diff --git a/src/lightning_lite/connector.py b/src/lightning_lite/connector.py
index 3f2299304f..1709f56f49 100644
--- a/src/lightning_lite/connector.py
+++ b/src/lightning_lite/connector.py
@@ -55,7 +55,7 @@ from lightning_lite.strategies import (
 from lightning_lite.strategies.ddp_spawn import _DDP_FORK_ALIASES
 from lightning_lite.utilities import _StrategyType, rank_zero_info, rank_zero_warn
 from lightning_lite.utilities.device_parser import determine_root_gpu_device
-from lightning_lite.utilities.imports import _HPU_AVAILABLE, _IPU_AVAILABLE, _IS_INTERACTIVE
+from lightning_lite.utilities.imports import _IS_INTERACTIVE
 
 _PLUGIN = Union[Precision, ClusterEnvironment, CheckpointIO]
 _PLUGIN_INPUT = Union[_PLUGIN, str]
@@ -109,7 +109,7 @@ class _Connector:
 
         # Raise an exception if there are conflicts between flags
         # Set each valid flag to `self._x_flag` after validation
-        # For devices: Assign gpus, ipus, etc. to the accelerator flag and devices flag
+        # For devices: Assign gpus, etc. to the accelerator flag and devices flag
         self._strategy_flag: Optional[Union[Strategy, str]] = None
         self._accelerator_flag: Optional[Union[Accelerator, str]] = None
         self._precision_input: Optional[_PRECISION_INPUT] = None
@@ -302,10 +302,6 @@ class _Connector:
         if self._accelerator_flag == "auto":
             if TPUAccelerator.is_available():
                 return "tpu"
-            if _IPU_AVAILABLE:
-                return "ipu"
-            if _HPU_AVAILABLE:
-                return "hpu"
             if MPSAccelerator.is_available():
                 return "mps"
             if CUDAAccelerator.is_available():
@@ -499,8 +495,6 @@ class _Connector:
             self.strategy.set_world_ranks()
         self.strategy._configure_launcher()
 
-        from lightning_lite.utilities import _IS_INTERACTIVE
-
         if _IS_INTERACTIVE and self.strategy.launcher and not self.strategy.launcher.is_interactive_compatible:
             raise RuntimeError(
                 f"`Lite(strategy={self._strategy_flag!r})` is not compatible with an interactive"
diff --git a/src/lightning_lite/plugins/io/xla_plugin.py b/src/lightning_lite/plugins/io/xla_plugin.py
index 75c13898eb..57e5fa7931 100644
--- a/src/lightning_lite/plugins/io/xla_plugin.py
+++ b/src/lightning_lite/plugins/io/xla_plugin.py
@@ -15,16 +15,13 @@ import os
 from typing import Any, Dict, Optional
 
 from lightning_utilities.core.apply_func import apply_to_collection
+from lightning_utilities.core.imports import RequirementCache
 
 from lightning_lite.accelerators.tpu import _XLA_AVAILABLE
 from lightning_lite.plugins.io.torch_plugin import TorchCheckpointIO
 from lightning_lite.utilities.cloud_io import get_filesystem
-from lightning_lite.utilities.imports import _OMEGACONF_AVAILABLE
 from lightning_lite.utilities.types import _PATH
 
-if _OMEGACONF_AVAILABLE:
-    from omegaconf import DictConfig, ListConfig, OmegaConf
-
 
 class XLACheckpointIO(TorchCheckpointIO):
     """CheckpointIO that utilizes :func:`xm.save` to save checkpoints for TPU training strategies."""
@@ -54,9 +51,10 @@ class XLACheckpointIO(TorchCheckpointIO):
             )
         fs = get_filesystem(path)
         fs.makedirs(os.path.dirname(path), exist_ok=True)
-        # Todo: TypeError: 'mappingproxy' object does not support item assignment
-        # Ref: https://github.com/pytorch/xla/issues/2773
-        if _OMEGACONF_AVAILABLE:
+        if RequirementCache("omegaconf"):
+            # workaround for https://github.com/pytorch/xla/issues/2773
+            from omegaconf import DictConfig, ListConfig, OmegaConf
+
             checkpoint = apply_to_collection(checkpoint, (DictConfig, ListConfig), OmegaConf.to_container)
         import torch_xla.core.xla_model as xm
 
diff --git a/src/lightning_lite/plugins/precision/deepspeed.py b/src/lightning_lite/plugins/precision/deepspeed.py
index 3817bf3aa3..265dfacfdb 100644
--- a/src/lightning_lite/plugins/precision/deepspeed.py
+++ b/src/lightning_lite/plugins/precision/deepspeed.py
@@ -21,10 +21,10 @@ from typing_extensions import Literal
 from lightning_lite.plugins.precision.precision import Precision
 from lightning_lite.plugins.precision.utils import _convert_fp_tensor
 from lightning_lite.utilities.enums import AMPType, PrecisionType
-from lightning_lite.utilities.imports import _APEX_AVAILABLE
 from lightning_lite.utilities.types import Steppable
 
