Revert "Revert "[AIR] Deprecations for 2.3""

python/ray/air/checkpoint.py

@@ -27,7 +27,7 @@
     upload_to_uri,
 )
 from ray.air.constants import PREPROCESSOR_KEY, CHECKPOINT_ID_ATTR
-from ray.util.annotations import Deprecated, DeveloperAPI, PublicAPI
+from ray.util.annotations import DeveloperAPI, PublicAPI
 
 if TYPE_CHECKING:
     from ray.data.preprocessor import Preprocessor
@@ -415,42 +415,6 @@ def to_dict(self) -> dict:
             checkpoint_data[PREPROCESSOR_KEY] = self._override_preprocessor
         return checkpoint_data
 
-    @classmethod
-    @Deprecated(
-        message="To restore a checkpoint from a remote object ref, call "
-        "`ray.get(obj_ref)` instead."
-    )
-    def from_object_ref(cls, obj_ref: ray.ObjectRef) -> "Checkpoint":
-        """Create checkpoint object from object reference.
-
-        Args:
-            obj_ref: ObjectRef pointing to checkpoint data.
-
-        Returns:
-            Checkpoint: checkpoint object.
-        """
-        raise DeprecationWarning(
-            "`from_object_ref` is deprecated and will be removed in a future Ray "
-            "version. To restore a Checkpoint from a remote object ref, call "
-            "`ray.get(obj_ref)` instead.",
-        )
-
-    @Deprecated(
-        message="To store the checkpoint in the Ray object store, call `ray.put(ckpt)` "
-        "instead of `ckpt.to_object_ref()`."
-    )
-    def to_object_ref(self) -> ray.ObjectRef:
-        """Return checkpoint data as object reference.
-
-        Returns:
-            ray.ObjectRef: ObjectRef pointing to checkpoint data.
-        """
-        raise DeprecationWarning(
-            "`to_object_ref` is deprecated and will be removed in a future Ray "
-            "version. To store the checkpoint in the Ray object store, call "
-            "`ray.put(ckpt)` instead of `ckpt.to_object_ref()`.",
-        )
-
     @classmethod
     def from_directory(cls, path: Union[str, os.PathLike]) -> "Checkpoint":
         """Create checkpoint object from directory.

python/ray/data/preprocessors/batch_mapper.py

@@ -80,23 +80,19 @@ def __init__(
                 Union[np.ndarray, Dict[str, np.ndarray]],
             ],
         ],
-        batch_format: Optional[BatchFormat] = None,
+        batch_format: Optional[BatchFormat],
         batch_size: Optional[Union[int, Literal["default"]]] = "default",
-        # TODO: Make batch_format required from user
         # TODO: Introduce a "zero_copy" format
         # TODO: We should reach consistency of args between BatchMapper and map_batches.
     ):
-        if not batch_format:
-            raise DeprecationWarning(
-                "batch_format is a required argument for BatchMapper from Ray 2.1."
-                "You must specify either 'pandas' or 'numpy' batch format."
-            )
 
         if batch_format not in [
             BatchFormat.PANDAS,
             BatchFormat.NUMPY,
         ]:
-            raise ValueError("BatchMapper only supports pandas and numpy batch format.")
+            raise ValueError(
+                "BatchMapper only supports 'pandas' or 'numpy' batch format."
+            )
 
         self.batch_format = batch_format
         self.batch_size = batch_size

python/ray/train/__init__.py

@@ -1,29 +1,13 @@
 from ray._private.usage import usage_lib
 from ray.train.backend import BackendConfig
 from ray.train.constants import TRAIN_DATASET_KEY
-from ray.train.train_loop_utils import (
-    get_dataset_shard,
-    load_checkpoint,
-    local_rank,
-    report,
-    save_checkpoint,
-    world_rank,
-    world_size,
-)
 from ray.train.trainer import TrainingIterator
 
 
 usage_lib.record_library_usage("train")
 
 __all__ = [
     "BackendConfig",
-    "get_dataset_shard",
-    "load_checkpoint",
-    "local_rank",
-    "report",
-    "save_checkpoint",
     "TrainingIterator",
-    "world_rank",
-    "world_size",
     "TRAIN_DATASET_KEY",
 ]

python/ray/train/tests/test_torch_trainer.py

@@ -378,16 +378,6 @@ def __getstate__(self):
     assert results.checkpoint
 
 
-def test_torch_prepare_model_deprecated():
-    model = torch.nn.Linear(1, 1)
-
-    with pytest.raises(DeprecationWarning):
-        train.torch.prepare_model(model, wrap_ddp=True)
-
-    with pytest.raises(DeprecationWarning):
-        train.torch.prepare_model(model, ddp_kwargs={"x": "y"})
-
-
 if __name__ == "__main__":
     import sys
 

python/ray/train/torch/train_loop_utils.py

@@ -53,9 +53,6 @@ def prepare_model(
     move_to_device: bool = True,
     parallel_strategy: Optional[str] = "ddp",
     parallel_strategy_kwargs: Optional[Dict[str, Any]] = None,
-    # Deprecated args.
-    wrap_ddp: bool = False,
-    ddp_kwargs: Optional[Dict[str, Any]] = None,
 ) -> torch.nn.Module:
     """Prepares the model for distributed execution.
 
@@ -76,18 +73,6 @@ def prepare_model(
             or "fsdp", respectively.
     """
 
-    if wrap_ddp:
-        raise DeprecationWarning(
-            "The `wrap_ddp` argument is deprecated as of Ray 2.1. Use the "
-            "`parallel_strategy` argument instead."
-        )
-
-    if ddp_kwargs:
-        raise DeprecationWarning(
-            "The `ddp_kwargs` argument is deprecated as of Ray 2.1. Use the "
-            "`parallel_strategy_kwargs` arg instead."
-        )
-
     if parallel_strategy == "fsdp" and FullyShardedDataParallel is None:
         raise ImportError(
             "FullyShardedDataParallel requires torch>=1.11.0. "