Merge branch 'main' into ahosler-patch-1

Author: ahosler

ads/aqua/app.py

@@ -6,14 +6,14 @@
 import os
 import traceback
 from dataclasses import fields
-from typing import Dict, Union
+from typing import Dict, Optional, Union
 
 import oci
 from oci.data_science.models import UpdateModelDetails, UpdateModelProvenanceDetails
 
 from ads import set_auth
 from ads.aqua import logger
-from ads.aqua.common.enums import Tags
+from ads.aqua.common.enums import ConfigFolder, Tags
 from ads.aqua.common.errors import AquaRuntimeError, AquaValueError
 from ads.aqua.common.utils import (
     _is_valid_mvs,
@@ -268,7 +268,12 @@ def if_artifact_exist(self, model_id: str, **kwargs) -> bool:
                 logger.info(f"Artifact not found in model {model_id}.")
                 return False
 
-    def get_config(self, model_id: str, config_file_name: str) -> Dict:
+    def get_config(
+        self,
+        model_id: str,
+        config_file_name: str,
+        config_folder: Optional[str] = ConfigFolder.CONFIG,
+    ) -> Dict:
         """Gets the config for the given Aqua model.
 
         Parameters
@@ -277,12 +282,17 @@ def get_config(self, model_id: str, config_file_name: str) -> Dict:
             The OCID of the Aqua model.
         config_file_name: str
             name of the config file
+        config_folder: (str, optional):
+            subfolder path where config_file_name needs to be searched
+             Defaults to `ConfigFolder.CONFIG`.
+             When searching inside model artifact directory , the value is ConfigFolder.ARTIFACT`
 
         Returns
         -------
         Dict:
             A dict of allowed configs.
         """
+        config_folder = config_folder or ConfigFolder.CONFIG
         oci_model = self.ds_client.get_model(model_id).data
         oci_aqua = (
             (
@@ -304,23 +314,26 @@ def get_config(self, model_id: str, config_file_name: str) -> Dict:
                 f"Base model found for the model: {oci_model.id}. "
                 f"Loading {config_file_name} for base model {base_model_ocid}."
             )
-            base_model = self.ds_client.get_model(base_model_ocid).data
-            artifact_path = get_artifact_path(base_model.custom_metadata_list)
+            if config_folder == ConfigFolder.ARTIFACT:
+                artifact_path = get_artifact_path(oci_model.custom_metadata_list)
+            else:
+                base_model = self.ds_client.get_model(base_model_ocid).data
+                artifact_path = get_artifact_path(base_model.custom_metadata_list)
         else:
             logger.info(f"Loading {config_file_name} for model {oci_model.id}...")
             artifact_path = get_artifact_path(oci_model.custom_metadata_list)
-
         if not artifact_path:
             logger.debug(
                 f"Failed to get artifact path from custom metadata for the model: {model_id}"
             )
             return config
 
-        config_path = f"{os.path.dirname(artifact_path)}/config/"
+        config_path = os.path.join(os.path.dirname(artifact_path), config_folder)
         if not is_path_exists(config_path):
-            config_path = f"{artifact_path.rstrip('/')}/config/"
-
-        config_file_path = f"{config_path}{config_file_name}"
+            config_path = os.path.join(artifact_path.rstrip("/"), config_folder)
+            if not is_path_exists(config_path):
+                config_path = f"{artifact_path.rstrip('/')}/"
+        config_file_path = os.path.join(config_path, config_file_name)
         if is_path_exists(config_file_path):
             try:
                 config = load_config(

ads/aqua/common/enums.py

@@ -92,3 +92,8 @@ class TextEmbeddingInferenceContainerParams(ExtendedEnum):
 
