azure-ai-evaluation 1.0.1__py3-none-any.whl → 1.13.3__py3-none-any.whl

azure/ai/evaluation/_evaluate/_eval_run.py

@@ -13,7 +13,7 @@ import uuid
 from typing import Any, Dict, List, Optional, Set, Type
 from urllib.parse import urlparse
 
-from promptflow._sdk.entities import Run
+from azure.ai.evaluation._legacy._adapters.entities import Run
 from typing_extensions import Self
 
 from azure.ai.evaluation._exceptions import ErrorBlame, ErrorCategory, ErrorTarget, EvaluationException
@@ -22,29 +22,12 @@ from azure.ai.evaluation._version import VERSION
 from azure.core.pipeline.policies import RetryPolicy
 from azure.core.rest import HttpResponse
 from azure.core.exceptions import HttpResponseError
+from azure.storage.blob import BlobServiceClient
+from azure.ai.evaluation._azure._clients import LiteMLClient
 
 LOGGER = logging.getLogger(__name__)
 
 
-# Handle optional import. The azure libraries are only present if
-# promptflow-azure is installed.
-try:
-    from azure.ai.ml import MLClient
-    from azure.ai.ml.entities._credentials import AccountKeyConfiguration  # pylint: disable=ungrouped-imports
-    from azure.ai.ml.entities._datastore.datastore import Datastore
-    from azure.storage.blob import BlobServiceClient
-except (ModuleNotFoundError, ImportError):
-    raise EvaluationException(  # pylint: disable=raise-missing-from
-        message=(
-            "The required packages for remote tracking are missing.\n"
-            'To resolve this, please install them by running "pip install azure-ai-evaluation[remote]".'
-        ),
-        target=ErrorTarget.EVALUATE,
-        category=ErrorCategory.MISSING_PACKAGE,
-        blame=ErrorBlame.USER_ERROR,
-    )
-
-
 @dataclasses.dataclass
 class RunInfo:
     """
@@ -93,15 +76,18 @@ class EvalRun(contextlib.AbstractContextManager):  # pylint: disable=too-many-in
     :type group_name: str
     :param workspace_name: The name of workspace/project used to track run.
     :type workspace_name: str
-    :param ml_client: The ml client used for authentication into Azure.
-    :type ml_client: azure.ai.ml.MLClient
+    :param management_client: The trace destination string to parse the AI ML workspace blob store from.
+    :type management_client:
+        ~azure.ai.evaluation._promptflow.azure._lite_azure_management_client.LiteMLClient
     :param promptflow_run: The promptflow run used by the
+    :type promptflow_run: Optional[promptflow._sdk.entities.Run]
+    :param tags: A dictionary of tags to be added to the evaluation run for tracking and organization purposes.
+    :type tags: Optional[Dict[str, str]]
     """
 
     _MAX_RETRIES = 5
     _BACKOFF_FACTOR = 2
     _TIMEOUT = 5
-    _SCOPE = "https://management.azure.com/.default"
 
     EVALUATION_ARTIFACT = "instance_results.jsonl"
 
@@ -112,17 +98,19 @@ class EvalRun(contextlib.AbstractContextManager):  # pylint: disable=too-many-in
         subscription_id: str,
         group_name: str,
         workspace_name: str,
-        ml_client: "MLClient",
+        management_client: LiteMLClient,
         promptflow_run: Optional[Run] = None,
+        tags: Optional[Dict[str, str]] = None,
     ) -> None:
         self._tracking_uri: str = tracking_uri
         self._subscription_id: str = subscription_id
         self._resource_group_name: str = group_name
         self._workspace_name: str = workspace_name
-        self._ml_client: Any = ml_client
+        self._management_client: LiteMLClient = management_client
         self._is_promptflow_run: bool = promptflow_run is not None
         self._run_name = run_name
         self._promptflow_run = promptflow_run
+        self._tags = tags or {}
         self._status = RunStatus.NOT_STARTED
         self._url_base: Optional[str] = None
         self._info: Optional[RunInfo] = None
@@ -184,16 +172,25 @@ class EvalRun(contextlib.AbstractContextManager):  # pylint: disable=too-many-in
             if self._promptflow_run is not None:
                 self._info = RunInfo(
                     self._promptflow_run.name,
-                    self._promptflow_run._experiment_name,  # pylint: disable=protected-access
+                    self._promptflow_run._experiment_name or "",  # pylint: disable=protected-access
                     self._promptflow_run.name,
                 )
             else:
                 url = f"https://{self._url_base}/mlflow/v2.0" f"{self._get_scope()}/api/2.0/mlflow/runs/create"
+
+                # Prepare tags: start with user tags, ensure mlflow.user is set
+                run_tags = self._tags.copy()
+                if "mlflow.user" not in run_tags:
+                    run_tags["mlflow.user"] = "azure-ai-evaluation"
+
+                # Convert tags to MLflow format
+                tags_list = [{"key": key, "value": value} for key, value in run_tags.items()]
+
                 body = {
                     "experiment_id": "0",
                     "user_id": "azure-ai-evaluation",
                     "start_time": int(time.time() * 1000),
-                    "tags": [{"key": "mlflow.user", "value": "azure-ai-evaluation"}],
+                    "tags": tags_list,
                 }
                 if self._run_name:
                     body["run_name"] = self._run_name
@@ -310,12 +307,8 @@ class EvalRun(contextlib.AbstractContextManager):  # pylint: disable=too-many-in
         """
         return f"https://{self._url_base}" "/mlflow/v2.0" f"{self._get_scope()}" f"/api/2.0/mlflow/runs/log-metric"
 
