Merge branch 'master' into ci/clear-cache

Author: akihironitta

.azure/hpu-tests.yml

@@ -39,7 +39,7 @@ jobs:
     cancelTimeoutInMinutes: "2"
     pool: intel-hpus
     container:
-      image: "vault.habana.ai/gaudi-docker/1.6.0/ubuntu20.04/habanalabs/pytorch-installer-1.12.0:latest"
+      image: "vault.habana.ai/gaudi-docker/1.6.1/ubuntu20.04/habanalabs/pytorch-installer-1.12.0:latest"
       options: "--runtime=habana -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --ipc=host --shm-size=4g --name cd-container -v /usr/bin/docker:/tmp/docker:ro"
     workspace:
       clean: all

dockers/ci-runner-hpu/Dockerfile

@@ -16,7 +16,7 @@
 #        gaudi-docker-agent:latest
 
 ARG DIST="latest"
-ARG GAUDI_VERSION="1.6.0"
+ARG GAUDI_VERSION="1.6.1"
 ARG PYTORCH_INSTALLER_VERSION="1.12.0"
 FROM vault.habana.ai/gaudi-docker/${GAUDI_VERSION}/ubuntu20.04/habanalabs/pytorch-installer-${PYTORCH_INSTALLER_VERSION}:${DIST}
 

pyproject.toml

@@ -59,7 +59,6 @@ module = [
     "pytorch_lightning.callbacks.progress.rich_progress",
     "pytorch_lightning.trainer.trainer",
     "pytorch_lightning.trainer.connectors.checkpoint_connector",
-    "pytorch_lightning.tuner.batch_size_scaling",
     "lightning_app.api.http_methods",
     "lightning_app.api.request_types",
     "lightning_app.cli.app-template.app",

src/pytorch_lightning/CHANGELOG.md

@@ -222,6 +222,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Removed the deprecated device attributes `Trainer.{devices,gpus,num_gpus,ipus,tpu_cores}` in favor of the accelerator-agnostic `Trainer.num_devices` ([#14829](https://github.com/Lightning-AI/lightning/pull/14829))
 
 
+- Removed the deprecated `LightningIPUModule` ([#14830](https://github.com/Lightning-AI/lightning/pull/14830))
+
+
 - Removed the deprecated `Logger.agg_and_log_metrics` hook in favour of `Logger.log_metrics` and the `agg_key_funcs` and `agg_default_func` arguments. ([#14840](https://github.com/Lightning-AI/lightning/pull/14840))
 
 

src/pytorch_lightning/callbacks/batch_size_finder.py

@@ -31,6 +31,8 @@
 class BatchSizeFinder(Callback):
     SUPPORTED_MODES = ("power", "binsearch")
 
+    optimal_batch_size: Optional[int]
+
     def __init__(
         self,
         mode: str = "power",

src/pytorch_lightning/strategies/ipu.py

@@ -24,7 +24,7 @@
 from lightning_lite.plugins import CheckpointIO, ClusterEnvironment
 from lightning_lite.plugins.precision.utils import _fp_to_half
 from lightning_lite.utilities.cloud_io import get_filesystem
-from pytorch_lightning.overrides.base import _LightningModuleWrapperBase, _LightningPrecisionModuleWrapperBase
+from pytorch_lightning.overrides.base import _LightningModuleWrapperBase
 from pytorch_lightning.plugins.precision import PrecisionPlugin
 from pytorch_lightning.strategies.parallel import ParallelStrategy
 from pytorch_lightning.strategies.strategy import TBroadcast
@@ -33,7 +33,6 @@
 from pytorch_lightning.utilities.data import _get_dataloader_init_args_and_kwargs, _reinstantiate_wrapped_cls
 from pytorch_lightning.utilities.exceptions import MisconfigurationException
 from pytorch_lightning.utilities.model_helpers import is_overridden
-from pytorch_lightning.utilities.rank_zero import rank_zero_deprecation
 from pytorch_lightning.utilities.types import STEP_OUTPUT
 
 if _POPTORCH_AVAILABLE:
@@ -42,28 +41,6 @@
     poptorch = None
 
 
-class LightningIPUModule(_LightningModuleWrapperBase):
-    """
-    .. deprecated:: v1.7.0
-        ``LightningIPUModule`` has been deprecated in v1.7.0 and will be removed in v1.9.0.
-    """
-
-    def __init__(
-        self,
-        forward_module: Optional[Union["pl.LightningModule", _LightningPrecisionModuleWrapperBase]] = None,
-        precision: Union[str, int] = 32,
-        pl_module: Optional[Union["pl.LightningModule", _LightningPrecisionModuleWrapperBase]] = None,
-    ) -> None:
-        rank_zero_deprecation("`LightningIPUModule` has been deprecated in v1.7.0 and will be removed in v1.8.0")
-        self._validate_init_arguments(pl_module, forward_module)
-        super().__init__(forward_module=(pl_module or forward_module))
-        self.precision = precision
-
-    def forward(self, *inputs: Any, **kwargs: Any) -> Any:
-        inputs = apply_to_collection(inputs, Tensor, function=_fp_to_half, precision=self.precision)
-        return super().forward(*inputs, **kwargs)
-
-
 class IPUStrategy(ParallelStrategy):
     """Plugin for training on IPU devices."""
 