     MODEL_ID = "model-id"
     PORT = "port"
+
+
+class ConfigFolder(ExtendedEnum):
+    CONFIG = "config"
+    ARTIFACT = "artifact"

ads/aqua/constants.py

@@ -10,6 +10,7 @@
 README = "README.md"
 LICENSE_TXT = "config/LICENSE.txt"
 DEPLOYMENT_CONFIG = "deployment_config.json"
+AQUA_MODEL_TOKENIZER_CONFIG = "tokenizer_config.json"
 COMPARTMENT_MAPPING_KEY = "service-model-compartment"
 CONTAINER_INDEX = "container_index.json"
 EVALUATION_REPORT_JSON = "report.json"

ads/aqua/extension/model_handler.py

@@ -14,6 +14,7 @@
 from ads.aqua.common.errors import AquaRuntimeError, AquaValueError
 from ads.aqua.common.utils import (
     get_hf_model_info,
+    is_valid_ocid,
     list_hf_models,
 )
 from ads.aqua.extension.base_handler import AquaAPIhandler
@@ -316,8 +317,30 @@ def post(self, *args, **kwargs):  # noqa: ARG002
         )
 
 
+class AquaModelTokenizerConfigHandler(AquaAPIhandler):
+    def get(self, model_id):
+        """
+        Handles requests for retrieving the Hugging Face tokenizer configuration of a specified model.
+        Expected request format: GET /aqua/models/<model-ocid>/tokenizer
+
+        """
+
+        path_list = urlparse(self.request.path).path.strip("/").split("/")
+        # Path should be /aqua/models/ocid1.iad.ahdxxx/tokenizer
+        # path_list=['aqua','models','<model-ocid>','tokenizer']
+        if (
+            len(path_list) == 4
+            and is_valid_ocid(path_list[2])
+            and path_list[3] == "tokenizer"
+        ):
+            return self.finish(AquaModelApp().get_hf_tokenizer_config(model_id))
+
+        raise HTTPError(400, f"The request {self.request.path} is invalid.")
+
+
 __handlers__ = [
     ("model/?([^/]*)", AquaModelHandler),
     ("model/?([^/]*)/license", AquaModelLicenseHandler),
+    ("model/?([^/]*)/tokenizer", AquaModelTokenizerConfigHandler),
     ("model/hf/search/?([^/]*)", AquaHuggingFaceHandler),
 ]

ads/aqua/model/model.py

@@ -15,6 +15,7 @@
 from ads.aqua import ODSC_MODEL_COMPARTMENT_OCID, logger
 from ads.aqua.app import AquaApp
 from ads.aqua.common.enums import (
+    ConfigFolder,
     CustomInferenceContainerTypeFamily,
     FineTuningContainerTypeFamily,
     InferenceContainerTypeFamily,
@@ -44,6 +45,7 @@
     AQUA_MODEL_ARTIFACT_CONFIG_MODEL_NAME,
     AQUA_MODEL_ARTIFACT_CONFIG_MODEL_TYPE,
     AQUA_MODEL_ARTIFACT_FILE,
+    AQUA_MODEL_TOKENIZER_CONFIG,
     AQUA_MODEL_TYPE_CUSTOM,
     HF_METADATA_FOLDER,
     LICENSE_TXT,
@@ -568,6 +570,26 @@ def _build_ft_metrics(
             training_final,
         ]
 
+    def get_hf_tokenizer_config(self, model_id):
+        """Gets the default chat template for the given Aqua model.
+
+        Parameters
+        ----------
+        model_id: str
+            The OCID of the Aqua model.
+
+        Returns
+        -------
+        str:
+            Chat template string.
+        """
+        config = self.get_config(
+            model_id, AQUA_MODEL_TOKENIZER_CONFIG, ConfigFolder.ARTIFACT
+        )
+        if not config:
+            logger.debug(f"Tokenizer config for model: {model_id} is not available.")
+        return config
+
     @staticmethod
     def to_aqua_model(
         model: Union[

