Fix loss function buffer support (#1203)

* Fix loss function buffer support

* Update CHANGELOG.md

CHANGELOG.md

@@ -22,6 +22,8 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 - Fixed DDP support for `VideoClassifier` ([#1189](https://github.com/PyTorchLightning/lightning-flash/pull/1189))
 
+- Fixed a bug where buffers in loss functions were not correctly registered in the `Task` ([#1203](https://github.com/PyTorchLightning/lightning-flash/pull/1203))
+
 ## [0.7.0] - 2022-02-15
 
 ### Added

flash/core/utilities/apply_func.py

@@ -13,12 +13,16 @@
 # limitations under the License.
 from typing import Callable, Dict, Mapping, Sequence, Type, Union
 
+from torch import nn
+
 
 def get_callable_name(fn_or_class: Union[Callable, object]) -> str:
     return getattr(fn_or_class, "__name__", fn_or_class.__class__.__name__).lower()
 
 
-def get_callable_dict(fn: Union[Callable, Mapping, Sequence]) -> Union[Dict, Mapping]:
+def get_callable_dict(fn: Union[nn.Module, Callable, Mapping, Sequence]) -> Union[Dict, Mapping]:
+    if isinstance(fn, nn.Module):
+        return nn.ModuleDict({get_callable_name(fn): fn})
     if isinstance(fn, Mapping):
         return fn
     if isinstance(fn, Sequence):

tests/core/test_model.py

@@ -488,3 +488,12 @@ def on_test_end(self, trainer: "pl.Trainer", pl_module: "pl.LightningModule") ->
     trainer = flash.Trainer(max_epochs=1, callbacks=CheckAccuracy(), gpus=torch.cuda.device_count())
     trainer.fit(task, train_dataloader=DataLoader(train_dataset), val_dataloaders=DataLoader(val_dataset))
     trainer.test(task, DataLoader(test_dataset))
+
+
+def test_loss_fn_buffer():
+    weight = torch.rand(10)
+    model = Task(loss_fn=nn.CrossEntropyLoss(weight=weight))
+    state_dict = model.state_dict()
+
+    assert len(state_dict) == 1
+    assert torch.allclose(state_dict["loss_fn.crossentropyloss.weight"], weight)