Merge branch 'master' into serve

README.md

@@ -146,7 +146,7 @@ class LitAutoEncoder(L.LightningModule):
 
     def training_step(self, batch, batch_idx):
         # training_step defines the train loop. It is independent of forward
-        x, y = batch
+        x, _ = batch
         x = x.view(x.size(0), -1)
         z = self.encoder(x)
         x_hat = self.decoder(z)

docs/source-app/code_samples/convert_pl_to_app/train.py

@@ -25,7 +25,7 @@ def forward(self, x):
     def training_step(self, batch, batch_idx):
         # training_step defines the train loop.
         # It is independent of forward
-        x, y = batch
+        x, _ = batch
         x = x.view(x.size(0), -1)
         z = self.encoder(x)
         x_hat = self.decoder(z)

docs/source-pytorch/common/evaluation_basic.rst

@@ -45,7 +45,7 @@ To add a test loop, implement the **test_step** method of the LightningModule
 
         def test_step(self, batch, batch_idx):
             # this is the test loop
-            x, y = batch
+            x, _ = batch
             x = x.view(x.size(0), -1)
             z = self.encoder(x)
             x_hat = self.decoder(z)
@@ -105,7 +105,7 @@ To add a validation loop, implement the **validation_step** method of the Lightn
 
         def validation_step(self, batch, batch_idx):
             # this is the validation loop
-            x, y = batch
+            x, _ = batch
             x = x.view(x.size(0), -1)
             z = self.encoder(x)
             x_hat = self.decoder(z)

docs/source-pytorch/common/notebooks.rst

@@ -57,7 +57,7 @@ Paste the following code block into a notebook cell:
             self.decoder = decoder
 
         def training_step(self, batch, batch_idx):
-            x, y = batch
+            x, _ = batch
             x = x.view(x.size(0), -1)
             z = self.encoder(x)
             x_hat = self.decoder(z)

docs/source-pytorch/model/train_model_basic.rst

@@ -68,7 +68,7 @@ The LightningModule is the full **recipe** that defines how your nn.Modules inte
 
         def training_step(self, batch, batch_idx):
             # training_step defines the train loop.
-            x, y = batch
+            x, _ = batch
             x = x.view(x.size(0), -1)
             z = self.encoder(x)
             x_hat = self.decoder(z)

docs/source-pytorch/starter/introduction.rst

@@ -129,7 +129,7 @@ A LightningModule enables your PyTorch nn.Module to play together in complex way
         def training_step(self, batch, batch_idx):
             # training_step defines the train loop.
             # it is independent of forward
-            x, y = batch
+            x, _ = batch
             x = x.view(x.size(0), -1)
             z = self.encoder(x)
             x_hat = self.decoder(z)

src/lightning/app/cli/react-ui-template/ui/yarn.lock

@@ -918,9 +918,9 @@ find-root@^1.1.0:
   integrity sha512-NKfW6bec6GfKc0SGx1e07QZY9PE99u0Bft/0rzSD5k3sO/vwkVUpDUKVm5Gpp5Ue3YfShPFTX2070tDs5kB9Ng==
 
 follow-redirects@^1.15.0:
-  version "1.15.4"
-  resolved "https://registry.yarnpkg.com/follow-redirects/-/follow-redirects-1.15.4.tgz#cdc7d308bf6493126b17ea2191ea0ccf3e535adf"
-  integrity sha512-Cr4D/5wlrb0z9dgERpUL3LrmPKVDsETIJhaCMeDfuFYcqa5bldGV6wBsAN6X/vxlXQtFBMrXdXxdL8CbDTGniw==
+  version "1.15.6"
+  resolved "https://registry.yarnpkg.com/follow-redirects/-/follow-redirects-1.15.6.tgz#7f815c0cda4249c74ff09e95ef97c23b5fd0399b"
+  integrity sha512-wWN62YITEaOpSK584EZXJafH1AGpO8RVgElfkuXbTOrPX4fIfOyEpW/CsiNd8JdYrAoOvafRTOEnvsO++qCqFA==
 
 form-data@^4.0.0:
   version "4.0.0"

src/pytorch_lightning/README.md

@@ -174,7 +174,7 @@ class LitAutoEncoder(pl.LightningModule):
 
     def training_step(self, batch, batch_idx):
         # training_step defines the train loop. It is independent of forward
-        x, y = batch
+        x, _ = batch
         x = x.view(x.size(0), -1)
         z = self.encoder(x)
         x_hat = self.decoder(z)

tests/tests_pytorch/accelerators/test_cpu.py

@@ -29,7 +29,7 @@ def test_availability():
 
 
 @RunIf(psutil=True)
-def test_get_device_stats(tmpdir):
+def test_get_device_stats(tmp_path):
     gpu_stats = CPUAccelerator().get_device_stats(Mock())
     fields = ["cpu_vm_percent", "cpu_percent", "cpu_swap_percent"]
 
@@ -38,7 +38,7 @@ def test_get_device_stats(tmpdir):
 
 
 @pytest.mark.parametrize("restore_after_pre_setup", [True, False])
-def test_restore_checkpoint_after_pre_setup(tmpdir, restore_after_pre_setup):
+def test_restore_checkpoint_after_pre_setup(tmp_path, restore_after_pre_setup):
     """Test to ensure that if restore_checkpoint_after_setup is True, then we only load the state after pre- dispatch
     is called."""
 
