Fix pytest warnings

tests/common/accuracy_control/test_calculate_drop.py

@@ -18,7 +18,7 @@
 
 
 @dataclass
-class TestCase:
+class AccuracyDropTestCase:
     initial_metric: float
     quantized_metric: float
     drop_type: DropType
@@ -31,57 +31,57 @@ class TestCase:
     "ts",
     [
         # ABSOLUTE
-        TestCase(
+        AccuracyDropTestCase(
             initial_metric=0.2923,
             quantized_metric=0.3185,
             drop_type=DropType.ABSOLUTE,
             expected_should_terminate=True,
             expected_accuracy_drop=-0.0262,
         ),
-        TestCase(
+        AccuracyDropTestCase(
             initial_metric=0.3185,
             quantized_metric=0.2923,
             drop_type=DropType.ABSOLUTE,
             expected_should_terminate=False,
             expected_accuracy_drop=0.0262,
         ),
-        TestCase(
+        AccuracyDropTestCase(
             initial_metric=-0.2923,
             quantized_metric=-0.3185,
             drop_type=DropType.ABSOLUTE,
             expected_should_terminate=False,
             expected_accuracy_drop=0.0262,
         ),
-        TestCase(
+        AccuracyDropTestCase(
             initial_metric=-0.3185,
             quantized_metric=-0.2923,
             drop_type=DropType.ABSOLUTE,
             expected_should_terminate=True,
             expected_accuracy_drop=-0.0262,
         ),
         # RELATIVE
-        TestCase(
+        AccuracyDropTestCase(
             initial_metric=0.2923,
             quantized_metric=0.3185,
             drop_type=DropType.RELATIVE,
             expected_should_terminate=True,
             expected_accuracy_drop=None,
         ),
-        TestCase(
+        AccuracyDropTestCase(
             initial_metric=0.3185,
             quantized_metric=0.2923,
             drop_type=DropType.RELATIVE,
             expected_should_terminate=False,
             expected_accuracy_drop=0.08226059,
         ),
-        TestCase(
+        AccuracyDropTestCase(
             initial_metric=-0.2923,
             quantized_metric=-0.3185,
             drop_type=DropType.RELATIVE,
             expected_should_terminate=False,
             expected_accuracy_drop=0.0896339,
         ),
-        TestCase(
+        AccuracyDropTestCase(
             initial_metric=-0.3185,
             quantized_metric=-0.2923,
             drop_type=DropType.RELATIVE,
@@ -90,7 +90,7 @@ class TestCase:
         ),
     ],
 )
-def test_calculate_accuracy_drop(ts: TestCase):
+def test_calculate_accuracy_drop(ts: AccuracyDropTestCase):
     should_terminate, accuracy_drop = calculate_accuracy_drop(
         ts.initial_metric, ts.quantized_metric, ts.max_drop, ts.drop_type
     )

tests/common/quantization/test_minmax.py

@@ -134,24 +134,23 @@ def test_mode_against_default_map(algo_params, is_error):
     qconf_attr_vs_constraint_dict_to_compare = {"mode": QuantizationScheme.SYMMETRIC}
 
     if is_error:
-        try:
+        with pytest.raises(nncf.ParameterNotSupportedError):
             minmax = MinMaxQuantization(**algo_params)
-        except nncf.ParameterNotSupportedError:
-            pytest.xfail("Caught expected error")
-    minmax = MinMaxQuantization(**algo_params)
-    for ref_parameter_name, ref_parameter_value in default_values_to_compare[mode_param].items():
-        parameter_value = getattr(minmax, ref_parameter_name)
-        assert parameter_value == ref_parameter_value
-
-        global_quantizer_constraints = getattr(minmax, "_global_quantizer_constraints")
-        assert (
-            global_quantizer_constraints[QuantizerGroup.ACTIVATIONS].qconf_attr_vs_constraint_dict
-            == qconf_attr_vs_constraint_dict_to_compare
-        )
-        assert (
-            global_quantizer_constraints[QuantizerGroup.WEIGHTS].qconf_attr_vs_constraint_dict
-            == qconf_attr_vs_constraint_dict_to_compare
-        )
+    else:
+        minmax = MinMaxQuantization(**algo_params)
+        for ref_parameter_name, ref_parameter_value in default_values_to_compare[mode_param].items():
+            parameter_value = getattr(minmax, ref_parameter_name)
+            assert parameter_value == ref_parameter_value
+
+            global_quantizer_constraints = getattr(minmax, "_global_quantizer_constraints")
+            assert (
+                global_quantizer_constraints[QuantizerGroup.ACTIVATIONS].qconf_attr_vs_constraint_dict
+                == qconf_attr_vs_constraint_dict_to_compare
+            )
+            assert (
+                global_quantizer_constraints[QuantizerGroup.WEIGHTS].qconf_attr_vs_constraint_dict
+                == qconf_attr_vs_constraint_dict_to_compare
+            )
 
