ruff format (#3308)

### Changes

Use [ruff format](https://docs.astral.sh/ruff/formatter/) instead of
black

Deviations from black: https://docs.astral.sh/ruff/formatter/black/

.gitignore

@@ -108,7 +108,7 @@ ENV/
 # snapshots
 *.tar
 
-# pylint config is left at dev's discretion, CI uses ruff/isort/black instead
+# pylint config is left at dev's discretion, CI uses ruff instead
 .pylintrc
 
 # VSCode

.pre-commit-config.yaml

@@ -7,12 +7,8 @@ repos:
     hooks:
       - id: ruff
         args: [--fix, --show-fixes]
-
-  - repo: https://github.com/psf/black
-    rev: 24.10.0
-    hooks:
-      - id: black
-        files: '^.*\.py'
+      - id: ruff-format
+        types_or: [ python, pyi ]
 
   - repo: https://github.com/igorshubovych/markdownlint-cli
     rev: v0.43.0

custom_version.py

@@ -43,7 +43,6 @@
 This ensures that `nncf/version.py` remains in its original state after the dynamic versioning process.
 """
 
-
 from __future__ import annotations
 
 import contextlib
@@ -98,7 +97,6 @@ def get_custom_version() -> str:
 
 
 def __getattr__(name: str) -> str:
-
     if name == "version":
         global version
         version = get_custom_version()

docs/styleguide/PyGuide.md

@@ -60,16 +60,16 @@ the [PEP 8 -- Style Guide for Python Code](https://www.python.org/dev/peps/pep-0
 
 ## 2 Automating Code Formatting
 
-To maintain consistency and readability throughout the codebase, we use the [black](https://github.com/psf/black)
-and [ruff](https://docs.astral.sh/ruff/) tools for formatting. Before committing any changes,
-it's important to run a pre-commit command to ensure that the code is properly formatted.
+To maintain consistency and readability throughout the codebase, we use the [ruff](https://docs.astral.sh/ruff/)
+tool for formatting. Before committing any changes, it's important to run a pre-commit command to ensure
+that the code is properly formatted.
 You can use the following commands for this:
 
 ```bash
 make pre-commit
 ```
 
-Also recommend configuring your IDE to run Black and Ruff tools automatically when saving files.
+Also recommend configuring your IDE to run Ruff tools automatically when saving files.
 
 Automatic code formatting is mandatory for all Python files, but you can disable it for specific cases if required:
 
@@ -87,7 +87,7 @@ import b
 import a
 ```
 
-Example for 'black':
+Example for 'ruff format':
 
 ```python
 arr1 = [

examples/llm_compression/openvino/tiny_llama_find_hyperparams/main.py

@@ -135,7 +135,7 @@ def print_results(optimized_model: ov.Model, similarity: float) -> None:
     else:
         print(best_params_info)
     footprint = Path(MODEL_PATH).with_suffix(".bin").stat().st_size
-    print(f"Memory footprint: {footprint / 2**20 :.2f} MB")
+    print(f"Memory footprint: {footprint / 2**20:.2f} MB")
     print(f"Similarity: {similarity:.2f}")
 
 

examples/tensorflow/object_detection/main.py

@@ -320,9 +320,8 @@ def run(config):
     # Create dataset
     train_builder, test_builder = get_dataset_builders(config, strategy.num_replicas_in_sync)
     train_dataset, test_dataset = train_builder.build(), test_builder.build()
-    train_dist_dataset, test_dist_dataset = strategy.experimental_distribute_dataset(
-        train_dataset
-    ), strategy.experimental_distribute_dataset(test_dataset)
+    train_dist_dataset = strategy.experimental_distribute_dataset(train_dataset)
+    test_dist_dataset = strategy.experimental_distribute_dataset(test_dataset)
 
