import pytest
import sagemaker.utils
import tests.integ as integ
from sagemaker.pytorch import PyTorch
from tests.integ import timeout
from tests.integ.test_pytorch import _upload_training_data
smdataparallel_dir = os.path.join(os.path.dirname(__file__), "..", "data",
"smdistributed_dataparallel")
@pytest.mark.skipif(
integ.test_region() not in integ.DATA_PARALLEL_TESTING_REGIONS,
reason=
"Only allow this test to run in IAD and CMH to limit usage of p3.16xlarge",
)
def test_smdataparallel_pt_mnist(
sagemaker_session,
pytorch_training_latest_version,
pytorch_training_latest_py_version,
):
job_name = sagemaker.utils.unique_name_from_base(
"pt-sm-distributed-dataparallel")
estimator = PyTorch(
entry_point="mnist_pt.py",
role="SageMakerRole",
source_dir=smdataparallel_dir,
instance_count=2,
# language governing permissions and limitations under the License.
from __future__ import absolute_import
import os
import pytest
from sagemaker.huggingface import HuggingFace
from tests import integ
from tests.integ import DATA_DIR, TRAINING_DEFAULT_TIMEOUT_MINUTES
from tests.integ.timeout import timeout
@pytest.mark.release
@pytest.mark.skipif(
integ.test_region() in integ.TRAINING_NO_P2_REGIONS,
reason="no ml.p2 instances in this region",
)
def test_huggingface_training(
sagemaker_session,
gpu_instance_type,
huggingface_training_latest_version,
huggingface_pytorch_latest_version,
):
with timeout(minutes=TRAINING_DEFAULT_TIMEOUT_MINUTES):
data_path = os.path.join(DATA_DIR, "huggingface")
hf = HuggingFace(
py_version="py36",
entry_point="examples/text-classification/run_glue.py",
role="SageMakerRole",
entry_point="mnist.py",
framework_version=pytorch_inference_latest_version,
py_version=pytorch_inference_latest_py_version,
sagemaker_session=sagemaker_session,
)
predictor = model.deploy(1, cpu_instance_type, endpoint_name=endpoint_name)
batch_size = 100
data = numpy.random.rand(batch_size, 1, 28, 28).astype(numpy.float32)
output = predictor.predict(data)
assert output.shape == (batch_size, 10)
@pytest.mark.skipif(
test_region() not in EI_SUPPORTED_REGIONS, reason="EI isn't supported in that specific region."
)
def test_deploy_model_with_accelerator(
sagemaker_session,
cpu_instance_type,
pytorch_eia_latest_version,
pytorch_eia_latest_py_version,
):
endpoint_name = "test-pytorch-deploy-eia-{}".format(sagemaker_timestamp())
model_data = sagemaker_session.upload_data(path=EIA_MODEL)
pytorch = PyTorchModel(
model_data,
"SageMakerRole",
entry_point=EIA_SCRIPT,
framework_version=pytorch_eia_latest_version,
py_version=pytorch_eia_latest_py_version,
sagemaker_session=sagemaker_session,
)
predictor = model.deploy(1,
cpu_instance_type,
endpoint_name=endpoint_name)
batch_size = 100
data = numpy.random.rand(batch_size, 1, 28, 28).astype(numpy.float32)
output = predictor.predict(data)
assert output.shape == (batch_size, 10)
@pytest.mark.skipif(PYTHON_VERSION == "py2",
reason="PyTorch EIA does not support Python 2.")
@pytest.mark.skipif(test_region() not in EI_SUPPORTED_REGIONS,
reason="EI isn't supported in that specific region.")