 
 @pytest.mark.parametrize(

tests/common/quantization/test_passes.py

@@ -26,30 +26,33 @@
 DATA_ROOT = TEST_ROOT / "common" / "data" / "reference_graphs"
 
 
-class TestModes(Enum):
+class ParameterTestModes(Enum):
     VALID = "valid"
     WRONG_TENSOR_SHAPE = "wrong_dropout_node"
     WRONG_PARALLEL_EDGES = "wrong_parallel_edges"
 
+    def __str__(self):
+        return self.value
+
 
 def _check_graphs(dot_file_name, nncf_graph) -> None:
     nx_graph = nncf_graph.get_graph_for_structure_analysis()
     path_to_dot = DATA_ROOT / dot_file_name
     compare_nx_graph_with_reference(nx_graph, path_to_dot, check_edge_attrs=True)
 
 
-@pytest.mark.parametrize("mode", [TestModes.VALID, TestModes.WRONG_TENSOR_SHAPE, TestModes.WRONG_PARALLEL_EDGES])
-def test_remove_nodes_and_reconnect_graph(mode: TestModes):
+@pytest.mark.parametrize("mode", ParameterTestModes)
+def test_remove_nodes_and_reconnect_graph(mode: ParameterTestModes):
     dot_reference_path_before = Path("passes") / "dropout_synthetic_model_before.dot"
     dot_reference_path_after = Path("passes") / "dropout_synthetic_model_after.dot"
     dropout_metatype = "DROPOUT_METATYPE"
     kwargs = {}
-    if mode != TestModes.VALID:
+    if mode != ParameterTestModes.VALID:
         kwargs.update({mode.value: True})
 
     nncf_graph = NNCFGraphDropoutRemovingCase(dropout_metatype, **kwargs).nncf_graph
 
-    if mode != TestModes.VALID:
+    if mode != ParameterTestModes.VALID:
         with pytest.raises(AssertionError):
             remove_nodes_and_reconnect_graph(nncf_graph, [dropout_metatype])
         return

tests/common/quantization/test_quantizer_removal.py

@@ -194,7 +194,7 @@ class Graph:
 
 
 @dataclass
-class TestCase:
+class ParameterTestCase:
     """
     :param node_name: Quantizer node's name. We want to remove this
         quantizer from the model.
@@ -212,42 +212,46 @@ class TestCase:
 
