Introduce Stateful PrecisionPlugin (#11638)

Co-authored-by: Rohit Gupta <[email protected]>
Co-authored-by: ananthsub <[email protected]>

CHANGELOG.md

@@ -108,6 +108,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Added a `_Stateful` support for `LightningDataModule` ([#11637](https://github.com/PyTorchLightning/pytorch-lightning/pull/11637))
 
 
+- Added `_Stateful` support for `PrecisionPlugin` ([#11638](https://github.com/PyTorchLightning/pytorch-lightning/pull/11638))
+
+
 - Added `Accelerator.is_available` to check device availability ([#11797](https://github.com/PyTorchLightning/pytorch-lightning/pull/11797))
 
 

pytorch_lightning/plugins/precision/apex_amp.py

@@ -93,9 +93,21 @@ def optimizer_step(
             return optimizer.step(**kwargs)
         return closure_result
 
+    def state_dict(self) -> Dict[str, Any]:
+        return amp.state_dict()
+
+    def load_state_dict(self, state_dict: Dict[str, Any]) -> None:
+        amp.load_state_dict(state_dict)
+
     def on_load_checkpoint(self, checkpoint: Dict[str, Any]) -> None:
-        if "amp_scaling_state" in checkpoint:
-            amp.load_state_dict(checkpoint["amp_scaling_state"])
+        """``ApexMixedPrecisionPlugin.on_load_checkpoint`` is deprecated in v1.6.
+
+        Lightning will auto-restore ApexMixedPrecisionPlugin state with ``ApexMixedPrecisionPlugin.load_state_dict``
+        instead
+        """
 
     def on_save_checkpoint(self, checkpoint: Dict[str, Any]) -> None:
-        checkpoint["amp_scaling_state"] = amp.state_dict()
+        """``ApexMixedPrecisionPlugin.on_save_checkpoint`` is deprecated in v1.6.
+
+        Lightning will auto-save ApexMixedPrecisionPlugin state with ``ApexMixedPrecisionPlugin.state_dict`` instead
+        """

pytorch_lightning/plugins/precision/native_amp.py

@@ -108,10 +108,24 @@ def forward_context(self) -> Generator[None, None, None]:
         with self.autocast_context_manager():
             yield
 
+    def state_dict(self) -> Dict[str, Any]:
+        if self.scaler is not None:
+            return self.scaler.state_dict()
+        return {}
+
+    def load_state_dict(self, state_dict: Dict[str, Any]) -> None:
+        if self.scaler is not None:
+            self.scaler.load_state_dict(state_dict)
+
     def on_load_checkpoint(self, checkpoint: Dict[str, Any]) -> None:
-        if self.scaler is not None and "native_amp_scaling_state" in checkpoint:
-            self.scaler.load_state_dict(checkpoint["native_amp_scaling_state"])
+        """``NativeMixedPrecisionPlugin.on_load_checkpoint`` is deprecated in v1.6.
+
+        Lightning will auto-restore NativeMixedPrecisionPlugin state with ``NativeMixedPrecisionPlugin.load_state_dict``
+        instead
+        """
 
     def on_save_checkpoint(self, checkpoint: Dict[str, Any]) -> None:
-        if self.scaler is not None:
-            checkpoint["native_amp_scaling_state"] = self.scaler.state_dict()
+        """``NativeMixedPrecisionPlugin.on_save_checkpoint`` is deprecated in v1.6.
+
+        Lightning will auto-save NativeMixedPrecisionPlugin state with ``NativeMixedPrecisionPlugin.state_dict`` instead
+        """

pytorch_lightning/plugins/precision/precision_plugin.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 import contextlib
 from functools import partial
-from typing import Any, Callable, Generator, List, Optional, Tuple, Union
+from typing import Any, Callable, Dict, Generator, List, Optional, Tuple, Union
 
 import torch
 from torch import Tensor
@@ -242,3 +242,20 @@ def teardown(self) -> None:
 
         It is the right place to release memory and free other resources.
         """
+
+    def state_dict(self) -> Dict[str, Any]:
+        """Called when saving a checkpoint, implement to generate precision plugin state_dict.
+
+        Returns:
+            A dictionary containing precision plugin state.
+        """
+        return {}
+
+    def load_state_dict(self, state_dict: Dict[str, Any]) -> None:
+        """Called when loading a checkpoint, implement to reload precision plugin state given precision plugin
+        state_dict.
+
+        Args:
+            state_dict: the precision plugin state returned by ``state_dict``.
+        """
+        pass

pytorch_lightning/trainer/connectors/checkpoint_connector.py

@@ -22,6 +22,7 @@
 
 import pytorch_lightning as pl
 from pytorch_lightning.plugins.environments import SLURMEnvironment
+from pytorch_lightning.plugins.precision import ApexMixedPrecisionPlugin, NativeMixedPrecisionPlugin
 from pytorch_lightning.trainer.states import TrainerFn
 from pytorch_lightning.utilities import _OMEGACONF_AVAILABLE
 from pytorch_lightning.utilities.cloud_io import get_filesystem
@@ -196,7 +197,7 @@ def restore_training_state(self) -> None:
             return
 
