
]
assert r == expected
bsize = 10
r = inspect(
simple_model,
(1, 32, 32),
input_dtype=torch.FloatTensor,
input_initializer=torch.zeros,
batch_size=bsize,
)
expected = [
L('Conv2d-1', [bsize, 1, 32, 32], [bsize, 6, 30, 30], 60, 0),
L('Conv2d-2', [bsize, 6, 15, 15], [bsize, 16, 13, 13], 880, 0),
L('Linear-3', [bsize, 576], [bsize, 120], 69240, 0),
L('Linear-4', [bsize, 120], [bsize, 84], 10164, 0),
L('Linear-5', [bsize, 84], [bsize, 10], 850, 0),
]
assert r == expected
def main():
net = SimpleNet()
batch_size = 100
ti.summary(net, (1, 32, 32), batch_size=batch_size)
info = ti.inspect(net, (1, 32, 32), batch_size=batch_size)
pprint.pprint(info)
def test_rnn(rnn):
bs = 12
r = inspect(rnn, [(6, 3)], batch_size=bs, input_initializer=torch.zeros)
expected = [
L('RNN-1', [bs, 6, 3], [[bs, 6, 5], [3, bs, 5]], 170, 0),
L('Linear-2', [6 * bs, 5], [6 * bs, 1], 6, 0),
]
assert r == expected
def test_inspect_multi_input(multi_input_net):
bs = 10
r = inspect(multi_input_net, [(1, 16, 16), (1, 28, 28)], batch_size=bs)
expected = [
L('Conv2d-1', [bs, 1, 16, 16], [bs, 1, 16, 16], 10, 0),
L('ReLU-2', [bs, 1, 16, 16], [bs, 1, 16, 16], 0, 0),
L('Conv2d-3', [bs, 1, 28, 28], [bs, 1, 28, 28], 10, 0),
L('ReLU-4', [bs, 1, 28, 28], [bs, 1, 28, 28], 0, 0),
]
assert r == expected
def test_simpleconv(simpleconv):
bs = 2
r = inspect(simpleconv, [(1, 16, 16), (1, 28, 28)], batch_size=bs)
expected = [
L('Conv2d-1', [bs, 1, 16, 16], [bs, 1, 16, 16], 10, 0),
L('ReLU-2', [bs, 1, 16, 16], [bs, 1, 16, 16], 0, 0),
L('Conv2d-3', [bs, 1, 28, 28], [bs, 1, 28, 28], 10, 0),
L('ReLU-4', [bs, 1, 28, 28], [bs, 1, 28, 28], 0, 0),
]
assert r == expected
def test_lstm_model(lstm_model):
bs = 10
r = inspect(lstm_model, [(1, 28)],
batch_size=bs,
input_initializer=torch.zeros)
out = [[10, 1, 100], [[1, 10, 100], [1, 10, 100]]]
expected = [
L('LSTM-1', [10, 1, 28], out, 52000, 0),
L('Linear-2', [10, 100], [10, 10], 1010, 0),
]
assert r == expected
def test_lstm_tagger_with_embedding(lstm_tagger):
bs = 10
r = inspect(
lstm_tagger,
[(1, 1)],
batch_size=bs,
input_initializer=torch.zeros,
input_dtype=torch.LongTensor,
)
expected = [
L('Embedding-1', [bs, 1, 1], [bs, 1, 1, 6], 30, 0),
L('LSTM-2', [bs, 1, 6], [[bs, 1, 6], [[1, 1, 6], [1, 1, 6]]], 336, 0),
L('Linear-3', [bs, 6], [bs, 3], 21, 0),
]
assert r == expected
def test_autoencoder(autoencoder):
bs = 10
r = inspect(autoencoder, [(3, 32, 32)], batch_size=bs)
expected = [
L('Conv2d-1', [bs, 3, 32, 32], [bs, 6, 28, 28], 456, 0),
L('ReLU-2', [bs, 6, 28, 28], [bs, 6, 28, 28], 0, 0),
L('Conv2d-3', [bs, 6, 28, 28], [bs, 16, 24, 24], 2416, 0),
L('ReLU-4', [bs, 16, 24, 24], [bs, 16, 24, 24], 0, 0),
L('ConvTranspose2d-5', [bs, 16, 24, 24], [bs, 6, 28, 28], 2406, 0),
L('ReLU-6', [bs, 6, 28, 28], [bs, 6, 28, 28], 0, 0),
L('ConvTranspose2d-7', [bs, 6, 28, 28], [bs, 3, 32, 32], 453, 0),
L('ReLU-8', [bs, 3, 32, 32], [bs, 3, 32, 32], 0, 0),
L('Sigmoid-9', [bs, 3, 32, 32], [bs, 3, 32, 32], 0, 0),
]
assert r == expected
def test_inspect_net_with_batch_norm(netbatchnorm):
bs = 10
r = inspect(netbatchnorm, (20, ), batch_size=bs)
expected = [
L('Linear-1', [bs, 20], [bs, 15], 300, 0),
L('BatchNorm1d-2', [bs, 15], [bs, 15], 30, 30),
L('Linear-3', [bs, 15], [bs, 15], 225, 0),
L('BatchNorm1d-4', [bs, 15], [bs, 15], 30, 30),
L('Linear-5', [bs, 15], [bs, 1], 16, 0),
]
assert r == expected
with io.StringIO() as buf:
network_info = summary(netbatchnorm, (20, ), file=buf, batch_size=bs)
expected_info = NetworkInfo(661, 601, 80 * bs, 488 * bs, 2644, 8324)
assert expected_info == network_info
def test_multi_input_net2(multi_input_net2):
bs = 10
r = inspect(multi_input_net2, [(3, 128, 1024), (4, )], batch_size=bs)
expected = [
L('Conv2d-1', [10, 3, 128, 1024], [10, 64, 128, 1024], 1792, 0),
L('MaxPool2d-2', [10, 64, 128, 1024], [10, 64, 64, 512], 0, 0),
L('Conv2d-3', [10, 64, 64, 512], [10, 96, 64, 512], 55392, 0),
L('MaxPool2d-4', [10, 96, 64, 512], [10, 96, 32, 256], 0, 0),
L('Conv2d-5', [10, 96, 32, 256], [10, 128, 32, 256], 110720, 0),
L('MaxPool2d-6', [10, 128, 32, 256], [10, 128, 16, 128], 0, 0),
L('Conv2d-7', [10, 128, 16, 128], [10, 192, 16, 128], 221376, 0),
L('AdaptiveAvgPool2d-8', [10, 192, 16, 128], [10, 192, 1, 1], 0, 0),
L('Linear-9', [10, 192], [10, 64], 12352, 0),
L('Linear-10', [10, 64], [10, 4], 260, 0),
]
assert r == expected
expected_info = NetworkInfo(401892, 401892, 62914560, 1289769280, 1607568,
1354291408)
with io.StringIO() as buf:
net_info = summary(multi_input_net2, [(3, 128, 1024), (4, )],
batch_size=bs,
file=buf)
assert net_info == expected_info