 TEST_CASES = {
     "simple_graph": [
-        TestCase(
+        ParameterTestCase(
             "fake_quantize_119",
             ["fake_quantize_139", "fake_quantize_162", "fake_quantize_119"],
             ["add_117", "conv2d_161"],
         ),
-        TestCase("fake_quantize_128", ["fake_quantize_134", "fake_quantize_128"], ["conv2d_127"]),
-        TestCase("fake_quantize_134", ["fake_quantize_134", "fake_quantize_128"], ["conv2d_127"]),
-        TestCase(
+        ParameterTestCase("fake_quantize_128", ["fake_quantize_134", "fake_quantize_128"], ["conv2d_127"]),
+        ParameterTestCase("fake_quantize_134", ["fake_quantize_134", "fake_quantize_128"], ["conv2d_127"]),
+        ParameterTestCase(
             "fake_quantize_139",
             ["fake_quantize_139", "fake_quantize_162", "fake_quantize_119"],
             ["add_117", "conv2d_161"],
         ),
-        TestCase("fake_quantize_147", ["fake_quantize_153", "fake_quantize_147"], ["conv2d_146"]),
-        TestCase("fake_quantize_153", ["fake_quantize_153", "fake_quantize_147"], ["conv2d_146"]),
-        TestCase(
+        ParameterTestCase("fake_quantize_147", ["fake_quantize_153", "fake_quantize_147"], ["conv2d_146"]),
+        ParameterTestCase("fake_quantize_153", ["fake_quantize_153", "fake_quantize_147"], ["conv2d_146"]),
+        ParameterTestCase(
             "fake_quantize_162",
             ["fake_quantize_139", "fake_quantize_162", "fake_quantize_119"],
             ["add_117", "conv2d_161"],
         ),
     ],
-    "graph_with_shapeof": [TestCase("fake_quantize_105", ["fake_quantize_105"], ["interpolate_115"])],
+    "graph_with_shapeof": [ParameterTestCase("fake_quantize_105", ["fake_quantize_105"], ["interpolate_115"])],
     "simple_graph_quantize_dequantize": [
-        TestCase("quantize_37", ["quantize_37", "dequantize_38", "quantize_39", "dequantize_40"], ["conv2d_41"]),
-        TestCase("quantize_39", ["quantize_37", "dequantize_38", "quantize_39", "dequantize_40"], ["conv2d_41"]),
+        ParameterTestCase(
+            "quantize_37", ["quantize_37", "dequantize_38", "quantize_39", "dequantize_40"], ["conv2d_41"]
+        ),
+        ParameterTestCase(
+            "quantize_39", ["quantize_37", "dequantize_38", "quantize_39", "dequantize_40"], ["conv2d_41"]
+        ),
         #
-        TestCase(
+        ParameterTestCase(
             "quantize_42",
             ["quantize_42", "dequantize_43", "quantize_44", "dequantize_45", "quantize_63", "dequantize_64"],
             ["conv2d_46", "add_65"],
         ),
-        TestCase(
+        ParameterTestCase(
             "quantize_44",
             ["quantize_42", "dequantize_43", "quantize_44", "dequantize_45", "quantize_63", "dequantize_64"],
             ["conv2d_46", "add_65"],
         ),
-        TestCase(
+        ParameterTestCase(
             "quantize_63",
             ["quantize_42", "dequantize_43", "quantize_44", "dequantize_45", "quantize_63", "dequantize_64"],
             ["conv2d_46", "add_65"],
@@ -295,7 +299,7 @@ def create_test_params():
 
 
 @pytest.mark.parametrize("nncf_graph,test_case", TEST_PARAMS, ids=IDS)
-def test_find_quantizer_nodes_to_cut(nncf_graph: NNCFGraph, test_case: TestCase):
+def test_find_quantizer_nodes_to_cut(nncf_graph: NNCFGraph, test_case: ParameterTestCase):
     quantizer_node = nncf_graph.get_node_by_name(test_case.node_name)
     # As test graphs are fully connected and does not have readvariable metatype,
     # this should work

tests/common/requirements.txt

@@ -1,5 +1,6 @@
 -c ../../constraints.txt
 pytest
 pytest-cov
+pytest-dependency
 pytest-mock
 pytest-xdist

tests/onnx/pytest.ini

@@ -1,7 +1,7 @@
 [pytest]
 markers =
     e2e_ptq: e2e ptq tests
-    e2e_eval_original_model: original model evaluation 
+    e2e_eval_reference_model: original model evaluation
 
 python_files = test_*
 xfail_strict = true

tests/onnx/quantization/test_classification_models_graph.py

@@ -23,21 +23,61 @@
 from tests.onnx.weightless_model import load_model_topology_with_zeros_weights
 