@@ -58,10 +58,10 @@ def load_checkpoint(self, checkpoint_path: Union[str, Path]) -> Dict[str, Any]:
             return super().load_checkpoint(checkpoint_path)
 
     model = BoringModel()
-    trainer = Trainer(default_root_dir=tmpdir, fast_dev_run=True)
+    trainer = Trainer(default_root_dir=tmp_path, fast_dev_run=True)
     trainer.fit(model)
 
-    checkpoint_path = os.path.join(tmpdir, "model.pt")
+    checkpoint_path = os.path.join(tmp_path, "model.pt")
     trainer.save_checkpoint(checkpoint_path)
 
     plugin = TestPlugin(
@@ -72,7 +72,7 @@ def load_checkpoint(self, checkpoint_path: Union[str, Path]) -> Dict[str, Any]:
     )
     assert plugin.restore_checkpoint_after_setup == restore_after_pre_setup
 
-    trainer = Trainer(default_root_dir=tmpdir, strategy=plugin, fast_dev_run=True)
+    trainer = Trainer(default_root_dir=tmp_path, strategy=plugin, fast_dev_run=True)
     trainer.fit(model, ckpt_path=checkpoint_path)
     for func in (trainer.test, trainer.validate, trainer.predict):
         plugin.setup_called = False

tests/tests_pytorch/accelerators/test_gpu.py

@@ -45,10 +45,10 @@ def test_get_nvidia_gpu_stats():
 
 @RunIf(min_cuda_gpus=1)
 @mock.patch("torch.cuda.set_device")
-def test_set_cuda_device(set_device_mock, tmpdir):
+def test_set_cuda_device(set_device_mock, tmp_path):
     model = BoringModel()
     trainer = Trainer(
-        default_root_dir=tmpdir,
+        default_root_dir=tmp_path,
         fast_dev_run=True,
         accelerator="gpu",
         devices=1,

tests/tests_pytorch/accelerators/test_mps.py

@@ -54,10 +54,10 @@ def test_trainer_mps_accelerator(accelerator_value):
 
 @RunIf(mps=True)
 @pytest.mark.parametrize("devices", [1, [0], "-1"])
-def test_single_gpu_model(tmpdir, devices):
+def test_single_gpu_model(tmp_path, devices):
     """Make sure single GPU works."""
     trainer_options = {
-        "default_root_dir": tmpdir,
+        "default_root_dir": tmp_path,
         "enable_progress_bar": False,
         "max_epochs": 1,
         "limit_train_batches": 0.1,

tests/tests_pytorch/accelerators/test_xla.py

@@ -52,7 +52,7 @@ def forward(self, x):
 
 @RunIf(tpu=True, standalone=True)
 @mock.patch.dict(os.environ, os.environ.copy(), clear=True)
-def test_resume_training_on_cpu(tmpdir):
+def test_resume_training_on_cpu(tmp_path):
     """Checks if training can be resumed from a saved checkpoint on CPU."""
     # Train a model on TPU
     model = BoringModel()
@@ -72,16 +72,16 @@ def test_resume_training_on_cpu(tmpdir):
     assert weight_tensor.device == torch.device("cpu")
 
     # Verify that training is resumed on CPU
-    trainer = Trainer(max_epochs=1, default_root_dir=tmpdir)
+    trainer = Trainer(max_epochs=1, default_root_dir=tmp_path)
     trainer.fit(model, ckpt_path=model_path)
 
 
 @RunIf(tpu=True)
 @mock.patch.dict(os.environ, os.environ.copy(), clear=True)
-def test_if_test_works_after_train(tmpdir):
+def test_if_test_works_after_train(tmp_path):
     """Ensure that .test() works after .fit()"""
     model = BoringModel()
-    trainer = Trainer(max_epochs=1, accelerator="tpu", devices="auto", default_root_dir=tmpdir, fast_dev_run=True)
+    trainer = Trainer(max_epochs=1, accelerator="tpu", devices="auto", default_root_dir=tmp_path, fast_dev_run=True)
     trainer.fit(model)
     out = trainer.test(model)
     assert len(out) == 1
@@ -159,13 +159,13 @@ def on_train_end(self):
 
