Merge remote-tracking branch 'origin/develop' into dl/torch/levit

Makefile

@@ -99,7 +99,7 @@ install-tensorflow-dev: install-tensorflow-test install-pre-commit install-pylin
 	pip install -r examples/post_training_quantization/tensorflow/mobilenet_v2/requirements.txt
 
 test-tensorflow:
-	pytest ${COVERAGE_ARGS} tests/common tests/tensorflow    \
+	pytest ${COVERAGE_ARGS} tests/tensorflow    \
 		--junitxml ${JUNITXML_PATH}         \
 		$(DATA_ARG)
 
@@ -128,7 +128,7 @@ install-torch-dev: install-torch-test install-pre-commit install-pylint
 	pip install -r examples/post_training_quantization/torch/ssd300_vgg16/requirements.txt
 
 test-torch:
-	pytest ${COVERAGE_ARGS} tests/common tests/torch -m "not weekly and not nightly" --junitxml ${JUNITXML_PATH} $(DATA_ARG)
+	pytest ${COVERAGE_ARGS} tests/torch -m "not weekly and not nightly" --junitxml ${JUNITXML_PATH} $(DATA_ARG)
 
 test-torch-nightly:
 	pytest ${COVERAGE_ARGS} tests/torch -m nightly --junitxml ${JUNITXML_PATH} $(DATA_ARG)

nncf/quantization/algorithms/algorithm.py

@@ -11,7 +11,7 @@
 
 from abc import ABC
 from abc import abstractmethod
-from typing import Dict, Optional, TypeVar
+from typing import List, Optional, TypeVar
 
 from nncf import Dataset
 from nncf.common.graph.graph import NNCFGraph
@@ -28,11 +28,11 @@ class Algorithm(ABC):
 
     @property
     @abstractmethod
-    def available_backends(self) -> Dict[str, BackendType]:
+    def available_backends(self) -> List[BackendType]:
         """
-        Returns dictionary of the available backends for the algorithm.
+        Returns list of the available backends for the algorithm.
 
-        :return: Dict of backends supported by the algorithm.
+        :return: List of backends supported by the algorithm.
         """
 
     @abstractmethod

nncf/quantization/algorithms/bias_correction/algorithm.py

@@ -32,7 +32,6 @@
 from nncf.common.utils.backend import copy_model
 from nncf.common.utils.backend import get_backend
 from nncf.quantization.algorithms.algorithm import Algorithm
-from nncf.quantization.algorithms.bias_correction.backend import ALGO_BACKENDS
 
 TModel = TypeVar("TModel")
 
@@ -104,8 +103,8 @@ def __init__(
             raise RuntimeError("BiasCorrection algorithm does not support apply_for_all_nodes=True yet")
 
     @property
-    def available_backends(self) -> Dict[str, BackendType]:
-        return ALGO_BACKENDS.registry_dict
+    def available_backends(self) -> List[BackendType]:
+        return [BackendType.ONNX, BackendType.OPENVINO]
 
     def _set_backend_entity(self, model: TModel) -> None:
         """

nncf/quantization/algorithms/bias_correction/backend.py

@@ -22,11 +22,9 @@
 from nncf.common.graph.transformations.commands import TransformationCommand
 from nncf.common.tensor import NNCFTensor
 from nncf.common.tensor_statistics.collectors import TensorStatisticCollectorBase
-from nncf.common.utils.registry import Registry
 
 TModel = TypeVar("TModel")
 OutputType = TypeVar("OutputType")
-ALGO_BACKENDS = Registry("algo_backends")
 
 
 # pylint:disable=too-many-public-methods

nncf/quantization/algorithms/bias_correction/onnx_backend.py

@@ -17,7 +17,6 @@
 from nncf.common.graph import NNCFGraph
 from nncf.common.graph import NNCFNode
 from nncf.common.graph.transformations.commands import TargetType
-from nncf.common.utils.backend import BackendType
 from nncf.onnx.graph.model_utils import remove_fq_from_inputs
 from nncf.onnx.graph.node_utils import get_bias_value
 from nncf.onnx.graph.node_utils import is_any_weight_quantized
@@ -33,12 +32,10 @@
 from nncf.onnx.statistics.collectors import ONNXNNCFCollectorTensorProcessor
 from nncf.onnx.statistics.collectors import ONNXRawStatisticCollector
 from nncf.onnx.tensor import ONNXNNCFTensor
-from nncf.quantization.algorithms.bias_correction.backend import ALGO_BACKENDS
 from nncf.quantization.algorithms.bias_correction.backend import BiasCorrectionAlgoBackend
 