def test_deploy_model_with_accelerator(sagemaker_session, cpu_instance_type):
endpoint_name = "test-pytorch-deploy-eia-{}".format(sagemaker_timestamp())
model_data = sagemaker_session.upload_data(path=EIA_MODEL)
pytorch = PyTorchModel(
model_data,
"SageMakerRole",
framework_version="1.3.1",
entry_point=EIA_SCRIPT,
sagemaker_session=sagemaker_session,
)
with timeout_and_delete_endpoint_by_name(endpoint_name, sagemaker_session):
predictor = pytorch.deploy(
initial_instance_count=1,
instance_type=cpu_instance_type,
endpoint_name = "test-mxnet-coach-deploy-{}".format(sagemaker_timestamp())
with timeout_and_delete_endpoint_by_name(endpoint_name, sagemaker_session):
predictor = estimator.deploy(
1, cpu_instance_type, entry_point="mxnet_deploy.py", endpoint_name=endpoint_name
)
observation = numpy.asarray([0, 0, 0, 0])
action = predictor.predict(observation)
assert 0 < action[0][0] < 1
assert 0 < action[0][1] < 1
@pytest.mark.skipif(
test_region() not in RL_SUPPORTED_REGIONS,
reason="Updated RL images aren't in {}".format(test_region()),
)
def test_coach_tf(sagemaker_session, coach_tensorflow_latest_version, cpu_instance_type):
estimator = _test_coach(
sagemaker_session,
RLFramework.TENSORFLOW,
coach_tensorflow_latest_version,
cpu_instance_type,
)
job_name = unique_name_from_base("test-coach-tf")
with timeout(minutes=15):
estimator.fit(job_name=job_name)
endpoint_name = "test-tf-coach-deploy-{}".format(sagemaker_timestamp())
horovod_dir = os.path.join(os.path.dirname(__file__), "..", "data", "horovod")
@pytest.fixture(scope="module")
def gpu_instance_type(request):
return "ml.p2.xlarge"
@pytest.mark.canary_quick
def test_hvd_cpu(sagemaker_session, cpu_instance_type, tmpdir):
__create_and_fit_estimator(sagemaker_session, cpu_instance_type, tmpdir)
@pytest.mark.canary_quick
@pytest.mark.skipif(integ.test_region() in integ.HOSTING_NO_P2_REGIONS,
reason="no ml.p2 instances in this region")
def test_hvd_gpu(sagemaker_session, gpu_instance_type, tmpdir):
__create_and_fit_estimator(sagemaker_session, gpu_instance_type, tmpdir)
@pytest.mark.local_mode
@pytest.mark.parametrize("instances, processes", [[1, 2], (2, 1), (2, 2)])
def test_horovod_local_mode(sagemaker_local_session, instances, processes,
tmpdir):
output_path = "file://%s" % tmpdir
job_name = sagemaker.utils.unique_name_from_base("tf-horovod")
estimator = TensorFlow(
entry_point=os.path.join(horovod_dir, "hvd_basic.py"),
role="SageMakerRole",
train_instance_count=2,
sagemaker_session=sagemaker_session,
)
train_input = mx.sagemaker_session.upload_data(
path=os.path.join(data_path, "train"), key_prefix="integ-test-data/mxnet_mnist/train"
)
test_input = mx.sagemaker_session.upload_data(
path=os.path.join(data_path, "test"), key_prefix="integ-test-data/mxnet_mnist/test"
)
mx.fit({"train": train_input, "test": test_input})
return mx.latest_training_job.name
@pytest.mark.skipif(
test_region() not in EDGE_PACKAGING_SUPPORTED_REGIONS,
reason="Edge packaging isn't supported in that specific region.",
)
def test_edge_packaging_job(mxnet_training_job, sagemaker_session):
estimator = MXNet.attach(mxnet_training_job, sagemaker_session=sagemaker_session)
model = estimator.compile_model(
target_instance_family="rasp3b",
input_shape={"data": [1, 1, 28, 28], "softmax_label": [1]},
output_path=estimator.output_path,
)
model.package_for_edge(
output_path=estimator.output_path,
role=estimator.role,
model_name="sdk-test-model",
model_version="1.0",
import sagemaker.utils
import tests.integ as integ
from sagemaker.tensorflow import TensorFlow
from tests.integ import test_region, timeout, HOSTING_NO_P3_REGIONS
horovod_dir = os.path.join(os.path.dirname(__file__), "..", "data", "horovod")
@pytest.fixture(
scope="session",
params=[
"ml.c4.xlarge",
pytest.param(
"ml.p3.2xlarge",
marks=pytest.mark.skipif(
test_region() in HOSTING_NO_P3_REGIONS,
