Merge branch 'develop' into memory-logger-tool

.github/action_configs/labeler.yml

@@ -1,49 +1,81 @@
 # See help here: https://github.com/marketplace/actions/labeler
 
 dependencies:
-  - 'requirements.txt'
-  - '**/setup.py'
+- any:
+  - changed-files:
+    - any-glob-to-any-file:
+      - 'requirements.txt'
+      - '**/setup.py'
 
 NNCF PT:
-  - 'examples/torch/**/*!(.md)'
-  - 'examples/post_training_quantization/torch/**/*!(.md)'
-  - 'nncf/torch/**/*!(.md)'
-  - 'tests/torch/**/*!(.md)'
-  - 'nncf/quantization/**/torch_backend.py'
+- any:
+  - changed-files:
+    - any-glob-to-any-file:
+      - 'examples/torch/**/*!(.md)'
+      - 'examples/post_training_quantization/torch/**/*!(.md)'
+      - 'nncf/torch/**/*!(.md)'
+      - 'tests/torch/**/*!(.md)'
+      - 'nncf/quantization/**/torch_backend.py'
 
 NNCF TF:
-  - 'examples/tensorflow/**/*!(.md)'
-  - 'examples/post_training_quantization/tensorflow/**/*!(.md)'
-  - 'nncf/tensorflow/**/*!(.md)'
-  - 'tests/tensorflow/**/*!(.md)'
-  - 'nncf/quantization/**/tf_backend.py'
+- any:
+  - changed-files:
+    - any-glob-to-any-file:
+      - 'examples/tensorflow/**/*!(.md)'
+      - 'examples/post_training_quantization/tensorflow/**/*!(.md)'
+      - 'nncf/tensorflow/**/*!(.md)'
+      - 'tests/tensorflow/**/*!(.md)'
+      - 'nncf/quantization/**/tf_backend.py'
 
 NNCF ONNX:
-  - 'examples/onnx/**/*!(.md)'
-  - 'examples/post_training_quantization/onnx/**/*!(.md)'
-  - 'nncf/onnx/**/*!(.md)'
-  - 'tests/onnx/**/*!(.md)'
-  - 'nncf/quantization/**/onnx_backend.py'
+- any:
+  - changed-files:
+    - any-glob-to-any-file:
+      - 'examples/onnx/**/*!(.md)'
+      - 'examples/post_training_quantization/onnx/**/*!(.md)'
+      - 'nncf/onnx/**/*!(.md)'
+      - 'tests/onnx/**/*!(.md)'
+      - 'nncf/quantization/**/onnx_backend.py'
 
 NNCF OpenVINO:
-  - 'examples/openvino/**/*!(.md)'
-  - 'examples/post_training_quantization/openvino/**/*!(.md)'
-  - 'nncf/openvino/**/*!(.md)'
-  - 'tests/openvino/**/*!(.md)'
-  - 'nncf/quantization/**/openvino_backend.py'
+- any:
+  - changed-files:
+    - any-glob-to-any-file:
+      - 'examples/openvino/**/*!(.md)'
+      - 'examples/post_training_quantization/openvino/**/*!(.md)'
+      - 'nncf/openvino/**/*!(.md)'
+      - 'tests/openvino/**/*!(.md)'
+      - 'nncf/quantization/**/openvino_backend.py'
 
 NNCF PTQ:
-  - 'nncf/quantization/**/*!(.md)'
-  - 'tests/post_training/**/*!(.md)'
+- any:
+  - changed-files:
+    - any-glob-to-any-file:
+      - 'nncf/quantization/**/*!(.md)'
+      - 'tests/post_training/**/*!(.md)'
 
 documentation:
-  - '**/*.md'
-  - 'docs/**/*'
+- any:
+  - changed-files:
+    - any-glob-to-any-file:
+      - '**/*.md'
+      - 'docs/**/*'
 
 experimental:
-  - 'nncf/experimental/**/*!(.md)'
+- any:
+  - changed-files:
+    - any-glob-to-any-file:
+      - 'nncf/experimental/**/*!(.md)'
 
 NNCF Common:
-  - 'examples/common/**/*!(.md)'
-  - 'nncf/common/**/*!(.md)'
-  - 'tests/common/**/*!(.md)'
+- any:
+  - changed-files:
+    - any-glob-to-any-file:
+      - 'examples/common/**/*!(.md)'
+      - 'nncf/common/**/*!(.md)'
+      - 'tests/common/**/*!(.md)'
+
+release target:
+- any:
+  - base-branch:
+    - '^release_v*'

