def test_ecs_pytorch_s3_plugin_training_gpu(gpu_only, ecs_container_instance,
pytorch_training, training_cmd,
ecs_cluster_name,
pt17_and_above_only):
"""
GPU resnet18 test for PyTorch Training using S3 plugin
Instance Type - p3.8xlarge
Given above parameters, registers a task with family named after this test, runs the task, and waits for
the task to be stopped before doing teardown operations of instance and cluster.
"""
_, image_framework_version = get_framework_and_version_from_tag(
pytorch_training)
if Version(image_framework_version) < Version("1.8"):
pytest.skip("S3 plugin is supported on PyTorch version >=1.8")
instance_id, cluster_arn = ecs_container_instance
num_gpus = ec2_utils.get_instance_num_gpus(instance_id)
ecs_utils.ecs_training_test_executor(ecs_cluster_name,
cluster_arn,
training_cmd,
pytorch_training,
instance_id,
num_gpus=num_gpus)
def test_ecs_pytorch_training_dgl_gpu(gpu_only, py3_only,
ecs_container_instance, pytorch_training,
training_cmd, ecs_cluster_name):
"""
GPU DGL test for PyTorch Training
Instance Type - p3.8xlarge
DGL is only supported in py3, hence we have used the "py3_only" fixture to ensure py2 images don't run
on this function.
Given above parameters, registers a task with family named after this test, runs the task, and waits for
the task to be stopped before doing teardown operations of instance and cluster.
"""
_, image_framework_version = get_framework_and_version_from_tag(
pytorch_training)
image_cuda_version = get_cuda_version_from_tag(pytorch_training)
if Version(image_framework_version) == Version(
"1.6") and image_cuda_version == "cu110":
pytest.skip("DGL does not suport CUDA 11 for PyTorch 1.6")
# TODO: Remove when DGL gpu test on ecs get fixed
if Version(image_framework_version) >= Version(
"1.10") and image_cuda_version == "cu113":
pytest.skip("ecs test for DGL gpu fails since pt 1.10")
instance_id, cluster_arn = ecs_container_instance
num_gpus = ec2_utils.get_instance_num_gpus(instance_id)
ecs_utils.ecs_training_test_executor(ecs_cluster_name,
cluster_arn,
training_cmd,
pytorch_training,
instance_id,
num_gpus=num_gpus)
def test_ecs_mxnet_training_gluonnlp_gpu(gpu_only, py3_only,
ecs_container_instance,
mxnet_training, training_cmd,
ecs_cluster_name):
"""
GPU Gluon NLP test for MXNet Training
Instance Type - p2.16xlarge
DGL is only supported in py3, hence we have used the "py3_only" fixture to ensure py2 images don't run
on this function.
Given above parameters, registers a task with family named after this test, runs the task, and waits for
the task to be stopped before doing teardown operations of instance and cluster.
"""
instance_id, cluster_arn = ecs_container_instance
num_gpus = ec2_utils.get_instance_num_gpus(instance_id)
ecs_utils.ecs_training_test_executor(ecs_cluster_name,
cluster_arn,
training_cmd,
mxnet_training,
instance_id,
num_gpus=num_gpus)
def test_ecs_mxnet_training_mnist_cpu(cpu_only, ecs_container_instance, mxnet_training, training_cmd, ecs_cluster_name):
"""
CPU mnist test for MXNet Training
Instance Type - c4.8xlarge
Given above parameters, registers a task with family named after this test, runs the task, and waits for
the task to be stopped before doing teardown operations of instance and cluster.
"""
instance_id, cluster_arn = ecs_container_instance
ecs_utils.ecs_training_test_executor(ecs_cluster_name, cluster_arn, training_cmd, mxnet_training, instance_id)
def test_ecs_pytorch_training_dgl_cpu(cpu_only, py3_only, ecs_container_instance, pytorch_training, training_cmd,
ecs_cluster_name):
"""
CPU DGL test for PyTorch Training
Instance Type - c5.12xlarge
DGL is only supported in py3, hence we have used the "py3_only" fixture to ensure py2 images don't run
on this function.
Given above parameters, registers a task with family named after this test, runs the task, and waits for
the task to be stopped before doing teardown operations of instance and cluster.
"""
instance_id, cluster_arn = ecs_container_instance
ecs_utils.ecs_training_test_executor(ecs_cluster_name, cluster_arn, training_cmd, pytorch_training, instance_id)
def test_ecs_pytorch_training_mnist_gpu(gpu_only, ecs_container_instance, pytorch_training, training_cmd,
ecs_cluster_name):
"""
GPU mnist test for PyTorch Training
Instance Type - p3.8xlarge
Given above parameters, registers a task with family named after this test, runs the task, and waits for
the task to be stopped before doing teardown operations of instance and cluster.
"""
instance_id, cluster_arn = ecs_container_instance
num_gpus = ec2_utils.get_instance_num_gpus(instance_id)
ecs_utils.ecs_training_test_executor(ecs_cluster_name, cluster_arn, training_cmd, pytorch_training, instance_id,
num_gpus=num_gpus)
def test_ecs_pytorch_s3_plugin_training_cpu(cpu_only, ecs_container_instance,
pytorch_training, training_cmd,
ecs_cluster_name):
"""
CPU resnet18 test for PyTorch Training using S3 plugin
Instance Type - c5.9xlarge
Given above parameters, registers a task with family named after this test, runs the task, and waits for
the task to be stopped before doing teardown operations of instance and cluster.
"""
instance_id, cluster_arn = ecs_container_instance
ecs_utils.ecs_training_test_executor(ecs_cluster_name, cluster_arn,
training_cmd, pytorch_training,
instance_id)
def test_ecs_mxnet_training_dgl_cpu(cpu_only, py3_only, ecs_container_instance, mxnet_training, training_cmd,
ecs_cluster_name):
"""
CPU DGL test for MXNet Training
Instance Type - c4.2xlarge
DGL is only supported in py3, hence we have used the "py3_only" fixture to ensure py2 images don't run
on this function.
Given above parameters, registers a task with family named after this test, runs the task, and waits for
the task to be stopped before doing teardown operations of instance and cluster.
"""
# TODO: remove/update this when DGL supports MXNet 1.9
_, framework_version = get_framework_and_version_from_tag(mxnet_training)
if Version(framework_version) >= Version('1.9.0'):
pytest.skip("Skipping DGL tests as DGL does not yet support MXNet 1.9")
instance_id, cluster_arn = ecs_container_instance
ecs_utils.ecs_training_test_executor(ecs_cluster_name, cluster_arn, training_cmd, mxnet_training, instance_id)
def test_ecs_pytorch_training_dgl_cpu(
cpu_only, py3_only, ecs_container_instance, pytorch_training, training_cmd, ecs_cluster_name
):
"""
CPU DGL test for PyTorch Training
Instance Type - c5.12xlarge
DGL is only supported in py3, hence we have used the "py3_only" fixture to ensure py2 images don't run
on this function.
Given above parameters, registers a task with family named after this test, runs the task, and waits for
the task to be stopped before doing teardown operations of instance and cluster.
"""
_, image_framework_version = get_framework_and_version_from_tag(pytorch_training)
# TODO: Remove when DGL gpu test on ecs get fixed
if Version(image_framework_version) in SpecifierSet("==1.10.*"):
pytest.skip("ecs test for DGL gpu fails for pt 1.10")
instance_id, cluster_arn = ecs_container_instance
ecs_utils.ecs_training_test_executor(ecs_cluster_name, cluster_arn, training_cmd, pytorch_training, instance_id)
