oracle-ads 2.12.10rc0__py3-none-any.whl → 2.13.0__py3-none-any.whl

ads/llm/langchain/plugins/embeddings/oci_data_science_model_deployment_endpoint.py

@@ -0,0 +1,184 @@
+#!/usr/bin/env python
+
+# Copyright (c) 2025 Oracle and/or its affiliates.
+# Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
+
+from typing import Any, Callable, Dict, List, Mapping, Optional
+
+import requests
+from langchain_core.embeddings import Embeddings
+from langchain_core.language_models.llms import create_base_retry_decorator
+from pydantic import BaseModel, Field
+
+DEFAULT_HEADER = {
+    "Content-Type": "application/json",
+}
+
+
+class TokenExpiredError(Exception):
+    pass
+
+
+def _create_retry_decorator(llm) -> Callable[[Any], Any]:
+    """Creates a retry decorator."""
+    errors = [requests.exceptions.ConnectTimeout, TokenExpiredError]
+    decorator = create_base_retry_decorator(
+        error_types=errors, max_retries=llm.max_retries
+    )
+    return decorator
+
+
+class OCIDataScienceEmbedding(BaseModel, Embeddings):
+    """Embedding model deployed on OCI Data Science Model Deployment.
+
+    Example:
+
+        .. code-block:: python
+
+            from ads.llm import OCIDataScienceEmbedding
+
+            embeddings = OCIDataScienceEmbedding(
+                endpoint="https://modeldeployment.us-ashburn-1.oci.customer-oci.com/<md_ocid>/predict",
+            )
+    """  # noqa: E501
+
+    auth: dict = Field(default_factory=dict, exclude=True)
+    """ADS auth dictionary for OCI authentication:
+    https://accelerated-data-science.readthedocs.io/en/latest/user_guide/cli/authentication.html.
+    This can be generated by calling `ads.common.auth.api_keys()`
+    or `ads.common.auth.resource_principal()`. If this is not
+    provided then the `ads.common.default_signer()` will be used."""
+
+    endpoint: str = ""
+    """The uri of the endpoint from the deployed Model Deployment model."""
+
+    model_kwargs: Optional[Dict] = None
+    """Keyword arguments to pass to the model."""
+
+    endpoint_kwargs: Optional[Dict] = None
+    """Optional attributes (except for headers) passed to the request.post
+    function.
+    """
+
+    max_retries: int = 1
+    """The maximum number of retries to make when generating."""
+
+    @property
+    def _identifying_params(self) -> Mapping[str, Any]:
+        """Get the identifying parameters."""
+        _model_kwargs = self.model_kwargs or {}
+        return {
+            **{"endpoint": self.endpoint},
+            **{"model_kwargs": _model_kwargs},
+        }
+
+    def _embed_with_retry(self, **kwargs) -> Any:
+        """Use tenacity to retry the call."""
+        retry_decorator = _create_retry_decorator(self)
+
+        @retry_decorator
+        def _completion_with_retry(**kwargs: Any) -> Any:
+            try:
+                response = requests.post(self.endpoint, **kwargs)
+                response.raise_for_status()
+                return response
+            except requests.exceptions.HTTPError as http_err:
+                if response.status_code == 401 and self._refresh_signer():
+                    raise TokenExpiredError() from http_err
+                else:
+                    raise ValueError(
+                        f"Server error: {str(http_err)}. Message: {response.text}"
+                    ) from http_err
+            except Exception as e:
+                raise ValueError(f"Error occurs by inference endpoint: {str(e)}") from e
+
+        return _completion_with_retry(**kwargs)
+
+    def _embedding(self, texts: List[str]) -> List[List[float]]:
+        """Call out to OCI Data Science Model Deployment Endpoint.
+
+        Args:
+            texts: A list of texts to embed.
+
+        Returns:
+            A list of list of floats representing the embeddings, or None if an
+            error occurs.
+        """
+        _model_kwargs = self.model_kwargs or {}
+        body = self._construct_request_body(texts, _model_kwargs)
+        request_kwargs = self._construct_request_kwargs(body)
+        response = self._embed_with_retry(**request_kwargs)
+        return self._proceses_response(response)
+
+    def _construct_request_kwargs(self, body: Any) -> dict:
+        """Constructs the request kwargs as a dictionary."""
+        from ads.model.common.utils import _is_json_serializable
+
+        _endpoint_kwargs = self.endpoint_kwargs or {}
+        headers = _endpoint_kwargs.pop("headers", DEFAULT_HEADER)
+        return (
+            dict(
+                headers=headers,
+                json=body,
+                auth=self.auth.get("signer"),
+                **_endpoint_kwargs,
+            )
+            if _is_json_serializable(body)
+            else dict(
+                headers=headers,
+                data=body,
+                auth=self.auth.get("signer"),
+                **_endpoint_kwargs,
+            )
+        )
+
+    def _construct_request_body(self, texts: List[str], params: dict) -> Any:
+        """Constructs the request body."""
+        return {"input": texts}
+
+    def _proceses_response(self, response: requests.Response) -> List[List[float]]:
+        """Extracts results from requests.Response."""
+        try:
+            res_json = response.json()
+            embeddings = res_json["data"][0]["embedding"]
+        except Exception as e:
+            raise ValueError(
+                f"Error raised by inference API: {e}.\nResponse: {response.text}"
+            ) from e
+        return embeddings
+
+    def embed_documents(
+        self,
+        texts: List[str],
+        chunk_size: Optional[int] = None,
+    ) -> List[List[float]]:
+        """Compute doc embeddings using OCI Data Science Model Deployment Endpoint.
+
+        Args:
+            texts: The list of texts to embed.
+            chunk_size: The chunk size defines how many input texts will
+                be grouped together as request. If None, will use the
+                chunk size specified by the class.
+
+        Returns:
+            List of embeddings, one for each text.
+        """
+        results = []
+        _chunk_size = (
+            len(texts) if (not chunk_size or chunk_size > len(texts)) else chunk_size
+        )
+        for i in range(0, len(texts), _chunk_size):
+            response = self._embedding(texts[i : i + _chunk_size])
+            results.extend(response)
+        return results
+
+    def embed_query(self, text: str) -> List[float]:
+        """Compute query embeddings using OCI Data Science Model Deployment Endpoint.
+
+        Args:
+            text: The text to embed.
+
+        Returns:
+            Embeddings for the text.
+        """
+        return self._embedding([text])[0]

