Fixes incorrect strategy init with HPUAccelerator (#19615)

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

@@ -408,21 +408,25 @@ def _choose_and_init_cluster_environment(self) -> ClusterEnvironment:
         return LightningEnvironment()
 
     def _choose_strategy(self) -> Union[Strategy, str]:
-        if self._accelerator_flag == "hpu":
-            if not _habana_available_and_importable():
-                raise ImportError(
-                    "You have asked for HPU but you miss install related integration."
-                    " Please run `pip install lightning-habana` or see for further instructions"
-                    " in https://github.com/Lightning-AI/lightning-Habana/."
-                )
-            if self._parallel_devices and len(self._parallel_devices) > 1:
-                from lightning_habana import HPUParallelStrategy
+        if _habana_available_and_importable():
+            from lightning_habana import HPUAccelerator
 
-                return HPUParallelStrategy.strategy_name
+            if self._accelerator_flag == "hpu" or isinstance(self._accelerator_flag, HPUAccelerator):
+                if self._parallel_devices and len(self._parallel_devices) > 1:
+                    from lightning_habana import HPUParallelStrategy
 
-            from lightning_habana import SingleHPUStrategy
+                    return HPUParallelStrategy.strategy_name
+
+                from lightning_habana import SingleHPUStrategy
+
+                return SingleHPUStrategy(device=torch.device("hpu"))
+        if self._accelerator_flag == "hpu" and not _habana_available_and_importable():
+            raise ImportError(
+                "You asked to run with HPU but you are missing a required dependency."
+                " Please run `pip install lightning-habana` or seek further instructions"
+                " in https://github.com/Lightning-AI/lightning-Habana/."
+            )
 
-            return SingleHPUStrategy(device=torch.device("hpu"))
         if self._accelerator_flag == "tpu" or isinstance(self._accelerator_flag, XLAAccelerator):
             if self._parallel_devices and len(self._parallel_devices) > 1:
                 return XLAStrategy.strategy_name