     # Training parameters
     epochs = config.epochs

examples/torch/common/models/classification/resnet_cifar10.py

@@ -169,8 +169,7 @@ def __init__(
             replace_stride_with_dilation = [False, False, False]
         if len(replace_stride_with_dilation) != 3:
             msg = (
-                "replace_stride_with_dilation should be None "
-                f"or a 3-element tuple, got {replace_stride_with_dilation}"
+                f"replace_stride_with_dilation should be None or a 3-element tuple, got {replace_stride_with_dilation}"
             )
             raise ValueError(msg)
         self.groups = groups

examples/torch/semantic_segmentation/metric/confusionmatrix.py

@@ -63,14 +63,14 @@ def add(self, predicted, target):
         assert predicted.shape[0] == target.shape[0], "number of targets and predicted outputs do not match"
 
         if np.ndim(predicted) != 1:
-            assert (
-                predicted.shape[1] == self.num_classes
-            ), "number of predictions does not match size of confusion matrix"
+            assert predicted.shape[1] == self.num_classes, (
+                "number of predictions does not match size of confusion matrix"
+            )
             predicted = np.argmax(predicted, 1)
         else:
-            assert (predicted.max() < self.num_classes) and (
-                predicted.min() >= 0
-            ), "predicted values are not between 0 and k-1"
+            assert (predicted.max() < self.num_classes) and (predicted.min() >= 0), (
+                "predicted values are not between 0 and k-1"
+            )
 
         if np.ndim(target) != 1:
             assert target.shape[1] == self.num_classes, "Onehot target does not match size of confusion matrix"

examples/torch/semantic_segmentation/metric/iou.py

@@ -65,9 +65,9 @@ def add(self, predicted, target):
         """
         # Dimensions check
         assert predicted.size(0) == target.size(0), "number of targets and predicted outputs do not match"
-        assert (
-            predicted.dim() == 3 or predicted.dim() == 4
-        ), "predictions must be of dimension (N, H, W) or (N, K, H, W)"
+        assert predicted.dim() == 3 or predicted.dim() == 4, (
+            "predictions must be of dimension (N, H, W) or (N, K, H, W)"
+        )
         assert target.dim() == 3 or target.dim() == 4, "targets must be of dimension (N, H, W) or (N, K, H, W)"
 
         # If the tensor is in categorical format convert it to integer format

nncf/common/accuracy_aware_training/training_loop.py

@@ -11,6 +11,7 @@
 """
 Implementations of training loops to be used for accuracy aware training.
 """
+
 import pathlib
 from abc import ABC
 from abc import abstractmethod
@@ -321,10 +322,7 @@ def __init__(
         super().__init__(compression_controller)
         self.adaptive_controller = self._get_adaptive_compression_ctrl(compression_controller)
         if self.adaptive_controller is None:
-            msg = (
-                "No compression algorithm supported by the accuracy-aware training "
-                "runner was specified in the config"
-            )
+            msg = "No compression algorithm supported by the accuracy-aware training runner was specified in the config"
             raise nncf.InternalError(msg)
 
         maximal_compression_rate = min(maximal_compression_rate, self.adaptive_controller.maximal_compression_rate)
@@ -606,9 +604,8 @@ def _interpolate_compression_step_update(
         nncf_logger.info(f"Compressed training history: {training_history}")
         training_history[minimal_compression_rate] = runner.maximal_accuracy_drop  # type: ignore
         training_history[maximal_compression_rate] = -full_compression_factor * runner.maximal_accuracy_drop  # type: ignore
-        compression_rates, evaluated_acc_budgets = cast(List[float], training_history.keys()), cast(
-            List[float], training_history.values()
-        )
+        compression_rates = cast(List[float], training_history.keys())
+        evaluated_acc_budgets = cast(List[float], training_history.values())
         interp_kind = "linear" if len(compression_rates) < 4 else "cubic"
         acc_budget_vs_comp_rate_curve = interp1d(compression_rates, evaluated_acc_budgets, kind=interp_kind)
         rate_interval = np.linspace(

nncf/common/deprecation.py

@@ -41,7 +41,6 @@ def deprecated(
     """
 
     def decorator(obj: TObj) -> TObj:
-
         if isinstance(obj, types.FunctionType):
 