ads/model/artifact_downloader.py

@@ -1,5 +1,4 @@
 #!/usr/bin/env python
-# -*- coding: utf-8; -*-
 
 # Copyright (c) 2022, 2024 Oracle and/or its affiliates.
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
@@ -12,9 +11,9 @@ from typing import Dict, Optional
 from zipfile import ZipFile
 
 from ads.common import utils
+from ads.common.object_storage_details import ObjectStorageDetails
 from ads.common.utils import extract_region
 from ads.model.service.oci_datascience_model import OCIDataScienceModel
-from ads.common.object_storage_details import ObjectStorageDetails
 
 
 class ArtifactDownloader(ABC):
@@ -169,9 +168,9 @@ class LargeArtifactDownloader(ArtifactDownloader):
 
     def _download(self):
         """Downloads model artifacts."""
-        self.progress.update(f"Importing model artifacts from catalog")
+        self.progress.update("Importing model artifacts from catalog")
 
-        if self.dsc_model.is_model_by_reference() and self.model_file_description:
+        if self.dsc_model._is_model_by_reference() and self.model_file_description:
             self.download_from_model_file_description()
             self.progress.update()
             return

ads/model/datascience_model.py

@@ -1,7 +1,6 @@
 #!/usr/bin/env python
-# -*- coding: utf-8; -*-
 
-# Copyright (c) 2022, 2024 Oracle and/or its affiliates.
+# Copyright (c) 2022, 2025 Oracle and/or its affiliates.
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
 
 import cgi
@@ -19,12 +18,14 @@ from jsonschema import ValidationError, validate
 
 from ads.common import oci_client as oc
 from ads.common import utils
-from ads.common.extended_enum import ExtendedEnumMeta
+from ads.common.extended_enum import ExtendedEnum
 from ads.common.object_storage_details import ObjectStorageDetails
+from ads.config import (
+    AQUA_SERVICE_MODELS_BUCKET as SERVICE_MODELS_BUCKET,
+)
 from ads.config import (
     COMPARTMENT_OCID,
     PROJECT_OCID,
-    AQUA_SERVICE_MODELS_BUCKET as SERVICE_MODELS_BUCKET,
 )
 from ads.feature_engineering.schema import Schema
 from ads.jobs.builders.base import Builder
@@ -80,14 +81,14 @@ class InvalidArtifactType(Exception):  # pragma: no cover
     pass
 
 
-class CustomerNotificationType(str, metaclass=ExtendedEnumMeta):
+class CustomerNotificationType(ExtendedEnum):
     NONE = "NONE"
     ALL = "ALL"
     ON_FAILURE = "ON_FAILURE"
     ON_SUCCESS = "ON_SUCCESS"
 
 
-class SettingStatus(str, metaclass=ExtendedEnumMeta):
+class SettingStatus(ExtendedEnum):
     """Enum to represent the status of retention settings."""
 