ads/aqua/modeldeployment/entities.py

@@ -41,6 +41,7 @@ class AquaDeployment(DataClassSerializable):
     id: str = None
     display_name: str = None
     aqua_service_model: bool = None
+    model_id: str = None
     aqua_model_name: str = None
     state: str = None
     description: str = None
@@ -97,7 +98,7 @@ def from_oci_model_deployment(
                 else None
             ),
         )
-
+        model_id = oci_model_deployment._model_deployment_configuration_details.model_configuration_details.model_id
         tags = {}
         tags.update(oci_model_deployment.freeform_tags or UNKNOWN_DICT)
         tags.update(oci_model_deployment.defined_tags or UNKNOWN_DICT)
@@ -110,6 +111,7 @@ def from_oci_model_deployment(
 
         return AquaDeployment(
             id=oci_model_deployment.id,
+            model_id=model_id,
             display_name=oci_model_deployment.display_name,
             aqua_service_model=aqua_service_model_tag is not None,
             aqua_model_name=aqua_model_name,

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 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 @@ def __init__(
 
     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

@@ -1778,7 +1778,7 @@ def _update_from_oci_dsc_model(
             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/model_metadata.py

@@ -1509,7 +1509,10 @@ def _from_oci_metadata(cls, metadata_list):
         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 @@ def from_dict(cls, data: Dict) -> "ModelTaxonomyMetadata":
         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
 
 

ads/model/service/oci_datascience_model.py

@@ -1,35 +1,32 @@
 #!/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/
 
 import logging
-import time
 from functools import wraps
 from io import BytesIO
 from typing import Callable, Dict, List, Optional
 
 import oci.data_science
-from ads.common import utils
-from ads.common.object_storage_details import ObjectStorageDetails
-from ads.common.oci_datascience import OCIDataScienceMixin
-from ads.common.oci_mixin import OCIWorkRequestMixin
-from ads.common.oci_resource import SEARCH_TYPE, OCIResource
-from ads.common.utils import extract_region
-from ads.common.work_request import DataScienceWorkRequest
-from ads.model.deployment import ModelDeployment
 from oci.data_science.models import (
     ArtifactExportDetailsObjectStorage,
     ArtifactImportDetailsObjectStorage,
     CreateModelDetails,
     ExportModelArtifactDetails,
     ImportModelArtifactDetails,
     UpdateModelDetails,
-    WorkRequest,
 )
 from oci.exceptions import ServiceError
 
+from ads.common.object_storage_details import ObjectStorageDetails
+from ads.common.oci_datascience import OCIDataScienceMixin
+from ads.common.oci_mixin import OCIWorkRequestMixin
+from ads.common.oci_resource import SEARCH_TYPE, OCIResource
+from ads.common.utils import extract_region
+from ads.common.work_request import DataScienceWorkRequest
+from ads.model.deployment import ModelDeployment
+
 logger = logging.getLogger(__name__)
 
 _REQUEST_INTERVAL_IN_SEC = 3
@@ -282,7 +279,7 @@ def get_artifact_info(self) -> Dict:
         msg="Model needs to be restored before the archived artifact content can be accessed."
     )
     def restore_archived_model_artifact(
-            self, restore_model_for_hours_specified: Optional[int] = None
+        self, restore_model_for_hours_specified: Optional[int] = None
     ) -> None:
         """Restores the archived model artifact.
 
@@ -304,7 +301,8 @@ def restore_archived_model_artifact(
         """
         return self.client.restore_archived_model_artifact(
             model_id=self.id,
-            restore_model_for_hours_specified=restore_model_for_hours_specified).headers["opc-work-request-id"]
+            restore_model_for_hours_specified=restore_model_for_hours_specified,
+        ).headers["opc-work-request-id"]
 
     @check_for_model_id(
         msg="Model needs to be saved to the Model Catalog before the artifact content can be read."
@@ -581,7 +579,7 @@ def from_id(cls, ocid: str) -> "OCIDataScienceModel":
             raise ValueError("Model OCID not provided.")
         return super().from_ocid(ocid)
 
-    def is_model_by_reference(self):
+    def _is_model_by_reference(self):
         """Checks if model is created by reference
         Returns
         -------