.github/workflows/labeler.yml

@@ -10,7 +10,7 @@ jobs:
       pull-requests: write
     runs-on: ubuntu-20.04
     steps:
-      - uses: actions/labeler@ac9175f8a1f3625fd0d4fb234536d26811351594 # v4.3.0
+      - uses: actions/labeler@8558fd74291d67161a8a78ce36a881fa63b766a9 # v5.0.0
         with:
           repo-token: "${{ secrets.GITHUB_TOKEN }}"
           configuration-path: '.github/action_configs/labeler.yml'

README.md

@@ -37,6 +37,7 @@ learning frameworks.
 | :------------------------------------------------------------------------------------------------------- | :-------: | :-------: | :-----------: | :-----------: |
 | [Post-Training Quantization](./docs/usage/post_training_compression/post_training_quantization/Usage.md) | Supported | Supported | Supported     | Supported     |
 | [Weights Compression](./docs/usage/post_training_compression/weights_compression/Usage.md)               | Supported | Supported | Not supported | Not supported |
+| [Activation Sparsity](./nncf/experimental/torch/sparsify_activations/ActivationSparsity.md)              | Not supported | Experimental |Not supported| Not supported |
 
 ### Training-Time Compression Algorithms
 
@@ -370,7 +371,6 @@ A list of notebooks demonstrating OpenVINO conversion and inference together wit
 | [Grammar Error Correction](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/grammar-correction)                                                                                                                                                                                                                        |            Post-Training Quantization             | OpenVINO  |                       NLP, Grammar Correction                        |
 | [LLM Instruction Following](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/llm-question-answering)                                                                                                                                                                                                                   |                Weight Compression                 | OpenVINO  |                      NLP, Instruction Following                      |
 | [Dolly 2.0](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/dolly-2-instruction-following)                                                                                                                                                                                                                            |                Weight Compression                 | OpenVINO  |                      NLP, Instruction Following                      |
-| [Stable-Zephyr-3b](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/stable-zephyr-3b-chatbot)                                                                                                                                                                                                                          |                Weight Compression                 | OpenVINO  |                            NLP, Chat Bot                             |
 | [LLM Chat Bots](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/llm-chatbot)                                                                                                                                                                                                                                          |                Weight Compression                 | OpenVINO  |                            NLP, Chat Bot                             |
 
 ### Post-Training Quantization Examples

docs/usage/post_training_compression/weights_compression/Usage.md

@@ -9,7 +9,7 @@ The Weights Compression algorithm is aimed at compressing the weights of the mod
 #### Supported modes
 
 By default, weights are compressed asymmetrically to 8-bit integer data type - "INT8_ASYM" mode.
-OpenVINO backend also supports 3 modes of mixed precision weight quantization with a 4-bit data type as a primary precision - INT4_SYM, INT4_ASYM, NF4, E2M1. The primary precision in case of INT4_SYM mode is signed 4-bit integer and weights are quantized to it [symmetrically](/docs/usage/training_time_compression/other_algorithms/LegacyQuantization.md#symmetric-quantization) without zero point. In case of INT4_ASYM mode - unsigned 4-bit integer and weight are quantized to it [asymmetrically](/docs/usage/training_time_compression/other_algorithms/LegacyQuantization.md#asymmetric-quantization) with a typical non-fixed zero point. In case of NF4 mode - [nf4](https://arxiv.org/pdf/2305.14314v1.pdf) data type without zero point. In case of E2M1 mode - [e2m1](https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf) data type without zero point and has 8bit [E8M0](https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf) scale.
+OpenVINO backend also supports 4 modes of mixed precision weight quantization with a 4-bit data type as a primary precision - INT4_SYM, INT4_ASYM, NF4, E2M1. The primary precision in case of INT4_SYM mode is signed 4-bit integer and weights are quantized to it [symmetrically](/docs/usage/training_time_compression/other_algorithms/LegacyQuantization.md#symmetric-quantization) without zero point. In case of INT4_ASYM mode - unsigned 4-bit integer and weight are quantized to it [asymmetrically](/docs/usage/training_time_compression/other_algorithms/LegacyQuantization.md#asymmetric-quantization) with a typical non-fixed zero point. In case of NF4 mode - [nf4](https://arxiv.org/pdf/2305.14314v1.pdf) data type without zero point. In case of E2M1 mode - [e2m1](https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf) data type without zero point and has 8bit [E8M0](https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf) scale.
 All 4-bit modes have a grouped quantization support, when small group of weights (e.g. 128) in the channel dimension share quantization parameters (scale).
 All embeddings, convolutions and last linear layers are always compressed to 8-bit integer data type. To quantize embeddings and last linear layers to 4-bit, use `all_layers=True`.
 Percent of the rest layers compressed to 4-bit can be configured by "ratio" parameter. E.g. ratio=0.9 means 90% of layers compressed to the corresponding 4-bit data type and the rest to 8-bit asymmetric integer data type.
@@ -538,5 +538,4 @@ List of notebooks demonstrating OpenVINO conversion and inference together with
 