     PENDING = "PENDING"
@@ -116,17 +117,17 @@ class ModelBackupSetting:
     """
 
     def __init__(
-            self,
-            is_backup_enabled: Optional[bool] = None,
-            backup_region: Optional[str] = None,
-            customer_notification_type: Optional[CustomerNotificationType] = None,
+        self,
+        is_backup_enabled: Optional[bool] = None,
+        backup_region: Optional[str] = None,
+        customer_notification_type: Optional[CustomerNotificationType] = None,
     ):
         self.is_backup_enabled = (
             is_backup_enabled if is_backup_enabled is not None else False
         )
         self.backup_region = backup_region
         self.customer_notification_type = (
-                customer_notification_type or CustomerNotificationType.NONE
+            customer_notification_type or CustomerNotificationType.NONE
         )
 
     def to_dict(self) -> Dict:
@@ -143,10 +144,7 @@ class ModelBackupSetting:
         return cls(
             is_backup_enabled=data.get("is_backup_enabled"),
             backup_region=data.get("backup_region"),
-            customer_notification_type=CustomerNotificationType(
-                data.get("customer_notification_type")
-            )
-                                       or None,
+            customer_notification_type=data.get("customer_notification_type") or None,
         )
 
     def to_json(self) -> str:
@@ -166,12 +164,15 @@ class ModelBackupSetting:
 
     def validate(self) -> bool:
         """Validates the backup settings details. Returns True if valid, False otherwise."""
-        return all([
-            isinstance(self.is_backup_enabled, bool),
-            not self.backup_region or isinstance(self.backup_region, str),
-            isinstance(self.customer_notification_type, str) and self.customer_notification_type in
-            CustomerNotificationType.values()
-        ])
+        return all(
+            [
+                isinstance(self.is_backup_enabled, bool),
+                not self.backup_region or isinstance(self.backup_region, str),
+                isinstance(self.customer_notification_type, str)
+                and self.customer_notification_type
+                in CustomerNotificationType.values(),
+            ]
+        )
 
     def __repr__(self):
         return self.to_yaml()
@@ -198,15 +199,15 @@ class ModelRetentionSetting:
     """
 
     def __init__(
-            self,
-            archive_after_days: Optional[int] = None,
-            delete_after_days: Optional[int] = None,
-            customer_notification_type: Optional[CustomerNotificationType] = None,
+        self,
+        archive_after_days: Optional[int] = None,
+        delete_after_days: Optional[int] = None,
+        customer_notification_type: Optional[CustomerNotificationType] = None,
     ):
         self.archive_after_days = archive_after_days
         self.delete_after_days = delete_after_days
         self.customer_notification_type = (
-                customer_notification_type or CustomerNotificationType.NONE
+            customer_notification_type or CustomerNotificationType.NONE
         )
 
     def to_dict(self) -> Dict:
@@ -223,10 +224,7 @@ class ModelRetentionSetting:
         return cls(
             archive_after_days=data.get("archive_after_days"),
             delete_after_days=data.get("delete_after_days"),
-            customer_notification_type=CustomerNotificationType(
-                data.get("customer_notification_type")
-            )
-                                       or None,
+            customer_notification_type=data.get("customer_notification_type") or None,
         )
 
     def to_json(self) -> str:
@@ -245,13 +243,23 @@ class ModelRetentionSetting:
 
