Merge pull request #1122 from openvinotoolkit/CVS-85318-nncf_precision

CVS-85318 nncf precision
openvinotoolkit · Jun 3, 2022 · f5d8097 · f5d8097
2 parents cc11743 + 561daf0
commit f5d8097
Show file tree

Hide file tree

Showing 3 changed files with 20 additions and 7 deletions.
diff --git a/external/mmdetection/detection_tasks/apis/detection/inference_task.py b/external/mmdetection/detection_tasks/apis/detection/inference_task.py
@@ -100,7 +100,7 @@ def __init__(self, task_environment: TaskEnvironment):
 
         # Set default model attributes.
         self._optimization_methods = []
-        self._precision = [ModelPrecision.FP16] if self._config.get('fp16', None) else [ModelPrecision.FP32]
+        self._precision = self._precision_from_config
         self._optimization_type = ModelOptimizationType.MO
 
         # Create and initialize PyTorch model.
@@ -113,6 +113,10 @@ def __init__(self, task_environment: TaskEnvironment):
         self._should_stop = False
         logger.info('Task initialization completed')
 
+    @property
+    def _precision_from_config(self):
+        return [ModelPrecision.FP16] if self._config.get('fp16', None) else [ModelPrecision.FP32]
+
     @property
     def _hyperparams(self):
         return self._task_environment.get_hyper_parameters(OTEDetectionConfig)
@@ -402,7 +406,8 @@ def export(self,
                     model = self._model.cpu()
                 pruning_transformation = OptimizationMethod.FILTER_PRUNING in self._optimization_methods
                 export_model(model, self._config, tempdir, target='openvino',
-                             pruning_transformation=pruning_transformation, precision=self._precision[0].name)
+                             pruning_transformation=pruning_transformation,
+                             precision=self._precision_from_config[0].name)
                 bin_file = [f for f in os.listdir(tempdir) if f.endswith('.bin')][0]
                 xml_file = [f for f in os.listdir(tempdir) if f.endswith('.xml')][0]
                 with open(os.path.join(tempdir, bin_file), "rb") as f:

diff --git a/external/mmdetection/detection_tasks/apis/detection/nncf_task.py b/external/mmdetection/detection_tasks/apis/detection/nncf_task.py
@@ -81,17 +81,17 @@ def _set_attributes_by_hyperparams(self):
         if quantization and pruning:
             self._nncf_preset = "nncf_quantization_pruning"
             self._optimization_methods = [OptimizationMethod.QUANTIZATION, OptimizationMethod.FILTER_PRUNING]
-            self._nncf_precision = [ModelPrecision.INT8]
+            self._precision = [ModelPrecision.INT8]
             return
         if quantization and not pruning:
             self._nncf_preset = "nncf_quantization"
             self._optimization_methods = [OptimizationMethod.QUANTIZATION]
-            self._nncf_precision = [ModelPrecision.INT8]
+            self._precision = [ModelPrecision.INT8]
             return
         if not quantization and pruning:
             self._nncf_preset = "nncf_pruning"
             self._optimization_methods = [OptimizationMethod.FILTER_PRUNING]
-            self._nncf_precision = [ModelPrecision.FP32]
+            self._precision = self._precision_from_config
             return
         raise RuntimeError('Not selected optimization algorithm')
 
@@ -249,7 +249,7 @@ def optimize(
         output_model.model_format = ModelFormat.BASE_FRAMEWORK
         output_model.optimization_type = self._optimization_type
         output_model.optimization_methods = self._optimization_methods
-        output_model.precision = self._nncf_precision
+        output_model.precision = self._precision
 
         self._is_training = False
 

diff --git a/external/mmdetection/tests/test_ote_api.py b/external/mmdetection/tests/test_ote_api.py
@@ -36,7 +36,7 @@
 from ote_sdk.entities.inference_parameters import InferenceParameters
 from ote_sdk.entities.model_template import TaskType, task_type_to_label_domain
 from ote_sdk.entities.metrics import Performance
-from ote_sdk.entities.model import ModelEntity, ModelFormat, ModelOptimizationType
+from ote_sdk.entities.model import ModelEntity, ModelFormat, ModelOptimizationType, ModelPrecision
 from ote_sdk.entities.model_template import parse_model_template
 from ote_sdk.entities.optimization_parameters import OptimizationParameters
 from ote_sdk.entities.resultset import ResultSetEntity
@@ -520,6 +520,14 @@ def end_to_end(
                 print(f'Performance of NNCF model: {nncf_performance.score.value:.4f}')
                 self.check_threshold(validation_performance, nncf_performance, nncf_perf_delta_tolerance,
                     'Too big performance difference after NNCF optimization.')
+
+                # Check whether optimize & export assigns correct model precision
+                nncf_task.export(ExportType.OPENVINO, nncf_model)
+
+                if nncf_task._hyperparams.nncf_optimization.enable_quantization:
+                    assert nncf_model.precision[0] == ModelPrecision.INT8
+                else:
+                    assert nncf_model.precision[0] == nncf_task._precision_from_config[0]
             else:
                 print('Skipped test of OTEDetectionNNCFTask. Required NNCF module.')