 - [LLM Instruction Following](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/llm-question-answering)
 - [Dolly 2.0](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/dolly-2-instruction-following)
-- [Stable-Zephyr-3b](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/stable-zephyr-3b-chatbot)
 - [LLM Chat Bots](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/llm-chatbot)

examples/post_training_quantization/tensorflow/mobilenet_v2/main.py

@@ -22,7 +22,6 @@
 
 import nncf
 
-tfds.display_progress_bar(enable=False)
 ROOT = Path(__file__).parent.resolve()
 WEIGHTS_URL = "https://huggingface.co/alexsu52/mobilenet_v2_imagenette/resolve/main/tf_model.h5"
 DATASET_CLASSES = 10

nncf/common/graph/graph.py

@@ -348,20 +348,52 @@ def get_input_edges(self, node: NNCFNode) -> List[NNCFGraphEdge]:
             edges.extend(self._get_edges(from_node, node))
         return sorted(edges, key=lambda x: x.input_port_id)
 
+    def get_input_edge_by_port_id(self, node: NNCFNode, port_id: int) -> NNCFGraphEdge:
+        """
+        Returns the input edge for a given node, where edge.input_port_id == port_id is True.
+
+        :param node: The node for which to retrieve the input edge.
+        :param port_id: The ID of the input port to filter the edges.
+        :return: An input edge connected to the specified input port ID of the
+            given node.
+        """
+        edges = [e for e in self.get_input_edges(node) if e.input_port_id == port_id]
+        if len(edges) == 0:
+            raise nncf.ValidationError(
+                f"Node {node.node_name} does not contain input edge connected to {port_id} port ID."
+            )
+
+        if len(edges) > 1:
+            raise nncf.InternalError(
+                "Unsupported graph. More than one edge was found for a given node by the specified input port ID."
+            )
+        return edges[0]
+
     def get_output_edges(self, node: NNCFNode) -> List[NNCFGraphEdge]:
         """
         Returns edges of output tensors sorted by output port ID.
 
         :param node: Producer node.
-        :return:  List of output edges for the node sorted by output port ID.
+        :return: List of output edges for the node sorted by output port ID.
         """
-
         output_nodes = self.get_next_nodes(node)
         edges = []
         for to_node in output_nodes:
             edges.extend(self._get_edges(node, to_node))
         return sorted(edges, key=lambda x: x.output_port_id)
 
+    def get_output_edges_by_port_id(self, node: NNCFNode, port_id: int) -> List[NNCFGraphEdge]:
+        """
+        Returns a list of output edges for a given node, filtered by the specified
+        output port ID (edge.output_port_id == port_id).
+
+        :param node: The node for which to retrieve the output edges.
+        :param port_id: The ID of the output port to filter the edges.
+        :return: A list of the output edges connected to the specified output port ID
+            of the given node.
+        """
+        return [e for e in self.get_output_edges(node) if e.output_port_id == port_id]
+
     def _get_edges(self, from_node: NNCFNode, to_node: NNCFNode) -> List[NNCFGraphEdge]:
         edges = []
         edge = self.get_edge(from_node, to_node)

nncf/common/graph/patterns/patterns.py

@@ -350,6 +350,9 @@ class HWFusedPatternNames(Enum):
     LINEAR_ARITHMETIC_ACTIVATIONS_ARITHMETIC = PatternDesc("linear_arithmetic_activations_arithmetic")
     LINEAR_BATCH_NORM = PatternDesc("linear_batch_norm")
     LINEAR_BATCH_NORM_ACTIVATIONS = PatternDesc("linear_batch_norm_activations")
+    # MaskRCNN_Resnet_Atrous
+    LINEAR_BATCH_TO_SPACE_SCALE_SHIFT_ACTIVATIONS = PatternDesc("linear_batch_to_space_scale_shift_activations")
+    LINEAR_BATCH_TO_SPACE_ARITHMETIC_ACTIVATIONS = PatternDesc("linear_batch_to_space_arithmetic_activations")
     LINEAR_BATCH_NORM_SCALE_SHIFT_ACTIVATIONS = PatternDesc("linear_batch_norm_scale_shift_activations")
     LINEAR_SCALE_SHIFT_ACTIVATIONS = PatternDesc("linear_scale_shift_activations")
     LINEAR_CONST_MULTIPLY = PatternDesc("linear_const_multiply")