     def validate(self) -> bool:
         """Validates the retention settings details. Returns True if valid, False otherwise."""
-        return all([
-            self.archive_after_days is None or (
-                    isinstance(self.archive_after_days, int) and self.archive_after_days >= 0),
-            self.delete_after_days is None or (isinstance(self.delete_after_days, int) and self.delete_after_days >= 0),
-            isinstance(self.customer_notification_type, str) and self.customer_notification_type in
-            CustomerNotificationType.values()
-        ])
+        return all(
+            [
+                self.archive_after_days is None
+                or (
+                    isinstance(self.archive_after_days, int)
+                    and self.archive_after_days >= 0
+                ),
+                self.delete_after_days is None
+                or (
+                    isinstance(self.delete_after_days, int)
+                    and self.delete_after_days >= 0
+                ),
+                isinstance(self.customer_notification_type, str)
+                and self.customer_notification_type
+                in CustomerNotificationType.values(),
+            ]
+        )
 
     def __repr__(self):
         return self.to_yaml()
@@ -278,13 +286,13 @@ class ModelRetentionOperationDetails:
     """
 
     def __init__(
-            self,
-            archive_state: Optional[SettingStatus] = None,
-            archive_state_details: Optional[str] = None,
-            delete_state: Optional[SettingStatus] = None,
-            delete_state_details: Optional[str] = None,
-            time_archival_scheduled: Optional[int] = None,
-            time_deletion_scheduled: Optional[int] = None,
+        self,
+        archive_state: Optional[SettingStatus] = None,
+        archive_state_details: Optional[str] = None,
+        delete_state: Optional[SettingStatus] = None,
+        delete_state_details: Optional[str] = None,
+        time_archival_scheduled: Optional[int] = None,
+        time_deletion_scheduled: Optional[int] = None,
     ):
         self.archive_state = archive_state
         self.archive_state_details = archive_state_details
@@ -308,9 +316,9 @@ class ModelRetentionOperationDetails:
     def from_dict(cls, data: Dict) -> "ModelRetentionOperationDetails":
         """Constructs retention operation details from a dictionary."""
         return cls(
-            archive_state=SettingStatus(data.get("archive_state")) or None,
+            archive_state=data.get("archive_state") or None,
             archive_state_details=data.get("archive_state_details"),
-            delete_state=SettingStatus(data.get("delete_state")) or None,
+            delete_state=data.get("delete_state") or None,
             delete_state_details=data.get("delete_state_details"),
             time_archival_scheduled=data.get("time_archival_scheduled"),
             time_deletion_scheduled=data.get("time_deletion_scheduled"),
@@ -334,8 +342,10 @@ class ModelRetentionOperationDetails:
         """Validates the retention operation details."""
         return all(
             [
-                self.archive_state is None or self.archive_state in SettingStatus.values(),
-                self.delete_state is None or self.delete_state in SettingStatus.values(),
+                self.archive_state is None
+                or self.archive_state in SettingStatus.values(),
+                self.delete_state is None
+                or self.delete_state in SettingStatus.values(),
                 self.time_archival_scheduled is None
                 or isinstance(self.time_archival_scheduled, int),
                 self.time_deletion_scheduled is None
@@ -368,10 +378,10 @@ class ModelBackupOperationDetails:
     """
 
     def __init__(
-            self,
-            backup_state: Optional[SettingStatus] = None,
-            backup_state_details: Optional[str] = None,
-            time_last_backup: Optional[int] = None,
+        self,
+        backup_state: Optional[SettingStatus] = None,
+        backup_state_details: Optional[str] = None,
+        time_last_backup: Optional[int] = None,
     ):
         self.backup_state = backup_state
         self.backup_state_details = backup_state_details
@@ -389,7 +399,7 @@ class ModelBackupOperationDetails:
     def from_dict(cls, data: Dict) -> "ModelBackupOperationDetails":
         """Constructs backup operation details from a dictionary."""
         return cls(
-            backup_state=SettingStatus(data.get("backup_state")) or None,
+            backup_state=data.get("backup_state") or None,
             backup_state_details=data.get("backup_state_details"),
             time_last_backup=data.get("time_last_backup"),
         )
@@ -411,8 +421,14 @@ class ModelBackupOperationDetails:
     def validate(self) -> bool:
         """Validates the backup operation details."""
         return not (
-                (self.backup_state is not None and self.backup_state not in SettingStatus.values()) or
-                (self.time_last_backup is not None and not isinstance(self.time_last_backup, int))
+            (
+                self.backup_state is not None
+                and self.backup_state not in SettingStatus.values()
+            )
+            or (
+                self.time_last_backup is not None
+                and not isinstance(self.time_last_backup, int)
+            )
         )
 
