test: replacing tmpdir with tmp_path in tests_pytorch/plugins (#…

…19648)

tests/tests_pytorch/plugins/precision/test_amp_integration.py

@@ -33,12 +33,12 @@ def configure_optimizers(self):
 
 
 @RunIf(min_cuda_gpus=1)
-def test_amp_fused_optimizer_parity(tmpdir):
+def test_amp_fused_optimizer_parity(tmp_path):
     def run(fused=False):
         seed_everything(1234)
         model = FusedOptimizerParityModel(fused)
         trainer = Trainer(
-            default_root_dir=tmpdir,
+            default_root_dir=tmp_path,
             accelerator="cuda",
             devices=1,
             precision="16-mixed",

tests/tests_pytorch/plugins/precision/test_double.py

@@ -142,22 +142,22 @@ def training_step(self, batch, batch_idx):
         DoublePrecisionBoringModelComplexBuffer,
     ],
 )
-def test_double_precision(tmpdir, boring_model):
+def test_double_precision(tmp_path, boring_model):
     model = boring_model()
 
-    trainer = Trainer(max_epochs=2, default_root_dir=tmpdir, fast_dev_run=2, precision="64-true", log_every_n_steps=1)
+    trainer = Trainer(max_epochs=2, default_root_dir=tmp_path, fast_dev_run=2, precision="64-true", log_every_n_steps=1)
     trainer.fit(model)
     trainer.test(model)
     trainer.predict(model)
 
 
 @RunIf(min_cuda_gpus=2)
-def test_double_precision_ddp(tmpdir):
+def test_double_precision_ddp(tmp_path):
     model = DoublePrecisionBoringModel()
 
     trainer = Trainer(
         max_epochs=1,
-        default_root_dir=tmpdir,
+        default_root_dir=tmp_path,
         strategy="ddp_spawn",
         accelerator="gpu",
         devices=2,

tests/tests_pytorch/plugins/test_amp_plugins.py

@@ -130,12 +130,12 @@ def configure_optimizers(self):
 
 @RunIf(min_cuda_gpus=2)
 @pytest.mark.parametrize("accum", [1, 2])
-def test_amp_gradient_unscale(tmpdir, accum: int):
+def test_amp_gradient_unscale(tmp_path, accum: int):
     model = TestPrecisionModel()
 
     trainer = Trainer(
         max_epochs=2,
-        default_root_dir=tmpdir,
+        default_root_dir=tmp_path,
         limit_train_batches=2,
         limit_val_batches=0,
         strategy="ddp_spawn",
@@ -153,7 +153,7 @@ def test_amp_gradient_unscale(tmpdir, accum: int):
 
 
 @RunIf(min_cuda_gpus=1)
-def test_amp_skip_optimizer(tmpdir):
+def test_amp_skip_optimizer(tmp_path):
     """Test that optimizers can be skipped when using amp."""
 
     class CustomBoringModel(BoringModel):
@@ -183,7 +183,7 @@ def configure_optimizers(self):
                 torch.optim.SGD(self.layer2.parameters(), lr=0.1),
             ]
 
-    trainer = Trainer(default_root_dir=tmpdir, accelerator="gpu", devices=1, fast_dev_run=1, precision="16-mixed")
+    trainer = Trainer(default_root_dir=tmp_path, accelerator="gpu", devices=1, fast_dev_run=1, precision="16-mixed")
     model = CustomBoringModel()
     trainer.fit(model)
 

tests/tests_pytorch/plugins/test_checkpoint_io_plugin.py

@@ -37,16 +37,16 @@ def remove_checkpoint(self, path: _PATH) -> None:
         os.remove(path)
 
 
-def test_checkpoint_plugin_called(tmpdir):
+def test_checkpoint_plugin_called(tmp_path):
     """Ensure that the custom checkpoint IO plugin and torch checkpoint IO plugin is called when saving/loading."""
     checkpoint_plugin = CustomCheckpointIO()
     checkpoint_plugin = MagicMock(wraps=checkpoint_plugin, spec=CustomCheckpointIO)
 
-    ck = ModelCheckpoint(dirpath=tmpdir, save_last=True)
+    ck = ModelCheckpoint(dirpath=tmp_path, save_last=True)
 
     model = BoringModel()
     trainer = Trainer(
-        default_root_dir=tmpdir,
+        default_root_dir=tmp_path,
         accelerator="cpu",
         strategy=SingleDeviceStrategy("cpu", checkpoint_io=checkpoint_plugin),
         callbacks=ck,
@@ -57,22 +57,22 @@ def test_checkpoint_plugin_called(tmpdir):
     )
     trainer.fit(model)
 