 _DEEPSPEED_AVAILABLE = RequirementCache("deepspeed")
+_APEX_AVAILABLE = RequirementCache("apex")
 if TYPE_CHECKING and _DEEPSPEED_AVAILABLE:
     import deepspeed
 
@@ -49,7 +49,7 @@ class DeepSpeedPrecision(Precision):
     def __init__(self, precision: Literal[16, 32, "bf16"], amp_type: str, amp_level: Optional[str] = None) -> None:
         if amp_type == AMPType.APEX:
             if not _APEX_AVAILABLE:
-                raise ImportError(
+                raise ModuleNotFoundError(
                     "You have asked for Apex AMP but `apex` is not installed."
                     " Install `apex` using this guide: https://github.com/NVIDIA/apex"
                 )
diff --git a/src/lightning_lite/utilities/__init__.py b/src/lightning_lite/utilities/__init__.py
index 63e9049492..17f37679f2 100644
--- a/src/lightning_lite/utilities/__init__.py
+++ b/src/lightning_lite/utilities/__init__.py
@@ -15,20 +15,6 @@
 
 from lightning_lite.utilities.apply_func import move_data_to_device  # noqa: F401
 from lightning_lite.utilities.enums import _AcceleratorType, _StrategyType, AMPType, LightningEnum  # noqa: F401
-
-# TODO(lite): Avoid importing protected attributes in `__init__.py` files
-from lightning_lite.utilities.imports import (  # noqa: F401
-    _HIVEMIND_AVAILABLE,
-    _HOROVOD_AVAILABLE,
-    _HPU_AVAILABLE,
-    _IPU_AVAILABLE,
-    _IS_INTERACTIVE,
-    _IS_WINDOWS,
-    _POPTORCH_AVAILABLE,
-    _TORCH_GREATER_EQUAL_1_10,
-    _TORCH_GREATER_EQUAL_1_11,
-    _TORCH_GREATER_EQUAL_1_12,
-)
 from lightning_lite.utilities.rank_zero import (  # noqa: F401
     rank_zero_deprecation,
     rank_zero_info,
diff --git a/src/lightning_lite/utilities/data.py b/src/lightning_lite/utilities/data.py
index ca50344567..afa0e988ca 100644
--- a/src/lightning_lite/utilities/data.py
+++ b/src/lightning_lite/utilities/data.py
@@ -175,11 +175,7 @@ def _dataloader_init_kwargs_resolve_sampler(
     disallow_batch_sampler: bool = False,
 ) -> Dict[str, Any]:
     """This function is used to handle the sampler, batch_sampler arguments associated within a DataLoader for its
-    re-instantiation.
-
-    If there are multiple devices in IPU mode, it is necessary to disallow BatchSampler that isn't instantiated
-    automatically, since `poptorch.DataLoader` will try to increase the batch_size
-    """
+    re-instantiation."""
     batch_sampler = getattr(dataloader, "batch_sampler")
 
     if batch_sampler is not None:
diff --git a/src/lightning_lite/utilities/distributed.py b/src/lightning_lite/utilities/distributed.py
index 9e7ea0142a..767de75411 100644
--- a/src/lightning_lite/utilities/distributed.py
+++ b/src/lightning_lite/utilities/distributed.py
@@ -4,11 +4,11 @@ from typing import Any, Iterable, Iterator, List, Optional, Sized, Tuple, Union
 
 import torch
 import torch.nn.functional as F
+from lightning_utilities.core.imports import module_available
 from torch import Tensor
 from torch.utils.data import Dataset, DistributedSampler, Sampler
 
 from lightning_lite.plugins.environments.cluster_environment import ClusterEnvironment
-from lightning_lite.utilities.imports import _HPU_AVAILABLE
 from lightning_lite.utilities.rank_zero import rank_zero_info
 from lightning_lite.utilities.types import ReduceOp
 
@@ -135,12 +135,16 @@ def sync_ddp(result: Tensor, group: Optional[Any] = None, reduce_op: Optional[Un
         op = reduce_op
 
     # WA for HPU. HPU doesn't support Long types, forcefully set it to float
-    if _HPU_AVAILABLE:
-        is_hpu_backend = os.environ.get("HCCL_DISTRIBUTED_BACKEND") == "1"
-        if is_hpu_backend:
-            if (result.type() == "torch.LongTensor") or (result.type() == "torch.hpu.LongTensor"):
-                rank_zero_info("Long tensor unsupported on HPU, casting to float")
-                result = result.float()
+    if module_available("habana_frameworks.torch.utils.library_loader"):
+        from habana_frameworks.torch.utils.library_loader import is_habana_available
+
+        if (
+            is_habana_available()
+            and os.environ.get("HCCL_DISTRIBUTED_BACKEND") == "1"
+            and result.type() in ("torch.LongTensor", "torch.hpu.LongTensor")
+        ):
+            rank_zero_info("Long tensor unsupported on HPU, casting to float")
+            result = result.float()
 