     def __repr__(self):
@@ -1042,7 +1058,7 @@ class DataScienceModel(Builder):
         elif json_string:
             json_data = json.loads(json_string)
         elif json_uri:
-            with open(json_uri, "r") as json_file:
+            with open(json_uri) as json_file:
                 json_data = json.load(json_file)
         else:
             raise ValueError("Must provide either a valid json string or URI location.")
@@ -1077,7 +1093,7 @@ class DataScienceModel(Builder):
         return self.get_spec(self.CONST_RETENTION_SETTING)
 
     def with_retention_setting(
-            self, retention_setting: Union[Dict, ModelRetentionSetting]
+        self, retention_setting: Union[Dict, ModelRetentionSetting]
     ) -> "DataScienceModel":
         """
         Sets the retention setting details for the model.
@@ -1106,7 +1122,7 @@ class DataScienceModel(Builder):
         return self.get_spec(self.CONST_BACKUP_SETTING)
 
     def with_backup_setting(
-            self, backup_setting: Union[Dict, ModelBackupSetting]
+        self, backup_setting: Union[Dict, ModelBackupSetting]
     ) -> "DataScienceModel":
         """
         Sets the model's backup setting details.
@@ -1368,8 +1384,8 @@ class DataScienceModel(Builder):
             shutil.rmtree(self.local_copy_dir, ignore_errors=True)
 
     def restore_model(
-            self,
-            restore_model_for_hours_specified: Optional[int] = None,
+        self,
+        restore_model_for_hours_specified: Optional[int] = None,
     ) -> None:
         """
         Restore archived model artifact.
@@ -1398,8 +1414,12 @@ class DataScienceModel(Builder):
 
         # Optional: Validate restore_model_for_hours_specified
         if restore_model_for_hours_specified is not None and (
-                not isinstance(restore_model_for_hours_specified, int) or restore_model_for_hours_specified <= 0):
-            raise ValueError("restore_model_for_hours_specified must be a positive integer.")
+            not isinstance(restore_model_for_hours_specified, int)
+            or restore_model_for_hours_specified <= 0
+        ):
+            raise ValueError(
+                "restore_model_for_hours_specified must be a positive integer."
+            )
 
         self.dsc_model.restore_archived_model_artifact(
             restore_model_for_hours_specified=restore_model_for_hours_specified,
@@ -1721,7 +1741,7 @@ class DataScienceModel(Builder):
             self.CONST_BACKUP_SETTING: ModelBackupSetting.to_dict,
             self.CONST_RETENTION_SETTING: ModelRetentionSetting.to_dict,
             self.CONST_BACKUP_OPERATION_DETAILS: ModelBackupOperationDetails.to_dict,
-            self.CONST_RETENTION_OPERATION_DETAILS: ModelRetentionOperationDetails.to_dict
+            self.CONST_RETENTION_OPERATION_DETAILS: ModelRetentionOperationDetails.to_dict,
         }
 
         # Update the main properties
@@ -1758,7 +1778,7 @@ class DataScienceModel(Builder):
             artifact_info = self.dsc_model.get_artifact_info()
             _, file_name_info = cgi.parse_header(artifact_info["Content-Disposition"])
 
-            if self.dsc_model.is_model_by_reference():
+            if self.dsc_model._is_model_by_reference():
                 _, file_extension = os.path.splitext(file_name_info["filename"])
                 if file_extension.lower() == ".json":
                     bucket_uri, _ = self._download_file_description_artifact()

ads/model/generic_model.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python
 
-# Copyright (c) 2022, 2024 Oracle and/or its affiliates.
+# Copyright (c) 2022, 2025 Oracle and/or its affiliates.
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
 
 import inspect
@@ -63,7 +63,7 @@ from ads.model.model_introspect import (
     ModelIntrospect,
 )
 from ads.model.model_metadata import (
-    ExtendedEnumMeta,
+    ExtendedEnum,
     Framework,
     MetadataCustomCategory,
     ModelCustomMetadata,
@@ -146,7 +146,7 @@ class ModelDeploymentRuntimeType:
     CONTAINER = "container"
 
 
-class DataScienceModelType(str, metaclass=ExtendedEnumMeta):
+class DataScienceModelType(ExtendedEnum):
     MODEL_DEPLOYMENT = "datasciencemodeldeployment"
     MODEL = "datasciencemodel"
 

ads/model/model_metadata.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python
 
