Replace register_backward_hook with register_full_backward_hook (meta-pytorch#720)

Summary:

register_backward_hook is deprecated.

Differential Revision: D68562558

Author: iden-kalemaj

.github/workflows/ci_gpu.yml

@@ -68,7 +68,7 @@ jobs:
           mkdir -p runs/mnist/data
           mkdir -p runs/mnist/test-reports
           python -c "import torch; exit(0) if torch.cuda.is_available() else exit(1)"
-          python examples/mnist.py --lr 0.25 --sigma 0.7 -c 1.5 --batch-size 64 --epochs 1 --data-root runs/mnist/data --n-runs 1 --device cuda
+          python examples/mnist.py --lr 0.25 --sigma 0.7 -c 1.5 --batch-size 64 --epochs 1 --data-root runs/mnist/data --n-runs 1 --device cuda --clip_per_layer
           python -c "import torch; accuracy = torch.load('run_results_mnist_0.25_0.7_1.5_64_1.pt'); exit(0) if (accuracy[0]>0.78 and accuracy[0]<0.95) else exit(1)"
 
       - name: Store MNIST test results
@@ -83,7 +83,7 @@ jobs:
           mkdir -p runs/cifar10/logs
           mkdir -p runs/cifar10/test-reports
           pip install tensorboard
-          python examples/cifar10.py --lr 0.1 --sigma 1.5 -c 10 --batch-size 2000 --epochs 10 --data-root runs/cifar10/data --log-dir runs/cifar10/logs --device cuda
+          python examples/cifar10.py --lr 0.1 --sigma 1.5 -c 10 --batch-size 2000 --epochs 10 --data-root runs/cifar10/data --log-dir runs/cifar10/logs --device cuda --clip_per_layer
           python -c "import torch; model = torch.load('model_best.pth.tar'); exit(0) if (model['best_acc1']>0.4 and model['best_acc1']<0.49) else exit(1)"
           python examples/cifar10.py --lr 0.1 --sigma 1.5 -c 10 --batch-size 2000 --epochs 10 --data-root runs/cifar10/data --log-dir runs/cifar10/logs --device cuda --grad_sample_mode no_op
           python -c "import torch; model = torch.load('model_best.pth.tar'); exit(0) if (model['best_acc1']>0.4 and model['best_acc1']<0.49) else exit(1)"

opacus/grad_sample/grad_sample_module.py

@@ -207,7 +207,7 @@ def add_hooks(
             )
 
             self.autograd_grad_sample_hooks.append(
-                module.register_backward_hook(
+                module.register_full_backward_hook(
                     partial(
                         self.capture_backprops_hook,
                         loss_reduction=loss_reduction,

opacus/tests/multigpu_gradcheck.py

@@ -206,6 +206,7 @@ def test_gradient_correct(self) -> None:
         )
         clipping_grad_sample_pairs.append(("ghost", "ghost"))
 
+        clipping_grad_sample_pairs = [("per_layer", "hooks")]
         for clipping, grad_sample_mode in clipping_grad_sample_pairs:
 
             weight_dp, weight_nodp = torch.zeros(10, 10), torch.zeros(10, 10)