Avoid circular imports when lightning-habana or lightning-graphcore is installed

src/lightning/pytorch/trainer/configuration_validator.py

@@ -16,7 +16,7 @@
 from lightning.fabric.utilities.warnings import PossibleUserWarning
 from lightning.pytorch.trainer.states import TrainerFn
 from lightning.pytorch.utilities.exceptions import MisconfigurationException
-from lightning.pytorch.utilities.imports import _LIGHTNING_GRAPHCORE_AVAILABLE
+from lightning.pytorch.utilities.imports import _lightning_graphcore_available
 from lightning.pytorch.utilities.model_helpers import is_overridden
 from lightning.pytorch.utilities.rank_zero import rank_zero_deprecation, rank_zero_warn
 from lightning.pytorch.utilities.signature_utils import is_param_in_hook_signature
@@ -123,7 +123,7 @@ def __verify_batch_transfer_support(trainer: "pl.Trainer") -> None:
     datahook_selector = trainer._data_connector._datahook_selector
     assert datahook_selector is not None
     for hook in batch_transfer_hooks:
-        if _LIGHTNING_GRAPHCORE_AVAILABLE:
+        if _lightning_graphcore_available():
             from lightning_graphcore import IPUAccelerator
 
             # TODO: This code could be done in a hook in the IPUAccelerator as it's a simple error check

src/lightning/pytorch/trainer/connectors/accelerator_connector.py

@@ -64,8 +64,8 @@
 from lightning.pytorch.utilities.imports import (
     _LIGHTNING_BAGUA_AVAILABLE,
     _LIGHTNING_COLOSSALAI_AVAILABLE,
-    _LIGHTNING_GRAPHCORE_AVAILABLE,
-    _LIGHTNING_HABANA_AVAILABLE,
+    _lightning_graphcore_available,
+    _lightning_habana_available,
 )
 from lightning.pytorch.utilities.rank_zero import rank_zero_info, rank_zero_warn
 
@@ -338,12 +338,12 @@ def _choose_auto_accelerator(self) -> str:
         """Choose the accelerator type (str) based on availability."""
         if XLAAccelerator.is_available():
             return "tpu"
-        if _LIGHTNING_GRAPHCORE_AVAILABLE:
+        if _lightning_graphcore_available():
             from lightning_graphcore import IPUAccelerator
 
             if IPUAccelerator.is_available():
                 return "ipu"
-        if _LIGHTNING_HABANA_AVAILABLE:
+        if _lightning_habana_available():
             from lightning_habana import HPUAccelerator
 
             if HPUAccelerator.is_available():
@@ -411,7 +411,7 @@ def _choose_and_init_cluster_environment(self) -> ClusterEnvironment:
 
     def _choose_strategy(self) -> Union[Strategy, str]:
         if self._accelerator_flag == "ipu":
-            if not _LIGHTNING_GRAPHCORE_AVAILABLE:
+            if not _lightning_graphcore_available():
                 raise ImportError(
                     "You have passed `accelerator='ipu'` but the IPU integration  is not installed."
                     " Please run `pip install lightning-graphcore` or check out"
@@ -421,7 +421,7 @@ def _choose_strategy(self) -> Union[Strategy, str]:
 
             return IPUStrategy.strategy_name
         if self._accelerator_flag == "hpu":
-            if not _LIGHTNING_HABANA_AVAILABLE:
+            if not _lightning_habana_available():
                 raise ImportError(
                     "You have asked for HPU but you miss install related integration."
                     " Please run `pip install lightning-habana` or see for further instructions"
@@ -490,7 +490,7 @@ def _check_and_init_precision(self) -> PrecisionPlugin:
         if isinstance(self._precision_plugin_flag, PrecisionPlugin):
             return self._precision_plugin_flag
 
-        if _LIGHTNING_GRAPHCORE_AVAILABLE:
+        if _lightning_graphcore_available():
             from lightning_graphcore import IPUAccelerator, IPUPrecision
 
             # TODO: For the strategies that have a fixed precision class, we don't really need this logic
@@ -500,7 +500,7 @@ def _check_and_init_precision(self) -> PrecisionPlugin:
             if isinstance(self.accelerator, IPUAccelerator):
                 return IPUPrecision(self._precision_flag)
 
-        if _LIGHTNING_HABANA_AVAILABLE:
+        if _lightning_habana_available():
             from lightning_habana import HPUAccelerator, HPUPrecisionPlugin
 
             if isinstance(self.accelerator, HPUAccelerator):
@@ -567,7 +567,7 @@ def _validate_precision_choice(self) -> None:
                     f"The `XLAAccelerator` can only be used with a `XLAPrecisionPlugin`,"
                     f" found: {self._precision_plugin_flag}."
                 )
-        if _LIGHTNING_HABANA_AVAILABLE:
+        if _lightning_habana_available():
             from lightning_habana import HPUAccelerator
 
             if isinstance(self.accelerator, HPUAccelerator) and self._precision_flag not in (
@@ -622,7 +622,7 @@ def _lazy_init_strategy(self) -> None:
                 f" found {self.strategy.__class__.__name__}."
             )
 