 @RunIf(tpu=True)
 @mock.patch.dict(os.environ, os.environ.copy(), clear=True)
-def test_manual_optimization_tpus(tmpdir):
+def test_manual_optimization_tpus(tmp_path):
     model = ManualOptimizationModel()
     model_copy = deepcopy(model)
 
     trainer = Trainer(
         max_epochs=1,
-        default_root_dir=tmpdir,
+        default_root_dir=tmp_path,
         limit_train_batches=3,
         limit_test_batches=0,
         limit_val_batches=0,
@@ -198,13 +198,13 @@ def test_strategy_choice_tpu_strategy():
 
 @RunIf(tpu=True)
 @mock.patch.dict(os.environ, os.environ.copy(), clear=True)
-def test_auto_parameters_tying_tpus(tmpdir):
+def test_auto_parameters_tying_tpus(tmp_path):
     model = WeightSharingModule()
     shared_params = find_shared_parameters(model)
 
     assert shared_params[0] == ["layer_1.weight", "layer_3.weight"]
 
-    trainer = Trainer(default_root_dir=tmpdir, limit_train_batches=3, accelerator="tpu", devices="auto", max_epochs=1)
+    trainer = Trainer(default_root_dir=tmp_path, limit_train_batches=3, accelerator="tpu", devices="auto", max_epochs=1)
     trainer.fit(model)
 
     assert torch.equal(model.layer_1.weight, model.layer_3.weight)
@@ -236,9 +236,9 @@ def forward(self, x):
 
 @RunIf(tpu=True)
 @mock.patch.dict(os.environ, os.environ.copy(), clear=True)
-def test_auto_parameters_tying_tpus_nested_module(tmpdir):
+def test_auto_parameters_tying_tpus_nested_module(tmp_path):
     model = NestedModule()
-    trainer = Trainer(default_root_dir=tmpdir, limit_train_batches=3, accelerator="tpu", devices="auto", max_epochs=1)
+    trainer = Trainer(default_root_dir=tmp_path, limit_train_batches=3, accelerator="tpu", devices="auto", max_epochs=1)
     trainer.fit(model)
 
     assert torch.all(torch.eq(model.net_a.layer.weight, model.net_b.layer.weight))

tests/tests_pytorch/conftest.py

@@ -271,8 +271,8 @@ def caplog(caplog):
 
 
 @pytest.fixture()
-def tmpdir_server(tmpdir):
-    Handler = partial(SimpleHTTPRequestHandler, directory=str(tmpdir))
+def tmpdir_server(tmp_path):
+    Handler = partial(SimpleHTTPRequestHandler, directory=str(tmp_path))
     from http.server import ThreadingHTTPServer
 
     with ThreadingHTTPServer(("localhost", 0), Handler) as server:

tests/tests_pytorch/test_cli.py

@@ -47,6 +47,7 @@
 from lightning.pytorch.utilities.exceptions import MisconfigurationException
 from lightning.pytorch.utilities.imports import _TORCHVISION_AVAILABLE
 from lightning_utilities import compare_version
+from lightning_utilities.core.imports import RequirementCache
 from lightning_utilities.test.warning import no_warning_call
 from tensorboard.backend.event_processing import event_accumulator
 from tensorboard.plugins.hparams.plugin_data_pb2 import HParamsPluginData
@@ -635,14 +636,14 @@ def test_cli_config_overwrite(cleandir):
         LightningCLI(BoringModel, save_config_kwargs={"overwrite": True}, trainer_defaults=trainer_defaults)
 
 
-def test_cli_config_filename(tmpdir):
+def test_cli_config_filename(tmp_path):
     with mock.patch("sys.argv", ["any.py", "fit"]):
         LightningCLI(
             BoringModel,
-            trainer_defaults={"default_root_dir": str(tmpdir), "logger": False, "max_steps": 1, "max_epochs": 1},
+            trainer_defaults={"default_root_dir": str(tmp_path), "logger": False, "max_steps": 1, "max_epochs": 1},
             save_config_kwargs={"config_filename": "name.yaml"},
         )
-    assert os.path.isfile(tmpdir / "name.yaml")
+    assert os.path.isfile(tmp_path / "name.yaml")
 
 
 @pytest.mark.parametrize("run", [False, True])
@@ -1481,6 +1482,12 @@ def __init__(self, activation: torch.nn.Module = lazy_instance(torch.nn.LeakyReL
         assert cli.model.activation is not model.activation
 
 
+@pytest.mark.xfail(
+    # https://github.com/omni-us/jsonargparse/issues/473
+    condition=(RequirementCache("jsonargparse>=4.27.6")),
+    strict=False,
+    reason="Breaking change for `lazy_instance` in jsonargparse",
+)
 def test_ddpstrategy_instantiation_and_find_unused_parameters(mps_count_0):
     strategy_default = lazy_instance(DDPStrategy, find_unused_parameters=True)
     with mock.patch("sys.argv", ["any.py", "--trainer.strategy.process_group_backend=group"]):