[Model Monitoring] Fix model endpoint labels and tag enrichment

mlrun/model_monitoring/stream_processing.py

@@ -917,6 +917,9 @@ def __init__(self, project: str, model_endpoint_store_target: str, **kwargs):
         self.model_endpoint_store_target = model_endpoint_store_target
 
     def do(self, event: dict):
+        # Remove labels from the event
+        event.pop(EventFieldType.LABELS)
+
         update_endpoint_record(
             project=self.project,
             endpoint_id=event.pop(EventFieldType.ENDPOINT_ID),

mlrun/serving/v2_serving.py

@@ -528,7 +528,11 @@ def _init_endpoint_record(
         return None
 
     # Generating version model value based on the model name and model version
-    if model.version:
+    if model.model_path and model.model_path.startswith("store://"):
+        # Enrich the model server with the model artifact metadata
+        model.get_model()
+        model.version = model.model_spec.tag
+        model.labels = model.model_spec.labels
         versioned_model_name = f"{model.name}:{model.version}"
     else:
         versioned_model_name = f"{model.name}:latest"

tests/system/model_monitoring/test_model_monitoring.py

@@ -250,7 +250,8 @@ class TestBasicModelMonitoring(TestMLRunSystem):
     def test_basic_model_monitoring(self) -> None:
         # Main validations:
         # 1 - a single model endpoint is created
-        # 2 - stream metrics are recorded as expected under the model endpoint
+        # 2 - model name, tag and values are recorded as expected under the model endpoint
+        # 3 - stream metrics are recorded as expected under the model endpoint
 
         # Deploy Model Servers
         project = self.project
@@ -278,6 +279,8 @@ def test_basic_model_monitoring(self) -> None:
         )
 
         model_name = "sklearn_RandomForestClassifier"
+        tag = "some-tag"
+        labels = {"framework": "sklearn", "mylabel": "l1"}
 
         # Upload the model through the projects API so that it is available to the serving function
         project.log_model(
@@ -286,12 +289,14 @@ def test_basic_model_monitoring(self) -> None:
             model_file="model.pkl",
             training_set=train_set,
             artifact_path=f"v3io:///projects/{project.metadata.name}",
+            tag=tag,
+            labels=labels,
         )
         # Add the model to the serving function's routing spec
         serving_fn.add_model(
             model_name,
             model_path=project.get_artifact_uri(
-                key=model_name, category="model", tag="latest"
+                key=model_name, category="model", tag=tag
             ),
         )
         if self.image is not None:
@@ -310,18 +315,34 @@ def test_basic_model_monitoring(self) -> None:
             )
             sleep(choice([0.01, 0.04]))
 
-        # Test metrics
         sleep(5)
-        self._assert_model_endpoint_metrics()
-
-    def _assert_model_endpoint_metrics(self) -> None:
         endpoints_list = mlrun.get_run_db().list_model_endpoints(
             self.project_name, metrics=["predictions_per_second"]
         )
         assert len(endpoints_list) == 1
 
         endpoint = endpoints_list[0]
 
+        self._assert_model_endpoint_tags_and_labels(
+            endpoint=endpoint, model_name=model_name, tag=tag, labels=labels
+        )
+
+        # Test metrics
+        self._assert_model_endpoint_metrics(endpoint=endpoint)
+
+    def _assert_model_endpoint_tags_and_labels(
+        self,
+        endpoint: mlrun.model_monitoring.model_endpoint.ModelEndpoint,
+        model_name: str,
+        tag: str,
+        labels: dict[str, str],
+    ) -> None:
+        assert endpoint.metadata.labels == labels
+        assert endpoint.spec.model == f"{model_name}:{tag}"
+
+    def _assert_model_endpoint_metrics(
+        self, endpoint: mlrun.model_monitoring.model_endpoint.ModelEndpoint
+    ) -> None:
         assert len(endpoint.status.metrics) > 0
         self._logger.debug("Model endpoint metrics", endpoint.status.metrics)