[WeightCompression] Ignored patterns introduction (#3562)

### Changes

Ignored patterns are being introduced to the WeightCompression algorithm

### Reason for changes

ROPE pattern breaks the logic of the WeightCompression algorithm,
required to be automatically ignored

### Related tickets
 huggingface/optimum-intel#1295
 164548

### Tests
test_rope_weight_compression

---------

Co-authored-by: Alexander Dokuchaev <alexander.dokuchaev@intel.com>

Author: daniil-lyakhov

src/nncf/common/graph/graph.py

@@ -810,3 +810,18 @@ def find_matching_subgraphs(self, patterns: GraphPattern, strict: bool = True) -
                 subgraph_list.append(self.get_node_by_key(node_key))
             output.append(subgraph_list)
         return output
+
+
+def get_node_names_matching_graph_pattern(nncf_graph: NNCFGraph, graph_pattern: GraphPattern) -> set[str]:
+    """
+    Returns the names of nodes in the given NNCFGraph that match the specified graph pattern.
+
+    :param nncf_graph: An instance of NNCFGraph to search for matching subgraphs.
+    :param graph_pattern: A GraphPattern instance used to identify matching subgraphs.
+    :return: A set of node names from the NNCFGraph that match the given graph pattern.
+    """
+    nncf_node_names = set()
+    for subgraph in nncf_graph.find_matching_subgraphs(graph_pattern, strict=False):
+        for nncf_node in subgraph:
+            nncf_node_names.add(nncf_node.node_name)
+    return nncf_node_names

src/nncf/quantization/algorithms/min_max/algorithm.py

@@ -22,6 +22,7 @@
 from nncf.common.factory import ModelTransformerFactory
 from nncf.common.graph.graph import NNCFGraph
 from nncf.common.graph.graph import NNCFNode
+from nncf.common.graph.graph import get_node_names_matching_graph_pattern
 from nncf.common.graph.operator_metatypes import OperatorMetatype
 from nncf.common.graph.patterns import GraphPattern
 from nncf.common.graph.patterns.manager import PatternsManager
@@ -585,32 +586,14 @@ def _get_ignored_names(
         user_ignored_names = get_ignored_node_names_from_ignored_scope(
             self._ignored_scope, nncf_graph, strict=self._ignored_scope.validate
         )
-        autogenerated_ignored_names = self._get_ignored_names_by_ignored_patterns(
-            inference_nncf_graph, ignored_patterns
-        )
+        autogenerated_ignored_names = get_node_names_matching_graph_pattern(inference_nncf_graph, ignored_patterns)
         autogenerated_ignored_names |= self._backend_entity.get_ignored_names_by_layer_attributes(inference_nncf_graph)
         autogenerated_ignored_names |= self._get_ignored_names_by_algorithm(inference_nncf_graph)
         ignored_names = {name: IgnoreReason.AUTOGENERATED for name in autogenerated_ignored_names}
         # User ignored scope has higher priority
         ignored_names.update({name: IgnoreReason.USER_REQUESTED for name in user_ignored_names})
         return ignored_names
 
-    def _get_ignored_names_by_ignored_patterns(
-        self, inference_nncf_graph: NNCFGraph, ignored_patterns: GraphPattern
-    ) -> set[str]:
-        """
-        Returns node names matched ignored_patterns.
-
-        :param nncf_graph: Inference graph without constant flows.
-        :param ignored_patterns: Ignored patterns.
-        :return: IgnoredScope with all node names matched ignored_patterns.
-        """
-        nncf_node_names = set()
-        for subgraph in inference_nncf_graph.find_matching_subgraphs(ignored_patterns, strict=False):
-            for nncf_node in subgraph:
-                nncf_node_names.add(nncf_node.node_name)
-        return nncf_node_names
-
     def _get_ignored_names_by_algorithm(self, inference_nncf_graph: NNCFGraph) -> set[str]:
         """
         Returns node names for ignored_algorithms matched `quantization`.

src/nncf/quantization/algorithms/weight_compression/algorithm.py

@@ -20,6 +20,7 @@
 from nncf.common.factory import StatisticsAggregatorFactory
 from nncf.common.graph.graph import NNCFGraph
 from nncf.common.graph.graph import NNCFNode
+from nncf.common.graph.graph import get_node_names_matching_graph_pattern
 from nncf.common.graph.transformations.commands import TargetType
 from nncf.common.logging import nncf_logger
 from nncf.common.logging.track_progress import track
@@ -419,6 +420,12 @@ def get_nodes_to_compress(self, nncf_graph: NNCFGraph) -> list[NNCFNode]:
         ignored_names = get_ignored_node_names_from_ignored_scope(
             self._ignored_scope, nncf_graph, strict=self._ignored_scope.validate
         )
+
+        autogenerated_ignored_names = get_node_names_matching_graph_pattern(
+            nncf_graph, self._backend_entity.get_ignored_patterns()
+        )
+        ignored_names = ignored_names.union(autogenerated_ignored_names)
+
         for node in nncf_graph.topological_sort():
             is_node_with_weights = self._backend_entity.is_node_with_weights(node, nncf_graph)
             is_within_scope = should_consider_scope(node.node_name, ignored_names)