             @wraps(obj)

nncf/common/initialization/dataloader.py

@@ -9,6 +9,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Interface for user-defined data usage during the compression algorithm initialization process."""
+
 from abc import ABC
 from abc import abstractmethod
 from typing import Any, Iterator

nncf/common/pruning/model_analysis.py

@@ -31,7 +31,8 @@ def get_position(nodes_list: List[NNCFNode], idx: int) -> Optional[int]:
 
 
 def merge_clusters_for_nodes(
-    nodes_to_merge: List[NNCFNode], clusterization: Clusterization  # type:ignore[type-arg]
+    nodes_to_merge: List[NNCFNode],
+    clusterization: Clusterization,  # type:ignore[type-arg]
 ) -> None:
     """
     Merges clusters to which nodes from nodes_to_merge belongs.

nncf/common/pruning/node_selector.py

@@ -249,7 +249,8 @@ def _check_all_closing_nodes_are_feasible(
         return can_prune_updated
 
     def _check_internal_groups_dim(
-        self, pruned_nodes_clusterization: Clusterization  # type: ignore[type-arg]
+        self,
+        pruned_nodes_clusterization: Clusterization,  # type: ignore[type-arg]
     ) -> Dict[int, PruningAnalysisDecision]:
         """
         Checks pruning dimensions of all nodes in each cluster group are equal and
@@ -310,7 +311,9 @@ def _should_prune_groups_analysis(
         return can_prune_updated
 
     def _filter_groups(
-        self, pruned_nodes_clusterization: Clusterization, can_prune: Dict[int, PruningAnalysisDecision]  # type: ignore[type-arg]
+        self,
+        pruned_nodes_clusterization: Clusterization,  # type: ignore[type-arg]
+        can_prune: Dict[int, PruningAnalysisDecision],
     ) -> None:
         """
         Check whether all nodes in group can be pruned based on user-defined constraints and
@@ -332,12 +335,12 @@ def _filter_groups(
                             cannot_prune_messages.append(message)
 
                 nncf_logger.debug(
-                    f'Could not prune node group [{", ".join(nodes_names)}], '
-                    f'reason: {", ".join(cannot_prune_messages)}.'
+                    f"Could not prune node group [{', '.join(nodes_names)}], "
+                    f"reason: {', '.join(cannot_prune_messages)}."
                 )
                 pruned_nodes_clusterization.delete_cluster(cluster.id)
             else:
-                nncf_logger.debug(f'Node group [{", ".join(nodes_names)}] will be pruned together.')
+                nncf_logger.debug(f"Node group [{', '.join(nodes_names)}] will be pruned together.")
 
     def _is_module_prunable(self, graph: NNCFGraph, node: NNCFNode) -> PruningAnalysisDecision:
         """

nncf/common/pruning/symbolic_mask.py

@@ -41,9 +41,9 @@ def merge_producers(cls, masks: List["SymbolicMask"]) -> List["SymbolicMaskProdu
         for mask in masks:
             for mask_producer in mask.mask_producers:
                 if mask_producer.id in merged_producers:
-                    assert (
-                        mask_producer.sparse_multiplier == merged_producers[mask_producer.id].sparse_multiplier
-                    ), f"Inconsistent sparse multiplier for NNCF node with id={mask_producer.id}"
+                    assert mask_producer.sparse_multiplier == merged_producers[mask_producer.id].sparse_multiplier, (
+                        f"Inconsistent sparse multiplier for NNCF node with id={mask_producer.id}"
+                    )
             merged_producers.update({p.id: p for p in mask.mask_producers})
         return list(merged_producers.values())
 

