Deprecate some methods for NNCF 2.6 release (#2083)

### Changes

Some methods were left for backward compatibility after #1584 and are
due for deletion before NNCF 2.6 release.

examples/torch/common/export.py

@@ -26,9 +26,9 @@ def export_model(ctrl: CompressionAlgorithmController, save_path: str, no_strip_
     model = ctrl.model if no_strip_on_export else ctrl.strip()
 
     model = model.eval().cpu()
-    input_names = generate_input_names_list(len(model.input_infos))
+    input_names = generate_input_names_list(len(model.nncf.input_infos))
     input_tensor_list = []
-    for info in model.input_infos:
+    for info in model.nncf.input_infos:
         input_shape = tuple([1] + list(info.shape)[1:])
         input_tensor_list.append(torch.rand(input_shape))
 

nncf/experimental/torch/sparsity/movement/structured_mask_handler.py

@@ -391,7 +391,7 @@ def _create_structured_mask_context_groups(
         module_vs_sparse_module_info_map = {minfo.module: minfo for minfo in sparsified_module_info_list}
 
         pruning_producing_types = ["linear"]
-        nncf_graph = nncf_network.get_original_graph()
+        nncf_graph = nncf_network.nncf.get_original_graph()
         pruning_groups = get_pruning_groups(
             nncf_graph, PT_EXPERIMENTAL_PRUNING_OPERATOR_METATYPES, pruning_producing_types
         )

nncf/torch/nncf_network.py

@@ -23,7 +23,6 @@
 from torch import nn
 
 from nncf import nncf_logger
-from nncf.common.deprecation import warning_deprecated
 from nncf.common.graph import NNCFNode
 from nncf.common.graph import NNCFNodeName
 from nncf.common.graph.definitions import MODEL_INPUT_OP_NAME
@@ -967,26 +966,6 @@ def nncf(self) -> NNCFNetworkInterface:
         # self._nncf is being set in the creation function defined in the NNCFNetworkMeta metaclass
         return self._nncf
 
-    def __getattr__(self, key):
-        """
-        Only defined for purposes of deprecation warnings. This method should be removed after v2.5.0.
-        """
-        try:
-            return super().__getattr__(key)
-        except AttributeError as e:
-            if hasattr(self._nncf, key):
-                warning_deprecated(
-                    "Old style of accessing NNCF-specific attributes and methods on NNCFNetwork "
-                    "objects is deprecated. "
-                    "Access the NNCF-specific attrs through the NNCFInterface, which is "
-                    "set up as an `nncf` attribute on the compressed model object.\n"
-                    "For instance, instead of `compressed_model.get_graph()` "
-                    "you should now write `compressed_model.nncf.get_graph()`.\n"
-                    "The old style will be removed after NNCF v2.5.0"
-                )
-                return getattr(self._nncf, key)
-            raise e
-
     def __setattr__(self, key, value):
         # If setting `forward`, set it on the original model.
         if key == "forward":
@@ -1004,16 +983,6 @@ def __setattr__(self, key, value):
             )
         super().__setattr__(key, value)
 
-    def get_nncf_wrapped_model(self) -> "NNCFNetwork":
-        warning_deprecated(
-            "Calls to NNCFNetwork.get_nncf_wrapped_model() are deprecated and will be removed "
-            "in NNCF v2.6.0.\n"
-            "Starting from NNCF v2.5.0, the compressed model object already inherits the original "
-            "class of the uncompressed model and the forward signature, so the call to "
-            ".get_nncf_wrapped_model() may be simply omitted."
-        )
-        return self
-
 
 class NNCFSkippingIter:
     """

tests/torch/pruning/experimental/test_nodes_grouping.py

@@ -474,7 +474,7 @@ def test_groups(desc: GroupTestDesc, mocker, tmp_path):
     pruning_producing_types = ["linear"]
     get_graph_spy = mocker.spy(BlockHierarchy, "_get_graph_for_visualization")
     not_filtered_groups = get_pruning_groups(
-        nncf_network.get_graph(), PT_EXPERIMENTAL_PRUNING_OPERATOR_METATYPES, pruning_producing_types, tmp_path
+        nncf_network.nncf.get_graph(), PT_EXPERIMENTAL_PRUNING_OPERATOR_METATYPES, pruning_producing_types, tmp_path
     )
 
     nx_graph = get_graph_spy.spy_return
@@ -511,7 +511,7 @@ def test_all_groups_valid(desc: GroupTestDesc):
     nncf_network = create_nncf_network(model, config)
     pruning_producing_types = ["linear"]
     all_groups = get_pruning_groups(
-        nncf_network.get_graph(), PT_EXPERIMENTAL_PRUNING_OPERATOR_METATYPES, pruning_producing_types
+        nncf_network.nncf.get_graph(), PT_EXPERIMENTAL_PRUNING_OPERATOR_METATYPES, pruning_producing_types
     )
     for group in all_groups:
         assert group.consumers

tests/torch/test_nncf_network.py

@@ -24,7 +24,6 @@
 from nncf.common.graph import NNCFNode
 from nncf.common.graph.operator_metatypes import UnknownMetatype
 from nncf.common.graph.transformations.commands import TargetType
-from nncf.common.logging.logger import NNCFDeprecationWarning
 from nncf.torch import register_module
 from nncf.torch.dynamic_graph.graph_tracer import ModelInputInfo
 from nncf.torch.dynamic_graph.operation_address import OperationAddress
@@ -584,13 +583,6 @@ def test_works_when_wrapped_with_dataparallel(simple_net):
     dp_model(torch.zeros([10, *simple_net.INPUT_SIZE[1:]], device="cuda"))
 
 
-def test_warns_on_old_style_calls(simple_net):
-    with pytest.warns(NNCFDeprecationWarning):
-        simple_net.get_graph()
-    with pytest.warns(NNCFDeprecationWarning):
-        simple_net.get_nncf_wrapped_model()
-
-
 def test_class_has_same_name_and_module_as_original(simple_net):
     assert simple_net.__class__.__name__ == SimplestModel.__name__
     assert simple_net.__class__.__module__ == SimplestModel.__module__
@@ -657,10 +649,10 @@ def test_reset_original_unbound_forward():
     inp = torch.ones((1,))
     assert nncf_network.forward(inp) == inp
 
-    nncf_network.set_original_unbound_forward(model.__class__.other_forward)
+    nncf_network.nncf.set_original_unbound_forward(model.__class__.other_forward)
     assert nncf_network.forward(inp) == inp * 2
 
-    nncf_network.reset_original_unbound_forward()
+    nncf_network.nncf.reset_original_unbound_forward()
     assert nncf_network.forward(inp) == inp