-        if _LIGHTNING_HABANA_AVAILABLE:
+        if _lightning_habana_available():
             from lightning_habana import HPUAccelerator, HPUParallelStrategy, SingleHPUStrategy
 
             if isinstance(self.accelerator, HPUAccelerator) and not isinstance(
@@ -641,7 +641,7 @@ def is_distributed(self) -> bool:
             DeepSpeedStrategy,
             XLAStrategy,
         ]
-        if _LIGHTNING_HABANA_AVAILABLE:
+        if _lightning_habana_available():
             from lightning_habana import HPUParallelStrategy
 
             distributed_strategies.append(HPUParallelStrategy)
@@ -694,7 +694,7 @@ def _register_external_accelerators_and_strategies() -> None:
         if "bagua" not in StrategyRegistry:
             BaguaStrategy.register_strategies(StrategyRegistry)
 
-    if _LIGHTNING_HABANA_AVAILABLE:
+    if _lightning_habana_available():
         from lightning_habana import HPUAccelerator, HPUParallelStrategy, SingleHPUStrategy
 
         # TODO: Prevent registering multiple times
@@ -705,7 +705,7 @@ def _register_external_accelerators_and_strategies() -> None:
         if "hpu_single" not in StrategyRegistry:
             SingleHPUStrategy.register_strategies(StrategyRegistry)
 
-    if _LIGHTNING_GRAPHCORE_AVAILABLE:
+    if _lightning_graphcore_available():
         from lightning_graphcore import IPUAccelerator, IPUStrategy
 
         # TODO: Prevent registering multiple times

src/lightning/pytorch/trainer/connectors/data_connector.py

@@ -34,7 +34,7 @@
 from lightning.pytorch.utilities.combined_loader import CombinedLoader
 from lightning.pytorch.utilities.data import _is_dataloader_shuffled, _update_dataloader
 from lightning.pytorch.utilities.exceptions import MisconfigurationException
-from lightning.pytorch.utilities.imports import _LIGHTNING_GRAPHCORE_AVAILABLE
+from lightning.pytorch.utilities.imports import _lightning_graphcore_available
 from lightning.pytorch.utilities.model_helpers import is_overridden
 from lightning.pytorch.utilities.rank_zero import rank_zero_warn, WarningCache
 from lightning.pytorch.utilities.types import EVAL_DATALOADERS, TRAIN_DATALOADERS
@@ -165,7 +165,7 @@ def attach_datamodule(
         datamodule.trainer = trainer
 
     def _requires_distributed_sampler(self, dataloader: DataLoader) -> bool:
-        if _LIGHTNING_GRAPHCORE_AVAILABLE:
+        if _lightning_graphcore_available():
             from lightning_graphcore import IPUAccelerator
 
             # `DistributedSampler` is never used with `poptorch.DataLoader`
@@ -190,7 +190,7 @@ def _prepare_dataloader(self, dataloader: object, shuffle: bool, mode: RunningSt
         if not isinstance(dataloader, DataLoader):
             return dataloader
 
-        if _LIGHTNING_GRAPHCORE_AVAILABLE:
+        if _lightning_graphcore_available():
             from lightning_graphcore import IPUAccelerator
 
             # IPUs use a custom `poptorch.DataLoader` which we might need to convert to

src/lightning/pytorch/trainer/setup.py

@@ -28,7 +28,7 @@
     XLAProfiler,
 )
 from lightning.pytorch.utilities.exceptions import MisconfigurationException
-from lightning.pytorch.utilities.imports import _LIGHTNING_GRAPHCORE_AVAILABLE, _LIGHTNING_HABANA_AVAILABLE
+from lightning.pytorch.utilities.imports import _lightning_graphcore_available, _lightning_habana_available
 from lightning.pytorch.utilities.rank_zero import rank_zero_info, rank_zero_warn
 
 
@@ -158,7 +158,7 @@ def _log_device_info(trainer: "pl.Trainer") -> None:
     num_tpu_cores = trainer.num_devices if isinstance(trainer.accelerator, XLAAccelerator) else 0
     rank_zero_info(f"TPU available: {XLAAccelerator.is_available()}, using: {num_tpu_cores} TPU cores")
 
-    if _LIGHTNING_GRAPHCORE_AVAILABLE:
+    if _lightning_graphcore_available():
         from lightning_graphcore import IPUAccelerator
 
         num_ipus = trainer.num_devices if isinstance(trainer.accelerator, IPUAccelerator) else 0
@@ -168,7 +168,7 @@ def _log_device_info(trainer: "pl.Trainer") -> None:
         ipu_available = False
     rank_zero_info(f"IPU available: {ipu_available}, using: {num_ipus} IPUs")
 
-    if _LIGHTNING_HABANA_AVAILABLE:
+    if _lightning_habana_available():
         from lightning_habana import HPUAccelerator
 
         num_hpus = trainer.num_devices if isinstance(trainer.accelerator, HPUAccelerator) else 0
@@ -192,13 +192,13 @@ def _log_device_info(trainer: "pl.Trainer") -> None:
     if XLAAccelerator.is_available() and not isinstance(trainer.accelerator, XLAAccelerator):
         rank_zero_warn("TPU available but not used. You can set it by doing `Trainer(accelerator='tpu')`.")
 