src/pytorch_lightning/tuner/batch_size_scaling.py

@@ -35,10 +35,10 @@ def scale_batch_size(
     init_val: int = 2,
     max_trials: int = 25,
     batch_arg_name: str = "batch_size",
-):
+) -> Optional[int]:
     if trainer.fast_dev_run:
         rank_zero_warn("Skipping batch size scaler since `fast_dev_run` is enabled.")
-        return
+        return None
 
     # Save initial model, that is loaded after batch size is found
     ckpt_path = os.path.join(trainer.default_root_dir, f".scale_batch_size_{uuid.uuid4()}.ckpt")
@@ -141,7 +141,12 @@ def __scale_batch_restore_params(trainer: "pl.Trainer", params: Dict[str, Any])
 
 
 def _run_power_scaling(
-    trainer: "pl.Trainer", pl_module: "pl.LightningModule", new_size: int, batch_arg_name: str, max_trials: int, params
+    trainer: "pl.Trainer",
+    pl_module: "pl.LightningModule",
+    new_size: int,
+    batch_arg_name: str,
+    max_trials: int,
+    params: Dict[str, Any],
 ) -> int:
     """Batch scaling mode where the size is doubled at each iteration until an OOM error is encountered."""
     # this flag is used to determine whether the previously scaled batch size, right before OOM, was a success or not
@@ -179,7 +184,12 @@ def _run_power_scaling(
 
 
 def _run_binary_scaling(
-    trainer: "pl.Trainer", pl_module: "pl.LightningModule", new_size: int, batch_arg_name: str, max_trials: int, params
+    trainer: "pl.Trainer",
+    pl_module: "pl.LightningModule",
+    new_size: int,
+    batch_arg_name: str,
+    max_trials: int,
+    params: Dict[str, Any],
 ) -> int:
     """Batch scaling mode where the size is initially is doubled at each iteration until an OOM error is
     encountered.
@@ -309,7 +319,7 @@ def _reset_dataloaders(trainer: "pl.Trainer", pl_module: "pl.LightningModule") -
         reset_fn(pl_module)
 
 
-def _try_loop_run(trainer: "pl.Trainer", params) -> None:
+def _try_loop_run(trainer: "pl.Trainer", params: Dict[str, Any]) -> None:
     if trainer.state.fn == "fit":
         loop = trainer.fit_loop
     else:

tests/tests_pytorch/deprecated_api/test_remove_1-10.py

@@ -33,7 +33,6 @@
 from pytorch_lightning.plugins.environments import LightningEnvironment
 from pytorch_lightning.strategies.bagua import LightningBaguaModule
 from pytorch_lightning.strategies.deepspeed import LightningDeepSpeedModule
-from pytorch_lightning.strategies.ipu import LightningIPUModule
 from pytorch_lightning.strategies.utils import on_colab_kaggle
 from pytorch_lightning.utilities.apply_func import (
     apply_to_collection,
@@ -85,7 +84,6 @@ def test_deprecated_amp_level():
         LightningBaguaModule,
         LightningDeepSpeedModule,
         pytest.param(LightningShardedDataParallel, marks=RunIf(fairscale=True)),
-        LightningIPUModule,
     ],
 )
 def test_v1_10_deprecated_pl_module_init_parameter(wrapper_class):

tests/tests_pytorch/deprecated_api/test_remove_1-8.py

@@ -20,7 +20,6 @@
 from pytorch_lightning import Callback, Trainer
 from pytorch_lightning.callbacks import ModelCheckpoint
 from pytorch_lightning.demos.boring_classes import BoringDataModule, BoringModel
-from pytorch_lightning.strategies.ipu import LightningIPUModule
 from pytorch_lightning.trainer.configuration_validator import _check_datamodule_checkpoint_hooks
 
 
@@ -257,11 +256,6 @@ def on_load_checkpoint(self, checkpoint):
         _check_datamodule_checkpoint_hooks(trainer)
 
 
-def test_v1_8_0_deprecated_lightning_ipu_module():
-    with pytest.deprecated_call(match=r"has been deprecated in v1.7.0 and will be removed in v1.8."):
-        _ = LightningIPUModule(BoringModel(), 32)
-
-
 def test_deprecated_mc_save_checkpoint():
     mc = ModelCheckpoint()
     trainer = Trainer()