 TORCHVISION_TEST_DATA = [
-    (ModelToTest("resnet18", [1, 3, 224, 224]), models.resnet18(pretrained=True), {}),
+    (
+        ModelToTest("resnet18", [1, 3, 224, 224]),
+        models.resnet18(weights=models.ResNet18_Weights.IMAGENET1K_V1),
+        {},
+    ),
     (
         ModelToTest("resnet50_cpu_spr", [1, 3, 224, 224]),
-        models.resnet50(pretrained=True),
+        models.resnet50(weights=models.ResNet50_Weights.IMAGENET1K_V1),
         {"target_device": TargetDevice.CPU_SPR},
     ),
-    (ModelToTest("mobilenet_v2", [1, 3, 224, 224]), models.mobilenet_v2(pretrained=True), {}),
-    (ModelToTest("mobilenet_v3_small", [1, 3, 224, 224]), models.mobilenet_v3_small(pretrained=True), {}),
-    (ModelToTest("inception_v3", [1, 3, 224, 224]), models.inception_v3(pretrained=True), {}),
-    (ModelToTest("googlenet", [1, 3, 224, 224]), models.googlenet(pretrained=True), {}),
-    (ModelToTest("vgg16", [1, 3, 224, 224]), models.vgg16(pretrained=True), {}),
-    (ModelToTest("shufflenet_v2_x1_0", [1, 3, 224, 224]), models.shufflenet_v2_x1_0(pretrained=True), {}),
-    (ModelToTest("squeezenet1_0", [1, 3, 224, 224]), models.squeezenet1_0(pretrained=True), {}),
-    (ModelToTest("densenet121", [1, 3, 224, 224]), models.densenet121(pretrained=True), {}),
-    (ModelToTest("mnasnet0_5", [1, 3, 224, 224]), models.mnasnet0_5(pretrained=True), {}),
+    (
+        ModelToTest("mobilenet_v2", [1, 3, 224, 224]),
+        models.mobilenet_v2(weights=models.MobileNet_V2_Weights.IMAGENET1K_V1),
+        {},
+    ),
+    (
+        ModelToTest("mobilenet_v3_small", [1, 3, 224, 224]),
+        models.mobilenet_v3_small(weights=models.MobileNet_V3_Small_Weights.IMAGENET1K_V1),
+        {},
+    ),
+    (
+        ModelToTest("inception_v3", [1, 3, 224, 224]),
+        models.inception_v3(weights=models.Inception_V3_Weights.IMAGENET1K_V1),
+        {},
+    ),
+    (
+        ModelToTest("googlenet", [1, 3, 224, 224]),
+        models.googlenet(weights=models.GoogLeNet_Weights.IMAGENET1K_V1),
+        {},
+    ),
+    (
+        ModelToTest("vgg16", [1, 3, 224, 224]),
+        models.vgg16(weights=models.VGG16_Weights.IMAGENET1K_V1),
+        {},
+    ),
+    (
+        ModelToTest("shufflenet_v2_x1_0", [1, 3, 224, 224]),
+        models.shufflenet_v2_x1_0(weights=models.ShuffleNet_V2_X1_0_Weights.IMAGENET1K_V1),
+        {},
+    ),
+    (
+        ModelToTest("squeezenet1_0", [1, 3, 224, 224]),
+        models.squeezenet1_0(weights=models.SqueezeNet1_0_Weights.IMAGENET1K_V1),
+        {},
+    ),
+    (
+        ModelToTest("densenet121", [1, 3, 224, 224]),
+        models.densenet121(weights=models.DenseNet121_Weights.IMAGENET1K_V1),
+        {},
+    ),
+    (
+        ModelToTest("mnasnet0_5", [1, 3, 224, 224]),
+        models.mnasnet0_5(weights=models.MNASNet0_5_Weights.IMAGENET1K_V1),
+        {},
+    ),
 ]
 
