Merge branch 'master' into refactor/num_ipus

Author: rohitgr7

.github/mergify.yml

@@ -50,6 +50,7 @@ pull_request_rules:
   - name: Not ready yet
     conditions:
       - or:
+        - label="has conflicts"
         - "#approved-reviews-by=0" # number of review approvals
         - "#changes-requested-reviews-by>=1" # no requested changes
     actions:

CHANGELOG.md

@@ -539,6 +539,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Deprecated `Trainer.ipus` in favor of `Trainer.num_devices` when IPU is used ([#12386](https://github.com/PyTorchLightning/pytorch-lightning/pull/12386))
 
 
+- Deprecated `Trainer.num_processes` in favor of `Trainer.num_devices` ([#12388](https://github.com/PyTorchLightning/pytorch-lightning/pull/12388))
+
+
 ### Removed
 
 - Removed deprecated parameter `method` in `pytorch_lightning.utilities.model_helpers.is_overridden` ([#10507](https://github.com/PyTorchLightning/pytorch-lightning/pull/10507))
@@ -729,12 +732,18 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Removed `AcceleratorConnector.root_gpu` property ([#12262](https://github.com/PyTorchLightning/pytorch-lightning/pull/12262))
 
 
+- Removed `AcceleratorConnector.tpu_id` property ([#12387](https://github.com/PyTorchLightning/pytorch-lightning/pull/12387))
+
+
 - Removed `AcceleratorConnector.num_gpus` property ([#12384](https://github.com/PyTorchLightning/pytorch-lightning/pull/12384))
 
 
 - Removed `AcceleratorConnector.num_ipus` property ([#12386](https://github.com/PyTorchLightning/pytorch-lightning/pull/12386))
 
 
+- Removed `AcceleratorConnector.num_processes` property ([#12388](https://github.com/PyTorchLightning/pytorch-lightning/pull/12388))
+
+
 ### Fixed
 
 - Fixed an issue where `ModelCheckpoint` could delete older checkpoints when `dirpath` has changed during resumed training ([#12045](https://github.com/PyTorchLightning/pytorch-lightning/pull/12045))

pytorch_lightning/trainer/connectors/accelerator_connector.py

@@ -784,10 +784,6 @@ def _lazy_init_strategy(self) -> None:
     def parallel_devices(self) -> List[Union[torch.device, int]]:
         return self._parallel_devices
 
-    @property
-    def num_processes(self) -> int:
-        return self.devices if self.devices is not None else 1
-
     @property
     def devices(self) -> int:
         if isinstance(self.strategy, SingleDeviceStrategy):
@@ -809,6 +805,12 @@ def tpu_id(self) -> Optional[int]:
                 return self._tpu_cores[0]
         return None
 
+    @property
+    def num_ipus(self) -> int:
+        if isinstance(self.accelerator, IPUAccelerator):
+            return self.devices
+        return 0
+
     @property
     def gpus(self) -> Optional[Union[List[int], str, int]]:
         return self._gpus

pytorch_lightning/trainer/trainer.py

@@ -2052,7 +2052,11 @@ def num_devices(self) -> int:
 
     @property
     def num_processes(self) -> int:
-        return self._accelerator_connector.num_processes
+        rank_zero_deprecation(
+            "`Trainer.num_processes` is deprecated in v1.6 and will be removed in v1.8. "
+            "Please use `Trainer.num_devices` instead."
+        )
+        return self.num_devices
 
     @property
     def root_gpu(self) -> Optional[int]:

tests/accelerators/test_accelerator_connector.py

@@ -533,7 +533,7 @@ def test_accelerator_cpu_with_devices(devices, plugin):
 
     trainer = Trainer(accelerator="cpu", devices=devices)
 
-    assert trainer.num_processes == devices
+    assert trainer.num_devices == devices
     assert isinstance(trainer.strategy, plugin)
     assert isinstance(trainer.accelerator, CPUAccelerator)
 
@@ -545,7 +545,7 @@ def test_accelerator_cpu_with_num_processes_priority():
     with pytest.warns(UserWarning, match="The flag `devices=8` will be ignored,"):
         trainer = Trainer(accelerator="cpu", devices=8, num_processes=num_processes)
 
-    assert trainer.num_processes == num_processes
+    assert trainer.num_devices == num_processes
 
