Convert subprocess test to standalone test (#14101)

Author: awaelchli

tests/tests_pytorch/run_standalone_tasks.sh

@@ -34,6 +34,10 @@ fi
 # test that a user can manually launch individual processes
 echo "Running manual ddp launch test"
 export PYTHONPATH="${PYTHONPATH}:$(pwd)"
-args="--trainer.accelerator gpu --trainer.devices 2 --trainer.strategy ddp --trainer.max_epochs=1 --trainer.limit_train_batches=1 --trainer.limit_val_batches=1 --trainer.limit_test_batches=1"
-MASTER_ADDR="localhost" MASTER_PORT=1234 LOCAL_RANK=1 python ../../examples/convert_from_pt_to_pl/image_classifier_5_lightning_datamodule.py ${args} &
-MASTER_ADDR="localhost" MASTER_PORT=1234 LOCAL_RANK=0 python ../../examples/convert_from_pt_to_pl/image_classifier_5_lightning_datamodule.py ${args}
+args="fit --trainer.accelerator gpu --trainer.devices 2 --trainer.strategy ddp --trainer.max_epochs=1 --trainer.limit_train_batches=1 --trainer.limit_val_batches=1 --trainer.limit_test_batches=1"
+MASTER_ADDR="localhost" MASTER_PORT=1234 LOCAL_RANK=1 python strategies/scripts/cli_script.py ${args} &
+MASTER_ADDR="localhost" MASTER_PORT=1234 LOCAL_RANK=0 python strategies/scripts/cli_script.py ${args}
+
+# test that ddp can launched as a module (-m option)
+echo "Running ddp example as module"
+python -m strategies.scripts.cli_script ${args}

tests/tests_pytorch/serve/__init__.py

@@ -0,0 +1,24 @@
+# Copyright The PyTorch Lightning team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""A trivial script that wraps a LightningCLI around the BoringModel and BoringDataModule."""
+from pytorch_lightning.cli import LightningCLI
+from pytorch_lightning.demos.boring_classes import BoringDataModule, BoringModel
+
+if __name__ == "__main__":
+    LightningCLI(
+        BoringModel,
+        BoringDataModule,
+        seed_everything_default=42,
+        save_config_overwrite=True,
+    )

tests/tests_pytorch/strategies/ddp_model.py

@@ -21,60 +21,41 @@
 from torch.nn.parallel.distributed import DistributedDataParallel
 
 import pytorch_lightning as pl
-from pytorch_lightning import Trainer
+from pytorch_lightning import seed_everything, Trainer
 from pytorch_lightning.callbacks import Callback
 from pytorch_lightning.demos.boring_classes import BoringModel
 from pytorch_lightning.strategies import DDPStrategy
+from tests_pytorch.helpers.datamodules import ClassifDataModule
 from tests_pytorch.helpers.runif import RunIf
-from tests_pytorch.strategies import ddp_model
-from tests_pytorch.utilities.distributed import call_training_script
+from tests_pytorch.helpers.simple_models import ClassificationModel
 
-CLI_ARGS = "--max_epochs 1 --accelerator gpu --devices 2 --strategy ddp"
 
+@RunIf(min_cuda_gpus=2, standalone=True)
+def test_multi_gpu_model_ddp_fit_only(tmpdir):
+    dm = ClassifDataModule()
+    model = ClassificationModel()
+    trainer = Trainer(default_root_dir=tmpdir, max_epochs=1, accelerator="gpu", devices=2, strategy="ddp")
+    trainer.fit(model, datamodule=dm)
 
-@RunIf(min_cuda_gpus=2)
-@pytest.mark.parametrize("as_module", [True, False])
-def test_multi_gpu_model_ddp_fit_only(tmpdir, as_module):
-    # call the script
-    call_training_script(ddp_model, CLI_ARGS, "fit", tmpdir, timeout=120, as_module=as_module)
 
-    # load the results of the script
-    result_path = os.path.join(tmpdir, "ddp.result")
-    result = torch.load(result_path)
+@RunIf(min_cuda_gpus=2, standalone=True)
+def test_multi_gpu_model_ddp_test_only(tmpdir):
+    dm = ClassifDataModule()
+    model = ClassificationModel()
+    trainer = Trainer(default_root_dir=tmpdir, max_epochs=1, accelerator="gpu", devices=2, strategy="ddp")
+    trainer.test(model, datamodule=dm)
 
-    # verify the file wrote the expected outputs
-    assert result["status"] == "complete"
 
+@RunIf(min_cuda_gpus=2, standalone=True)
+def test_multi_gpu_model_ddp_fit_test(tmpdir):
+    seed_everything(4321)
+    dm = ClassifDataModule()
+    model = ClassificationModel()
+    trainer = Trainer(default_root_dir=tmpdir, max_epochs=1, accelerator="gpu", devices=2, strategy="ddp")
+    trainer.fit(model, datamodule=dm)
+    result = trainer.test(model, datamodule=dm)
 
-@RunIf(min_cuda_gpus=2)
-@pytest.mark.parametrize("as_module", [True, False])
-def test_multi_gpu_model_ddp_test_only(tmpdir, as_module):
-    # call the script
-    call_training_script(ddp_model, CLI_ARGS, "test", tmpdir, as_module=as_module)
-
-    # load the results of the script
-    result_path = os.path.join(tmpdir, "ddp.result")
-    result = torch.load(result_path)
-
-    # verify the file wrote the expected outputs
-    assert result["status"] == "complete"
-
-
-@RunIf(min_cuda_gpus=2)
-@pytest.mark.parametrize("as_module", [True, False])
-def test_multi_gpu_model_ddp_fit_test(tmpdir, as_module):
-    # call the script
-    call_training_script(ddp_model, CLI_ARGS, "fit_test", tmpdir, timeout=20, as_module=as_module)
-
-    # load the results of the script
-    result_path = os.path.join(tmpdir, "ddp.result")
-    result = torch.load(result_path)
-
-    # verify the file wrote the expected outputs
-    assert result["status"] == "complete"
-
-    model_outs = result["result"]
-    for out in model_outs:
+    for out in result:
         assert out["test_acc"] > 0.7