-    def _get_token(self):
-        # We have to use lazy import because promptflow.azure
-        # is an optional dependency.
-        from promptflow.azure._utils._token_cache import ArmTokenCache  # pylint: disable=import-error,no-name-in-module
-
-        return ArmTokenCache().get_token(self._ml_client._credential)  # pylint: disable=protected-access
+    def _get_token(self) -> str:
+        return self._management_client.get_token().token
 
     def request_with_retry(
         self, url: str, method: str, json_dict: Dict[str, Any], headers: Optional[Dict[str, str]] = None
@@ -424,7 +417,7 @@ class EvalRun(contextlib.AbstractContextManager):  # pylint: disable=too-many-in
             LOGGER.warning("The run results file was not found, skipping artifacts upload.")
             return
         # First we will list the files and the appropriate remote paths for them.
-        root_upload_path = posixpath.join("promptflow", "PromptFlowArtifacts", self.info.run_name)
+        root_upload_path = posixpath.join("promptflow", "PromptFlowArtifacts", self.info.run_id)
         remote_paths: Dict[str, List[Dict[str, str]]] = {"paths": []}
         local_paths = []
         # Go over the artifact folder and upload all artifacts.
@@ -441,9 +434,12 @@ class EvalRun(contextlib.AbstractContextManager):  # pylint: disable=too-many-in
                 local_paths.append(local_file_path)
 
         # We will write the artifacts to the workspaceblobstore
-        datastore = self._ml_client.datastores.get_default(include_secrets=True)
+        datastore = self._management_client.workspace_get_default_datastore(
+            self._workspace_name, include_credentials=True
+        )
         account_url = f"{datastore.account_name}.blob.{datastore.endpoint}"
-        svc_client = BlobServiceClient(account_url=account_url, credential=self._get_datastore_credential(datastore))
+
+        svc_client = BlobServiceClient(account_url=account_url, credential=datastore.credential)
         try:
             for local, remote in zip(local_paths, remote_paths["paths"]):
                 blob_client = svc_client.get_blob_client(container=datastore.container_name, blob=remote["path"])
@@ -515,16 +511,6 @@ class EvalRun(contextlib.AbstractContextManager):  # pylint: disable=too-many-in
         except Exception as ex:  # pylint: disable=broad-exception-caught
             LOGGER.debug("Exception occurred while registering image artifact. ex: %s", ex)
 
-    def _get_datastore_credential(self, datastore: "Datastore"):
-        # Reference the logic in azure.ai.ml._artifact._artifact_utilities
-        # https://github.com/Azure/azure-sdk-for-python/blob/main/sdk/ml/azure-ai-ml/azure/ai/ml/_artifacts/_artifact_utilities.py#L103
-        credential = datastore.credentials
-        if isinstance(credential, AccountKeyConfiguration):
-            return credential.account_key
-        if hasattr(credential, "sas_token"):
-            return credential.sas_token
-        return self._ml_client.datastores._credential  # pylint: disable=protected-access
-
     def log_metric(self, key: str, value: float) -> None:
         """
         Log the metric to azure similar to how it is done by mlflow.