🐛 Bug

When using ModelPruning callback with verbosity level in {1,2}, the logged sparsity would be wrong when there are layers with multiple parameters to be pruned.
For example, when we add a ModelPruning(amount=0.2) callback to the BoringModel Colab link, we would get overall sparsity logged around 0.1 instead of 0.2:

Applied `L1Unstructured`. Pruned: 0/132 (0.00%) -> 13/132 (9.85%)
Applied `L1Unstructured` to `Linear(in_features=32, out_features=2, bias=True).weight` with amount=0.2. Pruned: 0 (0.00%) -> 13 (20.31%)
Applied `L1Unstructured` to `Linear(in_features=32, out_features=2, bias=True).bias` with amount=0.2. Pruned: 0 (0.00%) -> 0 (0.00%)

The problem is mainly due to L346 in the following code, where there might have layers counted multiple times:
https://github.com/Lightning-AI/lightning/blob/b59f80224843886459d54c828325683d770da746/src/pytorch_lightning/callbacks/pruning.py#L343-L353
Can be fixed easily by:

# count from prev/curr should be the same
total_params = sum(params for _, params in prev)

To Reproduce

https://gist.github.com/SungFeng-Huang/52d676869ad4e8a4a00ac3e29437ecdd

Expected behavior

Get sparsity logged around 0.2 instead of 0.1:

Applied `L1Unstructured`. Pruned: 0/132 (0.00%) -> 13/66 (19.70%)
Applied `L1Unstructured` to `Linear(in_features=32, out_features=2, bias=True).weight` with amount=0.2. Pruned: 0 (0.00%) -> 13 (20.31%)
Applied `L1Unstructured` to `Linear(in_features=32, out_features=2, bias=True).bias` with amount=0.2. Pruned: 0 (0.00%) -> 0 (0.00%)

Environment

• CUDA:
  • GPU:
    • Tesla T4
  • available: True
  • version: 11.3
• Packages:
  • numpy: 1.21.6
  • pyTorch_debug: False
  • pyTorch_version: 1.11.0+cu113
  • pytorch-lightning: 1.6.4
  • tqdm: 4.64.0
• System:
  • OS: Linux
  • architecture:
    • 64bit
  • processor: x86_64
  • python: 3.7.13
  • version: Proposal for help #1 SMP Sun Apr 24 10:03:06 PDT 2022

Additional context

cc @tchaton @carmocca