     # Sync all processes before reduction
     torch.distributed.barrier(group=group)
diff --git a/src/lightning_lite/utilities/enums.py b/src/lightning_lite/utilities/enums.py
index 567483b1e5..c1bb015010 100644
--- a/src/lightning_lite/utilities/enums.py
+++ b/src/lightning_lite/utilities/enums.py
@@ -61,21 +61,16 @@ class _StrategyType(LightningEnum):
     DDP = "ddp"
     DDP_SPAWN = "ddp_spawn"
     DDP_FORK = "ddp_fork"
-    TPU_SPAWN = "tpu_spawn"
     DEEPSPEED = "deepspeed"
-    HOROVOD = "horovod"
     DDP_SHARDED = "ddp_sharded"
     DDP_SHARDED_SPAWN = "ddp_sharded_spawn"
     DDP_FULLY_SHARDED = "ddp_fully_sharded"
-    BAGUA = "bagua"
-    HPU_PARALLEL = "hpu_parallel"
 
     @staticmethod
     def interactive_compatible_types() -> list[_StrategyType]:
         """Returns a list containing interactive compatible _StrategyTypes."""
         return [
             _StrategyType.DP,
-            _StrategyType.TPU_SPAWN,
             _StrategyType.DDP_FORK,
         ]
 
@@ -89,7 +84,5 @@ class _AcceleratorType(LightningEnum):
 
     CPU = "CPU"
     CUDA = "CUDA"
-    IPU = "IPU"
     TPU = "TPU"
-    HPU = "HPU"
     MPS = "MPS"
diff --git a/src/lightning_lite/utilities/imports.py b/src/lightning_lite/utilities/imports.py
index 737d1d7a4a..5ff3d656c7 100644
--- a/src/lightning_lite/utilities/imports.py
+++ b/src/lightning_lite/utilities/imports.py
@@ -16,7 +16,7 @@ import operator
 import platform
 import sys
 
-from lightning_utilities.core.imports import compare_version, module_available, package_available
+from lightning_utilities.core.imports import compare_version
 
 _IS_WINDOWS = platform.system() == "Windows"
 _IS_INTERACTIVE = hasattr(sys, "ps1")  # https://stackoverflow.com/a/64523765
@@ -28,24 +28,3 @@ _TORCH_LESSER_EQUAL_1_10_2 = compare_version("torch", operator.le, "1.10.2")
 _TORCH_GREATER_EQUAL_1_11 = compare_version("torch", operator.ge, "1.11.0")
 _TORCH_GREATER_EQUAL_1_12 = compare_version("torch", operator.ge, "1.12.0")
 _TORCH_GREATER_EQUAL_1_13 = compare_version("torch", operator.ge, "1.13.0", use_base_version=True)
-
-_APEX_AVAILABLE = module_available("apex.amp")
-_HABANA_FRAMEWORK_AVAILABLE = package_available("habana_frameworks")
-_HIVEMIND_AVAILABLE = package_available("hivemind")
-_HOROVOD_AVAILABLE = module_available("horovod.torch")
-_OMEGACONF_AVAILABLE = package_available("omegaconf")
-_POPTORCH_AVAILABLE = package_available("poptorch")
-
-if _POPTORCH_AVAILABLE:
-    import poptorch
-
-    _IPU_AVAILABLE = poptorch.ipuHardwareIsAvailable()
-else:
-    _IPU_AVAILABLE = False
-
-if _HABANA_FRAMEWORK_AVAILABLE:
-    from habana_frameworks.torch.utils.library_loader import is_habana_avaialble
-
-    _HPU_AVAILABLE = is_habana_avaialble()
-else:
-    _HPU_AVAILABLE = False
diff --git a/tests/tests_lite/plugins/precision/test_deepspeed.py b/tests/tests_lite/plugins/precision/test_deepspeed.py
index a0cee88b26..3a06956c31 100644
--- a/tests/tests_lite/plugins/precision/test_deepspeed.py
+++ b/tests/tests_lite/plugins/precision/test_deepspeed.py
@@ -27,9 +27,9 @@ def test_invalid_precision_with_deepspeed_precision():
 
 
 def test_deepspeed_precision_apex_not_installed(monkeypatch):
-    import lightning_lite.plugins.precision.deepspeed as deepspeed_apex
+    import lightning_lite.plugins.precision.deepspeed as deepspeed
 
