[RFC] Deprecate `on_configure_sharded_model` callback hook

[daniellepintz]

Proposed refactor

Deprecate on_configure_sharded_model callback/LM hooks in v1.6 for removal in v1.8
https://github.com/PyTorchLightning/pytorch-lightning/blob/5ad5ba54c0c477546d21daf75ac7b4748d7963a7/pytorch_lightning/callbacks/base.py#L60-L61

Motivation

I see no uses of this hook either externally or internally: https://www.grep.app/search?q=on_configure_sharded_model

Additional context

This hook was originally added in #6679

cc @justusschock @awaelchli @akihironitta @rohitgr7 @tchaton @carmocca @Borda @ninginthecloud @daniellepintz

[tchaton]

@SeanNaren I believe it has never been properly advertised. I think it would be fair to remove it as I don't really find a real use case for it.

[carmocca]

I agree with removal, the hook is really specific. Using it doesn't paint a good picture design-wise so I think it'll be good to clean up, now that it's not necessary.

[daniellepintz]

For anyone picking up this issue, you can follow #10940 as an example

[Piyush-97]

I will try to do this.

[awaelchli]

Yes!

[Piyush-97]

I don't know how to write the test for this.

[rohitgr7]

hey @Piyush-97
for the test you need to ensure that the deprecation warning is caught when someone uses this hook.
check this for reference.

[Piyush-97]

`def test_v1_8_0_on_configure_sharded_model(tmpdir):
class TestCallback(Callback):
def on_configure_sharded_model(self, trainer, model):
print("Configuring sharded model")

model = BoringModel()

trainer = Trainer(
    callbacks=[TestCallback()],
    max_epochs=1,
    fast_dev_run=True,
    enable_progress_bar=False,
    logger=False,
    default_root_dir=tmpdir,
    strategy='ddp_sharded'
)
with pytest.deprecated_call(
    match="The `on_configure_sharded_model` callback hook was deprecated in v1.6 and will be removed in v1.8."
):
    trainer.fit(model)`

I don't have GPU and I am not sure if this is right.