nncf/common/pruning/utils.py

@@ -233,9 +233,9 @@ def wrap(obj: TObj) -> TObj:
                 if name not in self._op_name_to_op_class:
                     self._op_name_to_op_class[name] = obj
                 else:
-                    assert (
-                        self._op_name_to_op_class[name] == obj
-                    ), "Inconsistent operator type registry - single patched op name maps to multiple metatypes!"
+                    assert self._op_name_to_op_class[name] == obj, (
+                        "Inconsistent operator type registry - single patched op name maps to multiple metatypes!"
+                    )
             return obj
 
         return wrap

nncf/common/quantization/quantizer_propagation/graph.py

@@ -191,7 +191,7 @@ def get_node_keys_by_metatype(self, metatype: Type[OperatorMetatype]) -> List[st
 
     @staticmethod
     def _insertion_point_to_quant_insertion_point(
-        ip: Union[PreHookInsertionPoint, PostHookInsertionPoint]
+        ip: Union[PreHookInsertionPoint, PostHookInsertionPoint],
     ) -> QuantizationInsertionPointBase:
         if isinstance(ip, PreHookInsertionPoint):
             return ActivationQuantizationInsertionPoint(ip.target_node_name, input_port_id=ip.input_port_id)
@@ -1389,7 +1389,8 @@ def create_quantizer_setup(
         )
         for pq_set in pq_sets_grouped_by_unified_scale:
             setup.register_unified_scale_group_with_types(
-                [pqid_vs_qpid[pq.id] for pq in pq_set], [pq.unified_scale_type for pq in pq_set]  # type: ignore
+                [pqid_vs_qpid[pq.id] for pq in pq_set],
+                [pq.unified_scale_type for pq in pq_set],  # type: ignore
             )
 
         setup = self._handle_output_quantizers_for_weights_as_outputs_ops(setup, pqid_vs_qpid, wao_op_node_key_vs_wq_id)

nncf/common/quantization/quantizer_propagation/solver.py

@@ -889,9 +889,8 @@ def _get_trait_for_op_meta_not_specified_in_hw_config(self, op_meta: Type[Operat
     def _get_operator_qconfigs_map(self) -> Dict[Type[OperatorMetatype], Optional[List[QuantizerConfig]]]:
         # TODO (vshampor): ensure that there are no name collisions between ops in different torch subpackages
         #  with the same name
-        retval: Dict[Type[OperatorMetatype], Optional[List[QuantizerConfig]]] = (
-            {}
-        )  # Metas not in retval will correspond to wildcard quantization
+        # Metas not in retval will correspond to wildcard quantization
+        retval: Dict[Type[OperatorMetatype], Optional[List[QuantizerConfig]]] = {}
         if self._hw_config is None:
             for trait, meta_list in self._default_trait_to_metatype_map.items():
                 if trait == QuantizationTrait.INPUTS_QUANTIZABLE:
@@ -1139,9 +1138,9 @@ def _setup_initial_quantizers_for_operator_node(
         for pred_ip_key in preds:
             pred_node = quant_prop_graph.nodes[pred_ip_key]
             pred_node_type = pred_node[QuantizerPropagationStateGraph.NODE_TYPE_NODE_ATTR]
-            assert QuantizerPropagationStateGraph.is_insertion_point(
-                pred_node_type
-            ), "Invalid insertion point graph supplied for quantizer propagation!"
+            assert QuantizerPropagationStateGraph.is_insertion_point(pred_node_type), (
+                "Invalid insertion point graph supplied for quantizer propagation!"
+            )
 
             ip = pred_node[QuantizerPropagationStateGraph.QUANT_INSERTION_POINT_DATA_NODE_ATTR]
             input_port_id = ip.input_port_id
@@ -1457,7 +1456,8 @@ def compatible_wo_requant(qconf: QuantizerConfig, other_qconf_list: List[Quantiz
 
         merged_qconfig_list_counter = Counter(merged_qconfig_list)
         resulting_branch_qconfig_lists: List[List[QuantizerConfig]] = [
-            None for _ in potential_qconfigs_for_each_branch  # type: ignore[misc]
+            None  # type: ignore[misc]
+            for _ in potential_qconfigs_for_each_branch
         ]
 