-    monkeypatch.setattr(deepspeed_apex, "_APEX_AVAILABLE", False)
+    monkeypatch.setattr(deepspeed, "_APEX_AVAILABLE", False)
     with pytest.raises(ImportError, match="You have asked for Apex AMP but `apex` is not installed."):
         DeepSpeedPrecision(precision=16, amp_type="apex")
 
diff --git a/tests/tests_lite/test_connector.py b/tests/tests_lite/test_connector.py
index 88fa081b0f..46ef1ad25f 100644
--- a/tests/tests_lite/test_connector.py
+++ b/tests/tests_lite/test_connector.py
@@ -221,7 +221,7 @@ def test_dist_backend_accelerator_mapping(*_):
 @RunIf(mps=False)
 @mock.patch("lightning_lite.accelerators.cuda.num_cuda_devices", return_value=2)
 def test_ipython_incompatible_backend_error(_, monkeypatch):
-    monkeypatch.setattr(lightning_lite.utilities, "_IS_INTERACTIVE", True)
+    monkeypatch.setattr(lightning_lite.connector, "_IS_INTERACTIVE", True)
     with pytest.raises(RuntimeError, match=r"strategy='ddp'\)`.*is not compatible"):
         _Connector(strategy="ddp", accelerator="gpu", devices=2)
 
@@ -238,21 +238,21 @@ def test_ipython_incompatible_backend_error(_, monkeypatch):
 
 @mock.patch("lightning_lite.accelerators.cuda.num_cuda_devices", return_value=2)
 def test_ipython_compatible_dp_strategy_gpu(_, monkeypatch):
-    monkeypatch.setattr(lightning_lite.utilities, "_IS_INTERACTIVE", True)
+    monkeypatch.setattr(lightning_lite.utilities.imports, "_IS_INTERACTIVE", True)
     connector = _Connector(strategy="dp", accelerator="gpu")
     assert connector.strategy.launcher is None
 
 
 @RunIf(skip_windows=True)
 def test_ipython_compatible_strategy_tpu(tpu_available, monkeypatch):
-    monkeypatch.setattr(lightning_lite.utilities, "_IS_INTERACTIVE", True)
+    monkeypatch.setattr(lightning_lite.utilities.imports, "_IS_INTERACTIVE", True)
     connector = _Connector(accelerator="tpu")
     assert connector.strategy.launcher.is_interactive_compatible
 
 
 @RunIf(skip_windows=True)
 def test_ipython_compatible_strategy_ddp_fork(monkeypatch):
-    monkeypatch.setattr(lightning_lite.utilities, "_IS_INTERACTIVE", True)
+    monkeypatch.setattr(lightning_lite.utilities.imports, "_IS_INTERACTIVE", True)
     connector = _Connector(strategy="ddp_fork", accelerator="cpu")
     assert connector.strategy.launcher.is_interactive_compatible
 
diff --git a/tests/tests_lite/utilities/test_imports.py b/tests/tests_lite/utilities/test_imports.py
index 1b48e626ce..b9812516c1 100644
--- a/tests/tests_lite/utilities/test_imports.py
+++ b/tests/tests_lite/utilities/test_imports.py
@@ -13,22 +13,9 @@
 # limitations under the License.
 from lightning_lite.strategies.deepspeed import _DEEPSPEED_AVAILABLE
 from lightning_lite.strategies.fairscale import _FAIRSCALE_AVAILABLE
-from lightning_lite.utilities.imports import (
-    _APEX_AVAILABLE,
-    _HOROVOD_AVAILABLE,
-    _OMEGACONF_AVAILABLE,
-    _POPTORCH_AVAILABLE,
-)
 
 
 def test_imports():
-    try:
-        import apex  # noqa
-    except ModuleNotFoundError:
-        assert not _APEX_AVAILABLE
-    else:
-        assert _APEX_AVAILABLE
-
     try:
         import deepspeed  # noqa
     except ModuleNotFoundError:
@@ -42,24 +29,3 @@ def test_imports():
         assert not _FAIRSCALE_AVAILABLE
     else:
         assert _FAIRSCALE_AVAILABLE
-
-    try:
-        import horovod.torch  # noqa
-    except ModuleNotFoundError:
-        assert not _HOROVOD_AVAILABLE
-    else:
-        assert _HOROVOD_AVAILABLE
-
-    try:
-        import omegaconf  # noqa
-    except ModuleNotFoundError:
-        assert not _OMEGACONF_AVAILABLE
-    else:
-        assert _OMEGACONF_AVAILABLE
-
-    try:
-        import poptorch  # noqa
-    except ModuleNotFoundError:
-        assert not _POPTORCH_AVAILABLE
-    else:
-        assert _POPTORCH_AVAILABLE