src/nncf/quantization/algorithms/weight_compression/backend.py

@@ -16,6 +16,7 @@
 from nncf.common.graph import NNCFGraph
 from nncf.common.graph import NNCFNode
 from nncf.common.graph.operator_metatypes import OperatorMetatype
+from nncf.common.graph.patterns.patterns import GraphPattern
 from nncf.common.graph.transformations.commands import TargetPoint
 from nncf.common.graph.transformations.commands import TargetType
 from nncf.common.tensor_statistics.collectors import TensorStatisticCollectorBase
@@ -256,6 +257,15 @@ def get_filter_fn_for_statistics(activation_port_id: int, algorithm_key: str) ->
         :return: Backend-specific callable to filter statistic containers according to its statistic point.
         """
 
+    @staticmethod
+    @abstractmethod
+    def get_ignored_patterns() -> GraphPattern:
+        """
+        Return backend-specific ignored patterns.
+
+        :return: backend-specific ignored patterns.
+        """
+
 
 class AWQAlgoBackend(WeightCompressionAlgoBackend):
     @staticmethod

src/nncf/quantization/algorithms/weight_compression/onnx_backend.py

@@ -21,6 +21,7 @@
 from nncf.common.graph import NNCFGraph
 from nncf.common.graph import NNCFNode
 from nncf.common.graph.operator_metatypes import OperatorMetatype
+from nncf.common.graph.patterns.patterns import GraphPattern
 from nncf.common.graph.transformations.commands import TargetType
 from nncf.common.graph.utils import get_reduction_axes
 from nncf.common.tensor_statistics.statistic_point import StatisticPoint
@@ -43,6 +44,7 @@
 from nncf.onnx.graph.onnx_helper import pack_4_bits
 from nncf.onnx.graph.onnx_helper import pack_int4_to_uint8
 from nncf.onnx.graph.transformations.commands import ONNXTargetPoint
+from nncf.onnx.quantization.ignored_patterns import create_rope
 from nncf.parameters import CompressionFormat
 from nncf.parameters import CompressWeightsMode
 from nncf.quantization.advanced_parameters import AdvancedCompressionParameters
@@ -458,3 +460,7 @@ def _replace_matmul_with_matmulnbits(
         del self.name_to_node_map[original_matmul.name]
         # Update the node mapping
         self.name_to_node_map[matmul_n_bits.name] = matmul_n_bits
+
+    @staticmethod
+    def get_ignored_patterns() -> GraphPattern:
+        return create_rope()

src/nncf/quantization/algorithms/weight_compression/openvino_backend.py

@@ -17,6 +17,7 @@
 from nncf.common.graph import NNCFGraph
 from nncf.common.graph import NNCFNode
 from nncf.common.graph.operator_metatypes import OperatorMetatype
+from nncf.common.graph.patterns.patterns import GraphPattern
 from nncf.common.graph.transformations.commands import TargetType
 from nncf.common.graph.utils import get_reduction_axes
 from nncf.common.tensor_statistics.statistic_point import StatisticPoint
@@ -41,6 +42,7 @@
 from nncf.openvino.graph.transformations.commands import OVTargetPoint
 from nncf.openvino.optimized_functions import clear_ov_model_cache
 from nncf.openvino.optimized_functions.models import OV_MODEL_CACHE
+from nncf.openvino.quantization.ignored_patterns import create_rope
 from nncf.openvino.rt_info import dump_parameters
 from nncf.openvino.statistics.collectors import OVMaxVarianceReducer
 from nncf.openvino.statistics.collectors import OVMeanAbsMaxReducer
@@ -367,6 +369,10 @@ def filter_func(point: StatisticPoint) -> bool:
 
         return filter_func
 
+    @staticmethod
+    def get_ignored_patterns() -> GraphPattern:
+        return create_rope()
+
 
 class OVTensorWeightCompressionAlgoBackend(OVWeightCompressionAlgoBackend):
     """