 
 # pylint:disable=too-many-public-methods
-@ALGO_BACKENDS.register(BackendType.ONNX)
 class ONNXBiasCorrectionAlgoBackend(BiasCorrectionAlgoBackend):
     @property
     def tensor_processor(self) -> ONNXNNCFCollectorTensorProcessor:

nncf/quantization/algorithms/bias_correction/openvino_backend.py

@@ -17,7 +17,6 @@
 from nncf.common.graph import NNCFGraph
 from nncf.common.graph import NNCFNode
 from nncf.common.graph.transformations.commands import TargetType
-from nncf.common.utils.backend import BackendType
 from nncf.experimental.common.tensor_statistics.collectors import TensorCollector
 from nncf.openvino.graph.metatypes.groups import FAKE_QUANTIZE_OPERATIONS
 from nncf.openvino.graph.model_utils import insert_null_biases
@@ -33,12 +32,10 @@
 from nncf.openvino.statistics.collectors import get_mean_statistic_collector
 from nncf.openvino.statistics.collectors import get_raw_stat_collector
 from nncf.openvino.tensor import OVNNCFTensor
-from nncf.quantization.algorithms.bias_correction.backend import ALGO_BACKENDS
 from nncf.quantization.algorithms.bias_correction.backend import BiasCorrectionAlgoBackend
 
 
 # pylint:disable=too-many-public-methods
-@ALGO_BACKENDS.register(BackendType.OPENVINO)
 class OVBiasCorrectionAlgoBackend(BiasCorrectionAlgoBackend):
     @property
     def tensor_processor(self) -> OVNNCFCollectorTensorProcessor:

nncf/quantization/algorithms/channel_alignment/algorithm.py

@@ -9,7 +9,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Dict, List, Optional, Tuple, TypeVar
+from typing import List, Optional, Tuple, TypeVar
 
 import numpy as np
 
@@ -28,7 +28,6 @@
 from nncf.common.utils.backend import BackendType
 from nncf.common.utils.backend import get_backend
 from nncf.quantization.algorithms.algorithm import Algorithm
-from nncf.quantization.algorithms.channel_alignment.backend import ALGO_BACKENDS
 from nncf.quantization.algorithms.channel_alignment.backend import ChannelAlignmentAlgoBackend
 from nncf.quantization.algorithms.channel_alignment.backend import LayoutDescriptor
 
@@ -75,8 +74,8 @@ def __init__(
         self._algorithm_key = f"CA_{hash(self)}"
 
     @property
-    def available_backends(self) -> Dict[str, BackendType]:
-        return ALGO_BACKENDS.registry_dict
+    def available_backends(self) -> List[BackendType]:
+        return [BackendType.OPENVINO]
 
     def _set_backend_entity(self, model: TModel) -> None:
         """

nncf/quantization/algorithms/channel_alignment/backend.py

@@ -21,10 +21,8 @@
 from nncf.common.graph.transformations.commands import TargetPoint
 from nncf.common.graph.transformations.commands import TargetType
 from nncf.common.tensor_statistics.collectors import TensorStatisticCollectorBase
-from nncf.common.utils.registry import Registry
 
 TModel = TypeVar("TModel")
-ALGO_BACKENDS = Registry("algo_backends")
 