reason="no ml.p3 instances in this region"),
),
],
)
def instance_type(request):
return request.param
@pytest.mark.canary_quick
def test_horovod(sagemaker_session, instance_type, tmpdir):
job_name = sagemaker.utils.unique_name_from_base("tf-horovod")
estimator = TensorFlow(
entry_point=os.path.join(horovod_dir, "test_hvd_basic.py"),
role="SageMakerRole",
train_instance_count=2,
import pytest
from sagemaker.huggingface import HuggingFace, TrainingCompilerConfig
from tests import integ
from tests.integ import DATA_DIR, TRAINING_DEFAULT_TIMEOUT_MINUTES
from tests.integ.timeout import timeout
@pytest.fixture(scope="module")
def gpu_instance_type(request):
return "ml.p3.2xlarge"
@pytest.mark.release
@pytest.mark.skipif(
integ.test_region() not in integ.TRAINING_COMPILER_SUPPORTED_REGIONS,
reason="SageMaker Training Compiler is not supported in this region",
)
@pytest.mark.skipif(
integ.test_region() in integ.TRAINING_NO_P3_REGIONS,
reason="no ml.p3 instances in this region",
)
def test_huggingface_pytorch(
sagemaker_session,
gpu_instance_type,
huggingface_training_compiler_latest_version,
huggingface_training_compiler_pytorch_latest_version,
):
with timeout(minutes=TRAINING_DEFAULT_TIMEOUT_MINUTES):
data_path = os.path.join(DATA_DIR, "huggingface")
sagemaker_session,
cpu_instance_type,
tmpdir,
):
_create_and_fit_estimator(
mxnet_training_latest_version,
mxnet_training_latest_py_version,
sagemaker_session,
cpu_instance_type,
tmpdir,
)
@pytest.mark.release
@pytest.mark.skipif(
integ.test_region() in integ.TRAINING_NO_P2_REGIONS
and integ.test_region() in integ.TRAINING_NO_P3_REGIONS,
reason="no ml.p2 or ml.p3 instances in this region",
)
def test_hvd_gpu(
mxnet_training_latest_version,
mxnet_training_latest_py_version,
sagemaker_session,
gpu_instance_type,
tmpdir,
):
_create_and_fit_estimator(
mxnet_training_latest_version,
mxnet_training_latest_py_version,
sagemaker_session,
gpu_instance_type,
# language governing permissions and limitations under the License.
from __future__ import absolute_import
import os
import pytest
from sagemaker.huggingface import HuggingFace
from tests import integ
from tests.integ import DATA_DIR, TRAINING_DEFAULT_TIMEOUT_MINUTES
from tests.integ.timeout import timeout
@pytest.mark.release
@pytest.mark.skipif(
integ.test_region() in integ.TRAINING_NO_P2_REGIONS, reason="no ml.p2 instances in this region"
)
def test_huggingface_training(
sagemaker_session,
gpu_instance_type,
huggingface_training_latest_version,
huggingface_pytorch_version,
):
with timeout(minutes=TRAINING_DEFAULT_TIMEOUT_MINUTES):
data_path = os.path.join(DATA_DIR, "huggingface")
hf = HuggingFace(
py_version="py36",
entry_point="examples/text-classification/run_glue.py",
role="SageMakerRole",
transformers_version=huggingface_training_latest_version,
import pytest
from sagemaker.huggingface import HuggingFace, HuggingFaceProcessor
from sagemaker.huggingface.model import HuggingFaceModel, HuggingFacePredictor
from sagemaker.utils import unique_name_from_base
from tests import integ
from tests.integ.utils import gpu_list, retry_with_instance_list
from tests.integ import DATA_DIR, TRAINING_DEFAULT_TIMEOUT_MINUTES
from tests.integ.timeout import timeout, timeout_and_delete_endpoint_by_name
ROLE = "SageMakerRole"
@pytest.mark.release
@pytest.mark.skipif(
integ.test_region() in integ.TRAINING_NO_P2_REGIONS
and integ.test_region() in integ.TRAINING_NO_P3_REGIONS,
reason="no ml.p2 or ml.p3 instances in this region",
)
@retry_with_instance_list(gpu_list(integ.test_region()))
def test_framework_processing_job_with_deps(
sagemaker_session,
huggingface_training_latest_version,
huggingface_training_pytorch_latest_version,
huggingface_pytorch_latest_training_py_version,
**kwargs,
):
with timeout(minutes=TRAINING_DEFAULT_TIMEOUT_MINUTES):
code_path = os.path.join(DATA_DIR, "dummy_code_bundle_with_reqs")
entry_point = "main_script.py"