-    if _LIGHTNING_GRAPHCORE_AVAILABLE:
+    if _lightning_graphcore_available():
         from lightning_graphcore import IPUAccelerator
 
         if IPUAccelerator.is_available() and not isinstance(trainer.accelerator, IPUAccelerator):
             rank_zero_warn("IPU available but not used. You can set it by doing `Trainer(accelerator='ipu')`.")
 
-    if _LIGHTNING_HABANA_AVAILABLE:
+    if _lightning_habana_available():
         from lightning_habana import HPUAccelerator
 
         if HPUAccelerator.is_available() and not isinstance(trainer.accelerator, HPUAccelerator):

src/lightning/pytorch/utilities/imports.py

@@ -41,5 +41,15 @@ def _try_import_module(module_name: str) -> bool:
         return False
 
 
-_LIGHTNING_GRAPHCORE_AVAILABLE = RequirementCache("lightning-graphcore") and _try_import_module("lightning_graphcore")
-_LIGHTNING_HABANA_AVAILABLE = RequirementCache("lightning-habana") and _try_import_module("lightning_habana")
+@functools.lru_cache(maxsize=1)
+def _lightning_graphcore_available() -> bool:
+    # This is defined as a function instead of a constant to avoid circular imports, because `lightning_graphcore`
+    # also imports Lightning
+    return bool(RequirementCache("lightning-graphcore")) and _try_import_module("lightning_graphcore")
+
+
+@functools.lru_cache(maxsize=1)
+def _lightning_habana_available() -> bool:
+    # This is defined as a function instead of a constant to avoid circular imports, because `lightning_habana`
+    # also imports Lightning
+    return bool(RequirementCache("lightning-habana")) and _try_import_module("lightning_habana")

tests/tests_pytorch/trainer/connectors/test_accelerator_connector.py

@@ -53,14 +53,14 @@
 from lightning.pytorch.strategies.launchers import _SubprocessScriptLauncher
 from lightning.pytorch.trainer.connectors.accelerator_connector import _AcceleratorConnector, _set_torch_flags
 from lightning.pytorch.utilities.exceptions import MisconfigurationException
-from lightning.pytorch.utilities.imports import _LIGHTNING_GRAPHCORE_AVAILABLE, _LIGHTNING_HABANA_AVAILABLE
+from lightning.pytorch.utilities.imports import _lightning_graphcore_available, _lightning_habana_available
 from tests_pytorch.conftest import mock_cuda_count, mock_mps_count, mock_tpu_available, mock_xla_available
 from tests_pytorch.helpers.runif import RunIf
 
-if _LIGHTNING_GRAPHCORE_AVAILABLE:
+if _lightning_graphcore_available():
     from lightning_graphcore import IPUAccelerator, IPUStrategy
 
-if _LIGHTNING_HABANA_AVAILABLE:
+if _lightning_habana_available():
     from lightning_habana import HPUAccelerator, SingleHPUStrategy
 
 
@@ -935,7 +935,7 @@ def _mock_tpu_available(value):
     assert connector.strategy.launcher.is_interactive_compatible
 
     # Single/Multi IPU: strategy is the same
-    if _LIGHTNING_GRAPHCORE_AVAILABLE:
+    if _lightning_graphcore_available():
         with monkeypatch.context():
             mock_cuda_count(monkeypatch, 0)
             mock_mps_count(monkeypatch, 0)
@@ -949,7 +949,7 @@ def _mock_tpu_available(value):
         assert connector.strategy.launcher is None
 
     # Single HPU
-    if _LIGHTNING_HABANA_AVAILABLE:
+    if _lightning_habana_available():
         import lightning_habana
 
         with monkeypatch.context():
@@ -967,7 +967,7 @@ def _mock_tpu_available(value):
     monkeypatch.undo()  # for some reason `.context()` is not working properly
     _mock_interactive()
 
-    if not is_interactive and _LIGHTNING_HABANA_AVAILABLE:  # HPU does not support interactive environments
+    if not is_interactive and _lightning_habana_available():  # HPU does not support interactive environments
         from lightning_habana import HPUParallelStrategy
 
         # Multi HPU