         if self._propagation_strategy == QuantizerPropagationRule.MERGE_WITH_POTENTIAL_REQUANTIZATION:

nncf/common/scopes.py

@@ -124,8 +124,7 @@ def check_scopes_in_graph(
 
     if not_matched_ignored_scopes or not_matched_target_scopes:
         err_message = (
-            "No match has been found among the model operations "
-            "for the following ignored/target scope definitions:\n"
+            "No match has been found among the model operations for the following ignored/target scope definitions:\n"
         )
         if not_matched_ignored_scopes:
             err_message += f" - ignored_scope: {not_matched_ignored_scopes}\n"

nncf/common/sparsity/statistics.py

@@ -86,8 +86,7 @@ def to_str(self) -> str:
         )
 
         pretty_string = (
-            f"Statistics of the sparsified model:\n{model_string}\n\n"
-            f"Statistics by sparsified layers:\n{layers_string}"
+            f"Statistics of the sparsified model:\n{model_string}\n\nStatistics by sparsified layers:\n{layers_string}"
         )
         return pretty_string
 
@@ -231,7 +230,6 @@ def to_str(self) -> str:
         )
 
         pretty_string = (
-            f"{self.model_statistics.to_str()}\n\n"
-            f"Statistics of the movement-sparsity algorithm:\n{algorithm_string}"
+            f"{self.model_statistics.to_str()}\n\nStatistics of the movement-sparsity algorithm:\n{algorithm_string}"
         )
         return pretty_string

nncf/config/__init__.py

@@ -11,4 +11,5 @@
 """
 NNCF configuration file schemata and associated structures.
 """
+
 from nncf.config.config import NNCFConfig as NNCFConfig

nncf/config/extractors.py

@@ -67,10 +67,7 @@ def extract_algo_specific_config(config: NNCFConfig, algo_name_to_match: str) ->
             matches.append(compression_algo_dict)
 
     if len(matches) > 1:
-        msg = (
-            f"Multiple algorithm configurations specified for the same "
-            f"algo {algo_name_to_match} in the NNCF config!"
-        )
+        msg = f"Multiple algorithm configurations specified for the same algo {algo_name_to_match} in the NNCF config!"
         raise nncf.ValidationError(msg)
     if not matches:
         msg = f"Did not find an algorithm configuration for algo {algo_name_to_match} in the NNCF config!"

nncf/config/schema.py

@@ -169,8 +169,7 @@ def validate_single_compression_algo_schema(
         )
         if algo_name in ALGO_NAME_VS_README_URL:
             e.message += (
-                f"or to the algorithm documentation for examples of the configs: "
-                f"{ALGO_NAME_VS_README_URL[algo_name]}"
+                f"or to the algorithm documentation for examples of the configs: {ALGO_NAME_VS_README_URL[algo_name]}"
             )
         raise e
 

nncf/config/schemata/algo/quantization.py

@@ -298,8 +298,7 @@
                 "items": [NUMBER, STRING],
                 "description": "A tuple of a bitwidth and a scope of the quantizer to assign the bitwidth to.",
             },
-            "description": "Manual settings for the quantizer bitwidths. Scopes are used to identify "
-            "the quantizers.",
+            "description": "Manual settings for the quantizer bitwidths. Scopes are used to identify the quantizers.",
             "examples": [
                 [
                     [2, "ResNet/NNCFConv2d[conv1]/conv2d_0|WEIGHT"],
@@ -463,8 +462,7 @@
         ),
         "quantize_inputs": with_attributes(
             BOOLEAN,
-            description="Whether the model inputs should be immediately quantized prior "
-            "to any other model operations.",
+            description="Whether the model inputs should be immediately quantized prior to any other model operations.",
             default=QUANTIZE_INPUTS,
         ),
         "quantize_outputs": with_attributes(