def test_initialization(tiny_model_2d):
config = tiny_model_2d["config"]
in_channels = config["input_channels"]
model = TinyConvNet2d(in_channels=in_channels)
ip = InferenceProcess(config=config, model=model)
ip.shutdown()
def test_training(tiny_model_2d):
config = tiny_model_2d["config"]
config["num_iterations_per_update"] = 10
in_channels = config["input_channels"]
model = TinyConvNet2d(in_channels=in_channels)
training = TrainingProcess(config=config, model=model)
try:
training.set_devices([torch.device("cpu")])
data = TikTensorBatch([
TikTensor(torch.zeros(in_channels, 15, 15), ((1, ), (1, ))),
TikTensor(torch.ones(in_channels, 9, 9), ((2, ), (2, ))),
])
labels = TikTensorBatch([
TikTensor(torch.ones(in_channels, 15, 15, dtype=torch.uint8),
((1, ), (1, ))),
TikTensor(torch.full((in_channels, 9, 9), 2, dtype=torch.uint8),
((2, ), (2, ))),
])
training.update_dataset("training", data, labels)
training.resume_training()
import time
time.sleep(10)
finally:
training.shutdown()
def test_training_in_proc(tiny_model_2d, log_queue):
config = tiny_model_2d["config"]
config["num_iterations_per_update"] = 10
in_channels = config["input_channels"]
model = TinyConvNet2d(in_channels=in_channels)
handler_conn, training_conn = mp.Pipe()
p = mp.Process(target=run,
kwargs={
"conn": training_conn,
"model": model,
"config": config,
"log_queue": log_queue
})
p.start()
client = create_client(ITraining, handler_conn)
try:
client.set_devices([torch.device("cpu")])
data = TikTensorBatch([
TikTensor(torch.zeros(in_channels, 15, 15), ((1, ), (1, ))),
TikTensor(torch.ones(in_channels, 9, 9), ((2, ), (2, ))),
])
labels = TikTensorBatch([
TikTensor(torch.ones(in_channels, 15, 15, dtype=torch.uint8),
((1, ), (1, ))),
TikTensor(torch.full((in_channels, 9, 9), 2, dtype=torch.uint8),
((2, ), (2, ))),
])
client.update_dataset("training", data, labels)
client.resume_training()
finally:
client.shutdown()
def test_initialization(tiny_model_2d):
config = tiny_model_2d["config"]
in_channels = config["input_channels"]
model = TinyConvNet2d(in_channels=in_channels)
dr = DryRunProcess(config=config, model=model)
dr.shutdown()
def test_set_devices(tiny_model_2d):
config = tiny_model_2d["config"]
in_channels = config["input_channels"]
model = TinyConvNet2d(in_channels=in_channels)
tp = TrainingProcess(config=config, model=model)
tp.set_devices([torch.device("cpu")])
tp.shutdown()
def test_inference2d(tiny_model_2d):
config = tiny_model_2d["config"]
in_channels = config["input_channels"]
model = TinyConvNet2d(in_channels=in_channels)
inference = InferenceProcess(config=config, model=model)
inference.set_devices([torch.device("cpu")])
data = TikTensor(torch.zeros(in_channels, 15, 15), (0, ))
pred = inference.forward(data)
assert isinstance(pred.result(timeout=10), TikTensor)
inference.shutdown()
def test_minimal_device_test(tiny_model_2d):
config = tiny_model_2d["config"]
in_channels = config["input_channels"]
model = TinyConvNet2d(in_channels=in_channels)
dr = DryRunProcess(config=config, model=model)
try:
assert dr.minimal_device_test([torch.device("cpu")])
finally:
dr.shutdown()
def test_given_training_shape(tiny_model_2d):
config = tiny_model_2d["config"]
config[TRAINING][TRAINING_SHAPE] = (1, 15, 15)
config[TRAINING][TRAINING_SHAPE_UPPER_BOUND] = (1, 15, 15)
in_channels = config["input_channels"]
model = TinyConvNet2d(in_channels=in_channels)
dr = DryRunProcess(config=config, model=model)
try:
dr.dry_run(devices=[torch.device("cpu")]).result(timeout=30)
finally:
dr.shutdown()
def test_malicious_training_shape(tiny_model_2d):
config = tiny_model_2d["config"]
config[TRAINING].update({
TRAINING_SHAPE: (1, 0, 20),
TRAINING_SHAPE_UPPER_BOUND: (1, 2, 2)
})
in_channels = config["input_channels"]
model = TinyConvNet2d(in_channels=in_channels)
dr = DryRunProcess(config=config, model=model)
try:
fut = dr.dry_run(devices=[torch.device("cpu")])
assert isinstance(fut.exception(timeout=20), ValueError)
finally:
dr.shutdown()
def test_inference2d_in_proc(tiny_model_2d, log_queue):
config = tiny_model_2d["config"]
in_channels = config["input_channels"]
model = TinyConvNet2d(in_channels=in_channels)
handler_conn, inference_conn = mp.Pipe()
p = mp.Process(target=run,
kwargs={
"conn": inference_conn,
"model": model,
"config": config,
"log_queue": log_queue
})
p.start()
client = create_client(IInference, handler_conn)
try:
client.set_devices([torch.device("cpu")])
data = TikTensor(torch.zeros(in_channels, 15, 15), (0, ))
f = client.forward(data)
f.result(timeout=10)
finally:
client.shutdown()