 
 @RunIf(min_gpus=2)
@@ -583,7 +583,7 @@ def test_validate_accelerator_and_devices():
 
     trainer = Trainer(accelerator="ddp_cpu", devices=2)
     assert isinstance(trainer.accelerator, CPUAccelerator)
-    assert trainer.num_processes == 2
+    assert trainer.num_devices == 2
 
 
 def test_set_devices_if_none_cpu():
@@ -962,7 +962,6 @@ def test_unsupported_ipu_choice(mock_ipu_acc_avail, monkeypatch):
 def test_devices_auto_choice_cpu(is_ipu_available_mock, is_tpu_available_mock, is_gpu_available_mock):
     trainer = Trainer(accelerator="auto", devices="auto")
     assert trainer.num_devices == 1
-    assert trainer.num_processes == 1
 
 
 @mock.patch("torch.cuda.is_available", return_value=True)

tests/deprecated_api/test_remove_1-8.py

@@ -1005,3 +1005,25 @@ def test_trainer_config_ipus(monkeypatch, trainer_kwargs, expected_ipus):
         " Please use `Trainer.num_devices` instead."
     ):
         trainer.ipus == expected_ipus
+
+
+@pytest.mark.parametrize(
+    ["trainer_kwargs", "expected_num_processes"],
+    [
+        ({}, 1),
+        ({"devices": 1}, 1),
+        ({"devices": 4}, 4),
+        ({"accelerator": "cpu", "devices": 1}, 0),
+        ({"accelerator": "gpu", "devices": 4}, 4),
+    ],
+)
+def test_trainer_num_processes(monkeypatch, trainer_kwargs, expected_num_processes):
+    if trainer_kwargs.get("accelerator") == "gpu":
+        monkeypatch.setattr(torch.cuda, "is_available", lambda: True)
+        monkeypatch.setattr(torch.cuda, "device_count", lambda: 16)
+    trainer = Trainer(**trainer_kwargs)
+    with pytest.deprecated_call(
+        match="`Trainer.num_processes` is deprecated in v1.6 and will be removed in v1.8. "
+        "Please use `Trainer.num_devices` instead."
+    ):
+        trainer.num_processes == expected_num_processes

tests/models/test_tpu.py

@@ -242,20 +242,11 @@ def test_dataloaders_passed_to_fit(tmpdir):
     assert trainer.state.finished, f"Training failed with {trainer.state}"
 
 
-@pytest.mark.parametrize(
-    ["tpu_cores", "expected_tpu_id"],
-    [(1, None), (8, None), ([1], 1), ([8], 8)],
-)
 @RunIf(tpu=True)
-def test_tpu_id_to_be_as_expected(tpu_cores, expected_tpu_id):
-    """Test if trainer.tpu_id is set as expected."""
-    assert Trainer(tpu_cores=tpu_cores)._accelerator_connector.tpu_id == expected_tpu_id
-
-
-def test_tpu_misconfiguration():
-    """Test if trainer.tpu_id is set as expected."""
+@pytest.mark.parametrize("tpu_cores", [[1, 8], "9, ", [9], [0], 2, 10])
+def test_tpu_misconfiguration(tpu_cores):
     with pytest.raises(MisconfigurationException, match="`tpu_cores` can only be"):
-        Trainer(tpu_cores=[1, 8])
+        Trainer(tpu_cores=tpu_cores)
 
 
 @pytest.mark.skipif(_TPU_AVAILABLE, reason="test requires missing TPU")
@@ -289,33 +280,6 @@ def test_broadcast(rank):
     xmp.spawn(test_broadcast, nprocs=8, start_method="fork")
 
 
-@pytest.mark.parametrize(
-    ["tpu_cores", "expected_tpu_id", "error_expected"],
-    [
-        (1, None, False),
-        (8, None, False),
-        ([1], 1, False),
-        ([8], 8, False),
-        ("1,", 1, False),
-        ("1", None, False),
-        ("9, ", 9, True),
-        ([9], 9, True),
-        ([0], 0, True),
-        (2, None, True),
-        (10, None, True),
-    ],
-)
-@RunIf(tpu=True)
-@pl_multi_process_test
-def test_tpu_choice(tmpdir, tpu_cores, expected_tpu_id, error_expected):
-    if error_expected:
-        with pytest.raises(MisconfigurationException, match=r".*tpu_cores` can only be 1, 8 or [<1-8>]*"):
-            Trainer(default_root_dir=tmpdir, tpu_cores=tpu_cores)
-    else:
-        trainer = Trainer(default_root_dir=tmpdir, tpu_cores=tpu_cores)
-        assert trainer._accelerator_connector.tpu_id == expected_tpu_id
-
-
 @pytest.mark.parametrize(
     ["cli_args", "expected"],
     [("--tpu_cores=8", {"tpu_cores": 8}), ("--tpu_cores=1,", {"tpu_cores": "1,"})],