 
 @dataclass

nncf/quantization/algorithms/channel_alignment/openvino_backend.py

@@ -19,7 +19,6 @@
 from nncf.common.graph.layer_attributes import ConvolutionLayerAttributes
 from nncf.common.graph.transformations.commands import TargetType
 from nncf.common.tensor_statistics.collectors import TensorStatisticCollectorBase
-from nncf.common.utils.backend import BackendType
 from nncf.experimental.common.tensor_statistics.collectors import MedianAggregator
 from nncf.experimental.common.tensor_statistics.collectors import TensorCollector
 from nncf.openvino.graph.layer_attributes import OVLayerAttributes
@@ -37,12 +36,10 @@
 from nncf.openvino.statistics.collectors import OVNNCFCollectorTensorProcessor
 from nncf.openvino.statistics.collectors import OVQuantileReducer
 from nncf.openvino.statistics.statistics import OVMinMaxTensorStatistic
-from nncf.quantization.algorithms.channel_alignment.backend import ALGO_BACKENDS
 from nncf.quantization.algorithms.channel_alignment.backend import ChannelAlignmentAlgoBackend
 from nncf.quantization.algorithms.channel_alignment.backend import LayoutDescriptor
 
 
-@ALGO_BACKENDS.register(BackendType.OPENVINO)
 class OVChannelAlignmentAlgoBackend(ChannelAlignmentAlgoBackend):
     @staticmethod
     def target_point(target_type: TargetType, target_node_name: str, port_id: int) -> OVTargetPoint:

nncf/quantization/algorithms/fast_bias_correction/algorithm.py

@@ -30,7 +30,6 @@
 from nncf.experimental.tensor import Tensor
 from nncf.experimental.tensor import functions as fns
 from nncf.quantization.algorithms.algorithm import Algorithm
-from nncf.quantization.algorithms.fast_bias_correction.backend import ALGO_BACKENDS
 
 TModel = TypeVar("TModel")
 TTensor = TypeVar("TTensor")
@@ -92,8 +91,8 @@ def __init__(
             raise RuntimeError("FastBiasCorrection algorithm does not support apply_for_all_nodes=True yet")
 
     @property
-    def available_backends(self) -> Dict[str, BackendType]:
-        return ALGO_BACKENDS.registry_dict
+    def available_backends(self) -> List[BackendType]:
+        return [BackendType.ONNX, BackendType.OPENVINO, BackendType.TORCH]
 
     def _set_backend_entity(self, model: TModel) -> None:
         """

nncf/quantization/algorithms/fast_bias_correction/backend.py

@@ -21,13 +21,11 @@
 from nncf.common.graph.transformations.commands import TargetType
 from nncf.common.graph.transformations.commands import TransformationCommand
 from nncf.common.tensor_statistics.collectors import TensorStatisticCollectorBase
-from nncf.common.utils.registry import Registry
 from nncf.experimental.tensor import Tensor
 
 TModel = TypeVar("TModel")
 TTensor = TypeVar("TTensor")
 OutputType = TypeVar("OutputType")
-ALGO_BACKENDS = Registry("algo_backends")
 
 
 class FastBiasCorrectionAlgoBackend(ABC):

nncf/quantization/algorithms/fast_bias_correction/onnx_backend.py

@@ -17,7 +17,6 @@
 from nncf.common.graph import NNCFGraph
 from nncf.common.graph import NNCFNode
 from nncf.common.graph.transformations.commands import TargetType
-from nncf.common.utils.backend import BackendType
 from nncf.experimental.tensor import Tensor
 from nncf.onnx.graph.node_utils import get_bias_value
 from nncf.onnx.graph.node_utils import is_any_weight_quantized
@@ -28,11 +27,9 @@
 from nncf.onnx.graph.transformations.commands import ONNXNullBiasInsertionCommand
 from nncf.onnx.graph.transformations.commands import ONNXTargetPoint
 from nncf.onnx.statistics.collectors import ONNXMeanStatisticCollector
-from nncf.quantization.algorithms.fast_bias_correction.backend import ALGO_BACKENDS
 from nncf.quantization.algorithms.fast_bias_correction.backend import FastBiasCorrectionAlgoBackend
 