 

tests/onnx/quantization/test_opset_versions.py

@@ -22,7 +22,7 @@
 
 @pytest.mark.parametrize("opset_version", TEST_OPSETS)
 def test_model_opset_version(tmp_path, opset_version):
-    model = models.mobilenet_v2(pretrained=True)
+    model = models.mobilenet_v2(weights=models.MobileNet_V2_Weights.IMAGENET1K_V1)
     input_shape = [1, 3, 224, 224]
     x = torch.randn(input_shape, requires_grad=False)
     torch.onnx.export(model, x, tmp_path / "model.onnx", opset_version=opset_version)

tests/onnx/test_nncf_graph_builder.py

@@ -42,16 +42,46 @@ def test_compare_nncf_graph_synthetic_models(model_cls_to_test):
 
 
 CLASSIFICATION_MODEL_DEF_AND_OBJ = [
-    (ModelToTest("resnet18", [1, 3, 224, 224]), models.resnet18(pretrained=True)),
-    (ModelToTest("mobilenet_v2", [1, 3, 224, 224]), models.mobilenet_v2(pretrained=True)),
-    (ModelToTest("mobilenet_v3_small", [1, 3, 224, 224]), models.mobilenet_v3_small(pretrained=True)),
-    (ModelToTest("inception_v3", [1, 3, 224, 224]), models.inception_v3(pretrained=True)),
-    (ModelToTest("googlenet", [1, 3, 224, 224]), models.googlenet(pretrained=True)),
-    (ModelToTest("vgg16", [1, 3, 224, 224]), models.vgg16(pretrained=True)),
-    (ModelToTest("shufflenet_v2_x1_0", [1, 3, 224, 224]), models.shufflenet_v2_x1_0(pretrained=True)),
-    (ModelToTest("squeezenet1_0", [1, 3, 224, 224]), models.squeezenet1_0(pretrained=True)),
-    (ModelToTest("densenet121", [1, 3, 224, 224]), models.densenet121(pretrained=True)),
-    (ModelToTest("mnasnet0_5", [1, 3, 224, 224]), models.mnasnet0_5(pretrained=True)),
+    (
+        ModelToTest("resnet18", [1, 3, 224, 224]),
+        models.resnet18(weights=models.ResNet18_Weights.IMAGENET1K_V1),
+    ),
+    (
+        ModelToTest("mobilenet_v2", [1, 3, 224, 224]),
+        models.mobilenet_v2(weights=models.MobileNet_V2_Weights.IMAGENET1K_V1),
+    ),
+    (
+        ModelToTest("mobilenet_v3_small", [1, 3, 224, 224]),
+        models.mobilenet_v3_small(weights=models.MobileNet_V3_Small_Weights.IMAGENET1K_V1),
+    ),
+    (
+        ModelToTest("inception_v3", [1, 3, 224, 224]),
+        models.inception_v3(weights=models.Inception_V3_Weights.IMAGENET1K_V1),
+    ),
+    (
+        ModelToTest("googlenet", [1, 3, 224, 224]),
+        models.googlenet(weights=models.GoogLeNet_Weights.IMAGENET1K_V1),
+    ),
+    (
+        ModelToTest("vgg16", [1, 3, 224, 224]),
+        models.vgg16(weights=models.VGG16_Weights.IMAGENET1K_V1),
+    ),
+    (
+        ModelToTest("shufflenet_v2_x1_0", [1, 3, 224, 224]),
+        models.shufflenet_v2_x1_0(weights=models.ShuffleNet_V2_X1_0_Weights.IMAGENET1K_V1),
+    ),
+    (
+        ModelToTest("squeezenet1_0", [1, 3, 224, 224]),
+        models.squeezenet1_0(weights=models.SqueezeNet1_0_Weights.IMAGENET1K_V1),
+    ),
+    (
+        ModelToTest("densenet121", [1, 3, 224, 224]),
+        models.densenet121(weights=models.DenseNet121_Weights.IMAGENET1K_V1),
+    ),
+    (
+        ModelToTest("mnasnet0_5", [1, 3, 224, 224]),
+        models.mnasnet0_5(weights=models.MNASNet0_5_Weights.IMAGENET1K_V1),
+    ),
 ]