-# Copyright (c) 2021, 2024 Oracle and/or its affiliates.
+# Copyright (c) 2021, 2025 Oracle and/or its affiliates.
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
 
 import json
@@ -21,7 +21,7 @@ from oci.util import to_dict
 
 from ads.common import logger
 from ads.common.error import ChangesNotCommitted
-from ads.common.extended_enum import ExtendedEnumMeta
+from ads.common.extended_enum import ExtendedEnum
 from ads.common.object_storage_details import ObjectStorageDetails
 from ads.common.serializer import DataClassSerializable
 from ads.dataset import factory
@@ -81,19 +81,19 @@ class MetadataDescriptionTooLong(ValueError):
         )
 
 
-class MetadataCustomPrintColumns(str, metaclass=ExtendedEnumMeta):
+class MetadataCustomPrintColumns(ExtendedEnum):
     KEY = "Key"
     VALUE = "Value"
     DESCRIPTION = "Description"
     CATEGORY = "Category"
 
 
-class MetadataTaxonomyPrintColumns(str, metaclass=ExtendedEnumMeta):
+class MetadataTaxonomyPrintColumns(ExtendedEnum):
     KEY = "Key"
     VALUE = "Value"
 
 
-class MetadataTaxonomyKeys(str, metaclass=ExtendedEnumMeta):
+class MetadataTaxonomyKeys(ExtendedEnum):
     USE_CASE_TYPE = "UseCaseType"
     FRAMEWORK = "Framework"
     FRAMEWORK_VERSION = "FrameworkVersion"
@@ -102,7 +102,7 @@ class MetadataTaxonomyKeys(str, metaclass=ExtendedEnumMeta):
     ARTIFACT_TEST_RESULT = "ArtifactTestResults"
 
 
-class MetadataCustomKeys(str, metaclass=ExtendedEnumMeta):
+class MetadataCustomKeys(ExtendedEnum):
     SLUG_NAME = "SlugName"
     CONDA_ENVIRONMENT = "CondaEnvironment"
     CONDA_ENVIRONMENT_PATH = "CondaEnvironmentPath"
@@ -121,7 +121,7 @@ class MetadataCustomKeys(str, metaclass=ExtendedEnumMeta):
     MODEL_FILE_NAME = "ModelFileName"
 
 
-class MetadataCustomCategory(str, metaclass=ExtendedEnumMeta):
+class MetadataCustomCategory(ExtendedEnum):
     PERFORMANCE = "Performance"
     TRAINING_PROFILE = "Training Profile"
     TRAINING_AND_VALIDATION_DATASETS = "Training and Validation Datasets"
@@ -129,7 +129,7 @@ class MetadataCustomCategory(str, metaclass=ExtendedEnumMeta):
     OTHER = "Other"
 
 
-class UseCaseType(str, metaclass=ExtendedEnumMeta):
+class UseCaseType(ExtendedEnum):
     BINARY_CLASSIFICATION = "binary_classification"
     REGRESSION = "regression"
     MULTINOMIAL_CLASSIFICATION = "multinomial_classification"
@@ -146,7 +146,7 @@ class UseCaseType(str, metaclass=ExtendedEnumMeta):
     OTHER = "other"
 
 
-class Framework(str, metaclass=ExtendedEnumMeta):
+class Framework(ExtendedEnum):
     SCIKIT_LEARN = "scikit-learn"
     XGBOOST = "xgboost"
     TENSORFLOW = "tensorflow"
@@ -1509,7 +1509,10 @@ class ModelTaxonomyMetadata(ModelMetadata):
         metadata = cls()
         for oci_item in metadata_list:
             item = ModelTaxonomyMetadataItem._from_oci_metadata(oci_item)
-            metadata[item.key].update(value=item.value)
+            if item.key in metadata.keys:
+                metadata[item.key].update(value=item.value)
+            else:
+                metadata._items.add(item)
         return metadata
 
     def to_dataframe(self) -> pd.DataFrame:
@@ -1562,7 +1565,10 @@ class ModelTaxonomyMetadata(ModelMetadata):
         metadata = cls()
         for item in data["data"]:
             item = ModelTaxonomyMetadataItem.from_dict(item)
-            metadata[item.key].update(value=item.value)
+            if item.key in metadata.keys:
+                metadata[item.key].update(value=item.value)
+            else:
+                metadata._items.add(item)
         return metadata