 
-@ALGO_BACKENDS.register(BackendType.ONNX)
 class ONNXFastBiasCorrectionAlgoBackend(FastBiasCorrectionAlgoBackend):
     @property
     def types_to_insert_bias(self):

nncf/quantization/algorithms/fast_bias_correction/openvino_backend.py

@@ -17,7 +17,6 @@
 from nncf.common.graph import NNCFGraph
 from nncf.common.graph import NNCFNode
 from nncf.common.graph.transformations.commands import TargetType
-from nncf.common.utils.backend import BackendType
 from nncf.experimental.common.tensor_statistics.collectors import TensorCollector
 from nncf.experimental.tensor import Tensor
 from nncf.openvino.graph.metatypes.groups import FAKE_QUANTIZE_OPERATIONS
@@ -28,11 +27,9 @@
 from nncf.openvino.graph.transformations.commands import OVModelExtractionCommand
 from nncf.openvino.graph.transformations.commands import OVTargetPoint
 from nncf.openvino.statistics.collectors import get_mean_statistic_collector
-from nncf.quantization.algorithms.fast_bias_correction.backend import ALGO_BACKENDS
 from nncf.quantization.algorithms.fast_bias_correction.backend import FastBiasCorrectionAlgoBackend
 
 
-@ALGO_BACKENDS.register(BackendType.OPENVINO)
 class OVFastBiasCorrectionAlgoBackend(FastBiasCorrectionAlgoBackend):
     @staticmethod
     def target_point(target_type: TargetType, target_node_name: str, port_id: int) -> OVTargetPoint:

nncf/quantization/algorithms/fast_bias_correction/torch_backend.py

@@ -18,10 +18,8 @@
 from nncf.common.graph import NNCFNode
 from nncf.common.graph.definitions import NNCFGraphNodeType
 from nncf.common.graph.transformations.commands import TargetType
-from nncf.common.utils.backend import BackendType
 from nncf.experimental.common.tensor_statistics.collectors import TensorCollector
 from nncf.experimental.tensor import Tensor
-from nncf.quantization.algorithms.fast_bias_correction.backend import ALGO_BACKENDS
 from nncf.quantization.algorithms.fast_bias_correction.backend import FastBiasCorrectionAlgoBackend
 from nncf.torch.graph.transformations.command_creation import create_bias_correction_command
 from nncf.torch.graph.transformations.commands import PTBiasCorrectionCommand
@@ -35,7 +33,6 @@
 from nncf.torch.tensor_statistics.collectors import get_mean_statistic_collector
 
 
-@ALGO_BACKENDS.register(BackendType.TORCH)
 class PTFastBiasCorrectionAlgoBackend(FastBiasCorrectionAlgoBackend):
     TARGET_TYPE_TO_PT_INS_TYPE_MAP = {
         TargetType.PRE_LAYER_OPERATION: TargetType.OPERATOR_PRE_HOOK,

nncf/quantization/algorithms/min_max/algorithm.py

@@ -51,7 +51,6 @@
 from nncf.quantization.advanced_parameters import QuantizationParameters
 from nncf.quantization.advanced_parameters import changes_asdict
 from nncf.quantization.algorithms.algorithm import Algorithm
-from nncf.quantization.algorithms.min_max.backend import ALGO_BACKENDS
 from nncf.quantization.fake_quantize import calculate_quantizer_parameters
 from nncf.quantization.fake_quantize import get_quantizer_narrow_range
 from nncf.quantization.passes import transform_to_inference_graph
@@ -177,8 +176,8 @@ def _reset_cache(self):
         self._unified_scale_groups = []
 
     @property
-    def available_backends(self) -> Dict[str, BackendType]:
-        return ALGO_BACKENDS.registry_dict
+    def available_backends(self) -> List[BackendType]:
+        return [BackendType.ONNX, BackendType.OPENVINO, BackendType.TORCH]
 
     def _get_quantizer_constraints(
         self, group: QuantizerGroup, preset: QuantizationPreset, quantization_params: Optional[QuantizationParameters]

nncf/quantization/algorithms/min_max/backend.py

@@ -23,14 +23,12 @@
 from nncf.common.quantization.structs import QuantizerConfig
 from nncf.common.tensor_statistics.collectors import TensorStatisticCollectorBase
 from nncf.common.tensor_statistics.statistics import MinMaxTensorStatistic
-from nncf.common.utils.registry import Registry
 from nncf.parameters import ModelType
 from nncf.parameters import TargetDevice
 from nncf.quantization.fake_quantize import FakeQuantizeParameters
 from nncf.quantization.range_estimator import RangeEstimatorParameters
 
 TModel = TypeVar("TModel")
-ALGO_BACKENDS = Registry("algo_backends")
 
