mark `OptimizerLoop.backward` method protected (#9514)

Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>

CHANGELOG.md

@@ -77,6 +77,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
     * Removed `TrainingBatchLoop.backward()`; manual optimization now calls directly into `Accelerator.backward()` and automatic optimization handles backward in new `OptimizerLoop` ([#9265](https://github.com/PyTorchLightning/pytorch-lightning/pull/9265))
     * Extracted `ManualOptimization` logic from `TrainingBatchLoop` into its own separate loop class ([#9266](https://github.com/PyTorchLightning/pytorch-lightning/pull/9266))
     * Added `OutputResult` and `ManualResult` classes ([#9437](https://github.com/PyTorchLightning/pytorch-lightning/pull/9437), [#9424](https://github.com/PyTorchLightning/pytorch-lightning/pull/9424))
+    * Marked `OptimizerLoop.backward` as protected ([#9514](https://github.com/PyTorchLightning/pytorch-lightning/pull/9514))
 
 
 - Added support for saving and loading state of multiple callbacks of the same type ([#7187](https://github.com/PyTorchLightning/pytorch-lightning/pull/7187))

pytorch_lightning/loops/optimization/optimizer_loop.py

@@ -221,7 +221,7 @@ class OptimizerLoop(Loop):
         outputs, self.outputs = self.outputs, []  # free memory
         return outputs
 
-    def backward(
+    def _backward(
         self, loss: Tensor, optimizer: torch.optim.Optimizer, opt_idx: int, *args: Any, **kwargs: Any
     ) -> Tensor:
         """Performs the backward step.
@@ -337,7 +337,7 @@ class OptimizerLoop(Loop):
             return None
 
         def backward_fn(loss: Tensor) -> Tensor:
-            self.backward(loss, optimizer, opt_idx)
+            self._backward(loss, optimizer, opt_idx)
 
             # check if model weights are nan
             if self.trainer.terminate_on_nan:

tests/trainer/test_trainer.py

@@ -961,7 +961,7 @@ def test_gradient_clipping_by_norm(tmpdir, precision):
         gradient_clip_val=1.0,
     )
 
-    old_backward = trainer.fit_loop.epoch_loop.batch_loop.optimizer_loop.backward
+    old_backward = trainer.fit_loop.epoch_loop.batch_loop.optimizer_loop._backward
 
     def backward(*args, **kwargs):
         # test that gradient is clipped correctly
@@ -971,7 +971,7 @@ def test_gradient_clipping_by_norm(tmpdir, precision):
         assert (grad_norm - 1.0).abs() < 0.01, f"Gradient norm != 1.0: {grad_norm}"
         return ret_val
 
-    trainer.fit_loop.epoch_loop.batch_loop.optimizer_loop.backward = backward
+    trainer.fit_loop.epoch_loop.batch_loop.optimizer_loop._backward = backward
     trainer.fit(model)
 
 
@@ -996,7 +996,7 @@ def test_gradient_clipping_by_value(tmpdir, precision):
         default_root_dir=tmpdir,
     )
 
-    old_backward = trainer.fit_loop.epoch_loop.batch_loop.optimizer_loop.backward
+    old_backward = trainer.fit_loop.epoch_loop.batch_loop.optimizer_loop._backward
 
     def backward(*args, **kwargs):
         # test that gradient is clipped correctly
@@ -1009,7 +1009,7 @@ def test_gradient_clipping_by_value(tmpdir, precision):
         ), f"Gradient max value {grad_max} != grad_clip_val {grad_clip_val} ."
         return ret_val
 
-    trainer.fit_loop.epoch_loop.batch_loop.optimizer_loop.backward = backward
+    trainer.fit_loop.epoch_loop.batch_loop.optimizer_loop._backward = backward
     trainer.fit(model)