-    ckpt_files = {fn.name for fn in Path(tmpdir).glob("*.ckpt")}
+    ckpt_files = {fn.name for fn in Path(tmp_path).glob("*.ckpt")}
     assert ckpt_files == {"epoch=1-step=2.ckpt", "last.ckpt"}
-    assert trainer.checkpoint_callback.best_model_path == tmpdir / "epoch=1-step=2.ckpt"
-    assert trainer.checkpoint_callback.last_model_path == tmpdir / "last.ckpt"
+    assert trainer.checkpoint_callback.best_model_path == str(tmp_path / "epoch=1-step=2.ckpt")
+    assert trainer.checkpoint_callback.last_model_path == str(tmp_path / "last.ckpt")
     assert checkpoint_plugin.save_checkpoint.call_count == 4
     assert checkpoint_plugin.remove_checkpoint.call_count == 1
 
     trainer.test(model, ckpt_path=ck.last_model_path)
-    checkpoint_plugin.load_checkpoint.assert_called_with(tmpdir / "last.ckpt")
+    checkpoint_plugin.load_checkpoint.assert_called_with(str(tmp_path / "last.ckpt"))
 
     checkpoint_plugin.reset_mock()
-    ck = ModelCheckpoint(dirpath=tmpdir, save_last=True)
+    ck = ModelCheckpoint(dirpath=tmp_path, save_last=True)
 
     model = BoringModel()
     trainer = Trainer(
-        default_root_dir=tmpdir,
+        default_root_dir=tmp_path,
         accelerator="cpu",
         strategy=SingleDeviceStrategy("cpu"),
         plugins=[checkpoint_plugin],
@@ -84,19 +84,19 @@ def test_checkpoint_plugin_called(tmpdir):
     )
     trainer.fit(model)
 
-    ckpt_files = {fn.name for fn in Path(tmpdir).glob("*.ckpt")}
+    ckpt_files = {fn.name for fn in Path(tmp_path).glob("*.ckpt")}
     assert ckpt_files == {"epoch=1-step=2.ckpt", "last.ckpt", "epoch=1-step=2-v1.ckpt", "last-v1.ckpt"}
-    assert trainer.checkpoint_callback.best_model_path == tmpdir / "epoch=1-step=2-v1.ckpt"
-    assert trainer.checkpoint_callback.last_model_path == tmpdir / "last-v1.ckpt"
+    assert trainer.checkpoint_callback.best_model_path == str(tmp_path / "epoch=1-step=2-v1.ckpt")
+    assert trainer.checkpoint_callback.last_model_path == str(tmp_path / "last-v1.ckpt")
     assert checkpoint_plugin.save_checkpoint.call_count == 4
     assert checkpoint_plugin.remove_checkpoint.call_count == 1
 
     trainer.test(model, ckpt_path=ck.last_model_path)
     checkpoint_plugin.load_checkpoint.assert_called_once()
-    checkpoint_plugin.load_checkpoint.assert_called_with(tmpdir / "last-v1.ckpt")
+    checkpoint_plugin.load_checkpoint.assert_called_with(str(tmp_path / "last-v1.ckpt"))
 
 
-def test_async_checkpoint_plugin(tmpdir):
+def test_async_checkpoint_plugin(tmp_path):
     """Ensure that the custom checkpoint IO plugin and torch checkpoint IO plugin is called when async saving and
     loading."""
 
@@ -111,11 +111,11 @@ def on_fit_start(self):
             base_ckpt_io.save_checkpoint = Mock(wraps=base_ckpt_io.save_checkpoint)
             base_ckpt_io.remove_checkpoint = Mock(wraps=base_ckpt_io.remove_checkpoint)
 
-    ck = ModelCheckpoint(dirpath=tmpdir, save_top_k=2, monitor="step", mode="max")
+    ck = ModelCheckpoint(dirpath=tmp_path, save_top_k=2, monitor="step", mode="max")
 
     model = CustomBoringModel()
     trainer = Trainer(
-        default_root_dir=tmpdir,
+        default_root_dir=tmp_path,
         plugins=[checkpoint_plugin],
         callbacks=ck,
         max_epochs=3,