 
 # pylint:disable=too-many-public-methods

nncf/quantization/algorithms/min_max/onnx_backend.py

@@ -20,7 +20,6 @@
 from nncf.common.hardware.config import HWConfig
 from nncf.common.quantization.structs import QuantizationMode
 from nncf.common.quantization.structs import QuantizerConfig
-from nncf.common.utils.backend import BackendType
 from nncf.onnx.graph.metatypes import onnx_metatypes as om
 from nncf.onnx.graph.metatypes.groups import MATMUL_METATYPES
 from nncf.onnx.graph.node_utils import get_input_edges_mapping
@@ -39,14 +38,12 @@
 from nncf.parameters import TargetDevice
 from nncf.quantization.advanced_parameters import AggregatorType
 from nncf.quantization.advanced_parameters import StatisticsType
-from nncf.quantization.algorithms.min_max.backend import ALGO_BACKENDS
 from nncf.quantization.algorithms.min_max.backend import MinMaxAlgoBackend
 from nncf.quantization.fake_quantize import FakeQuantizeParameters
 from nncf.quantization.range_estimator import RangeEstimatorParameters
 
 
 # pylint:disable=too-many-public-methods
-@ALGO_BACKENDS.register(BackendType.ONNX)
 class ONNXMinMaxAlgoBackend(MinMaxAlgoBackend):
     @property
     def mat_mul_metatypes(self) -> List[OperatorMetatype]:

nncf/quantization/algorithms/min_max/openvino_backend.py

@@ -21,7 +21,6 @@
 from nncf.common.quantization.structs import QuantizationMode
 from nncf.common.quantization.structs import QuantizerConfig
 from nncf.common.tensor_statistics.collectors import ReductionAxes
-from nncf.common.utils.backend import BackendType
 from nncf.experimental.common.tensor_statistics.collectors import AGGREGATORS_MAP
 from nncf.experimental.common.tensor_statistics.collectors import TensorCollector
 from nncf.openvino.graph.layer_attributes import OVLayerAttributes
@@ -40,13 +39,11 @@
 from nncf.parameters import TargetDevice
 from nncf.quantization.advanced_parameters import RangeEstimatorParameters
 from nncf.quantization.advanced_parameters import StatisticsType
-from nncf.quantization.algorithms.min_max.backend import ALGO_BACKENDS
 from nncf.quantization.algorithms.min_max.backend import MinMaxAlgoBackend
 from nncf.quantization.fake_quantize import FakeQuantizeParameters
 
 
 # pylint:disable=too-many-public-methods
-@ALGO_BACKENDS.register(BackendType.OPENVINO)
 class OVMinMaxAlgoBackend(MinMaxAlgoBackend):
     @property
     def mat_mul_metatypes(self) -> List[OperatorMetatype]:

nncf/quantization/algorithms/min_max/torch_backend.py

@@ -23,13 +23,11 @@
 from nncf.common.hardware.config import HWConfig
 from nncf.common.quantization.structs import QuantizationMode
 from nncf.common.quantization.structs import QuantizerConfig
-from nncf.common.utils.backend import BackendType
 from nncf.experimental.common.tensor_statistics.collectors import AGGREGATORS_MAP
 from nncf.experimental.common.tensor_statistics.collectors import TensorCollector
 from nncf.parameters import ModelType
 from nncf.parameters import TargetDevice
 from nncf.quantization.advanced_parameters import StatisticsType
-from nncf.quantization.algorithms.min_max.backend import ALGO_BACKENDS
 from nncf.quantization.algorithms.min_max.backend import MinMaxAlgoBackend
 from nncf.quantization.fake_quantize import FakeQuantizeParameters
 from nncf.quantization.range_estimator import RangeEstimatorParameters
@@ -50,7 +48,6 @@
 
 
 # pylint:disable=too-many-public-methods
-@ALGO_BACKENDS.register(BackendType.TORCH)
 class PTMinMaxAlgoBackend(MinMaxAlgoBackend):
     TARGET_TYPE_TO_PT_INS_TYPE_MAP = {
         TargetType.PRE_LAYER_OPERATION: TargetType.OPERATOR_PRE_HOOK,