         # restore precision plugin (scaler etc.)
-        self.trainer.precision_plugin.on_load_checkpoint(self._loaded_checkpoint)
+        self.restore_precision_plugin_state()
 
         # restore loops and their progress
         self.restore_loops()
@@ -206,6 +207,21 @@ def restore_training_state(self) -> None:
             # restore optimizers and schedulers state
             self.restore_optimizers_and_schedulers()
 
+    def restore_precision_plugin_state(self) -> None:
+        """Restore the precision plugin state from the pre-loaded checkpoint."""
+        prec_plugin = self.trainer.precision_plugin
+        prec_plugin.on_load_checkpoint(self._loaded_checkpoint)
+        if prec_plugin.__class__.__qualname__ in self._loaded_checkpoint:
+            prec_plugin.load_state_dict(self._loaded_checkpoint[prec_plugin.__class__.__qualname__])
+
+        # old checkpoints compatibility
+        if "amp_scaling_state" in self._loaded_checkpoint and isinstance(prec_plugin, ApexMixedPrecisionPlugin):
+            prec_plugin.load_state_dict(self._loaded_checkpoint["amp_scaling_state"])
+        if "native_amp_scaling_state" in self._loaded_checkpoint and isinstance(
+            prec_plugin, NativeMixedPrecisionPlugin
+        ):
+            prec_plugin.load_state_dict(self._loaded_checkpoint["native_amp_scaling_state"])
+
     def restore_callbacks(self) -> None:
         """Restores all callbacks from the pre-loaded checkpoint."""
         if not self._loaded_checkpoint:
@@ -318,9 +334,8 @@ def dump_checkpoint(self, weights_only: bool = False) -> dict:
                 'callbacks':                 "callback specific state"[] # if not weights_only
                 'optimizer_states':          "PT optim's state_dict"[]   # if not weights_only
                 'lr_schedulers':             "PT sched's state_dict"[]   # if not weights_only
-                'native_amp_scaling_state':  PT amp's state_dict         # if not weights_only and use native amp
-                'amp_scaling_state':         Apex's state_dict           # if not weights_only and use apex amp
                 'state_dict':                Model's state_dict (e.g. network weights)
+                precision_plugin.__class__.__qualname__:  precision plugin state_dict # if not weights_only
                 CHECKPOINT_HYPER_PARAMS_NAME:
                 CHECKPOINT_HYPER_PARAMS_KEY:
                 CHECKPOINT_HYPER_PARAMS_TYPE:
@@ -357,7 +372,12 @@ def dump_checkpoint(self, weights_only: bool = False) -> dict:
                 lr_schedulers.append(config.scheduler.state_dict())
             checkpoint["lr_schedulers"] = lr_schedulers
 
-            self.trainer.precision_plugin.on_save_checkpoint(checkpoint)
+            # precision plugin
+            prec_plugin = self.trainer.precision_plugin
+            prec_plugin_state_dict = prec_plugin.state_dict()
+            if prec_plugin_state_dict:
+                checkpoint[prec_plugin.__class__.__qualname__] = prec_plugin_state_dict
+            prec_plugin.on_save_checkpoint(checkpoint)
 
         # dump hyper-parameters
         if model.hparams:

tests/models/test_hooks.py

@@ -493,10 +493,8 @@ def training_step(self, batch, batch_idx):
         "state_dict": ANY,
         "loops": ANY,
     }
-    if kwargs.get("amp_backend") == "native":
-        saved_ckpt["native_amp_scaling_state"] = ANY
-    elif kwargs.get("amp_backend") == "apex":
-        saved_ckpt["amp_scaling_state"] = ANY
+    if kwargs.get("amp_backend") == "native" or kwargs.get("amp_backend") == "apex":
+        saved_ckpt[trainer.precision_plugin.__class__.__qualname__] = ANY
     device = torch.device("cuda:0" if "gpus" in kwargs else "cpu")
     expected = [
         dict(name="Callback.on_init_start", args=(trainer,)),