src/nncf/quantization/algorithms/weight_compression/torch_backend.py

@@ -66,6 +66,7 @@
 from nncf.torch.model_graph_manager import split_const_name
 from nncf.torch.model_transformer import PTModelTransformer
 from nncf.torch.nncf_network import NNCFNetwork
+from nncf.torch.quantization.ignored_patterns import create_rope
 from nncf.torch.quantization.layers import QUANTIZATION_MODULES
 from nncf.torch.quantization.layers import INT4AsymmetricWeightsDecompressor
 from nncf.torch.quantization.layers import INT4SymmetricWeightsDecompressor
@@ -510,6 +511,10 @@ def transform_model(
 
         return transformed_model
 
+    @staticmethod
+    def get_ignored_patterns() -> GraphPattern:
+        return create_rope()
+
 
 class PTAWQAlgoAlgoBackend(AWQAlgoBackend, PTWeightCompressionAlgoBackend):
     @staticmethod

src/nncf/quantization/algorithms/weight_compression/torch_fx_backend.py

@@ -20,6 +20,7 @@
 from nncf.common.graph.graph import NNCFGraph
 from nncf.common.graph.graph import NNCFNode
 from nncf.common.graph.operator_metatypes import OperatorMetatype
+from nncf.common.graph.patterns.patterns import GraphPattern
 from nncf.common.graph.transformations.commands import TargetType
 from nncf.common.graph.transformations.layout import TransformationLayout
 from nncf.common.tensor_statistics.statistic_point import StatisticPoint
@@ -53,6 +54,7 @@
 from nncf.torch.graph.transformations.commands import PTTargetPoint
 from nncf.torch.model_graph_manager import get_const_node
 from nncf.torch.model_graph_manager import get_weight_tensor_port_ids
+from nncf.torch.quantization.ignored_patterns import create_rope
 from nncf.torch.quantization.layers import INT4AsymmetricWeightsDecompressor
 from nncf.torch.quantization.layers import INT4SymmetricWeightsDecompressor
 from nncf.torch.quantization.layers import INT8AsymmetricWeightsDecompressor
@@ -273,6 +275,10 @@ def transform_model(
 
         return transformed_model
 
+    @staticmethod
+    def get_ignored_patterns() -> GraphPattern:
+        return create_rope()
+
 
 class FXMixedPrecisionAlgoBackend(MixedPrecisionAlgoBackend, FXWeightCompressionAlgoBackend):
     @staticmethod

tests/cross_fw/test_templates/template_test_weights_compression.py

@@ -79,6 +79,10 @@ def cast_to(x: TTensor, dtype: TensorDataType) -> TTensor:
     def get_matmul_model() -> TModel:
         """Returns a backend model for test_data_based_criterion."""
 
+    @abstractmethod
+    def get_RoPE_model() -> TModel:
+        """Returns a backend model for test_rope_weight_compression."""
+
     @pytest.mark.parametrize(
         ("mode", "ref_act_score", "ref_score"),
         (
@@ -323,6 +327,24 @@ def test_awq_with_ignored_scope(self):
         int4_num_nodes = self.get_num_int4_nodes(compressed_model)
         assert int4_num_nodes == int4_ref_num_compressed
 
+    def test_rope_weight_compression(self):
+        model = self.get_RoPE_model()
+        sz = 8
+        n_samples = 10
+
+        dataset = Dataset([self.to_tensor(np.ones([1, i + 1, sz], dtype=np.float32)) for i in range(n_samples)])
+        compressed_model = compress_weights(
+            model,
+            mode=CompressWeightsMode.INT4_SYM,
+            ratio=1.0,
+            group_size=-1,
+            dataset=dataset,
+        )
+
+        int4_ref_num_compressed = 0
+        int4_num_nodes = self.get_num_int4_nodes(compressed_model)
+        assert int4_num_nodes == int4_ref_num_compressed
+
     @staticmethod
     @abstractmethod
     def get_reference_for_test_awq_scale_reference() -> dict[str, Tensor]:

tests/openvino/native/quantization/test_weights_compression.py

@@ -71,6 +71,7 @@
 from tests.openvino.native.models import MatMul
 from tests.openvino.native.models import ModelNamedConsts
 from tests.openvino.native.models import OVReferenceModel
+from tests.openvino.native.models import RoPEModel
 from tests.openvino.native.models import SequentialMatmulModel
 from tests.openvino.native.models import WeightsModel
 from tests.openvino.native.quantization.test_fq_params_calculation import REFERENCE_SCALES_DIR
@@ -1754,6 +1755,10 @@ class TestOVTemplateWeightCompression(TemplateWeightCompression):
     def get_matmul_model() -> ov.Model:
         return IdentityMatmul().ov_model
 
+    @staticmethod
+    def get_RoPE_model() -> ov.Model:
+        return RoPEModel().ov_model
+
     @staticmethod
     def get_sequential_matmul_model() -> ov.Model:
         return SequentialMatmulModel().ov_model