oracle-ads 2.12.2__py3-none-any.whl → 2.12.3__py3-none-any.whl

ads/aqua/modeldeployment/deployment.py

@@ -3,6 +3,7 @@
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
 
 import logging
+import shlex
 from typing import Dict, List, Optional, Union
 
 from ads.aqua.app import AquaApp, logger
@@ -23,7 +24,7 @@ from ads.aqua.common.utils import (
     get_params_list,
     get_resource_name,
     get_restricted_params_by_container,
-    load_config,
+    validate_cmd_var,
 )
 from ads.aqua.constants import (
     AQUA_MODEL_ARTIFACT_FILE,
@@ -44,11 +45,10 @@ from ads.aqua.ui import ModelFormat
 from ads.common.object_storage_details import ObjectStorageDetails
 from ads.common.utils import get_log_links
 from ads.config import (
-    AQUA_CONFIG_FOLDER,
+    AQUA_DEPLOYMENT_CONTAINER_CMD_VAR_METADATA_NAME,
     AQUA_DEPLOYMENT_CONTAINER_METADATA_NAME,
-    AQUA_DEPLOYMENT_CONTAINER_OVERRIDE_FLAG_METADATA_NAME,
+    AQUA_DEPLOYMENT_CONTAINER_URI_METADATA_NAME,
     AQUA_MODEL_DEPLOYMENT_CONFIG,
-    AQUA_MODEL_DEPLOYMENT_CONFIG_DEFAULTS,
     COMPARTMENT_OCID,
 )
 from ads.model.datascience_model import DataScienceModel
@@ -87,26 +87,29 @@ class AquaDeploymentApp(AquaApp):
 
     @telemetry(entry_point="plugin=deployment&action=create", name="aqua")
     def create(
-            self,
-            model_id: str,
-            instance_shape: str,
-            display_name: str,
-            instance_count: int = None,
-            log_group_id: str = None,
-            access_log_id: str = None,
-            predict_log_id: str = None,
-            compartment_id: str = None,
-            project_id: str = None,
-            description: str = None,
-            bandwidth_mbps: int = None,
-            web_concurrency: int = None,
-            server_port: int = None,
-            health_check_port: int = None,
-            env_var: Dict = None,
-            container_family: str = None,
-            memory_in_gbs: Optional[float] = None,
-            ocpus: Optional[float] = None,
-            model_file: Optional[str] = None,
+        self,
+        model_id: str,
+        instance_shape: str,
+        display_name: str,
+        instance_count: int = None,
+        log_group_id: str = None,
+        access_log_id: str = None,
+        predict_log_id: str = None,
+        compartment_id: str = None,
+        project_id: str = None,
+        description: str = None,
+        bandwidth_mbps: int = None,
+        web_concurrency: int = None,
+        server_port: int = None,
+        health_check_port: int = None,
+        env_var: Dict = None,
+        container_family: str = None,
+        memory_in_gbs: Optional[float] = None,
+        ocpus: Optional[float] = None,
+        model_file: Optional[str] = None,
+        private_endpoint_id: Optional[str] = None,
+        container_image_uri: Optional[None] = None,
+        cmd_var: List[str] = None,
     ) -> "AquaDeployment":
         """
         Creates a new Aqua deployment
@@ -153,6 +156,13 @@ class AquaDeploymentApp(AquaApp):
             The ocpu count for the shape selected.
         model_file: str
             The file used for model deployment.
+        private_endpoint_id: str
+            The private endpoint id of model deployment.
+        container_image_uri: str
+            The image of model deployment container runtime, ignored for service managed containers.
+            Required parameter for BYOC based deployments if this parameter was not set during model registration.
+        cmd_var: List[str]
+            The cmd of model deployment container runtime.
         Returns
         -------
         AquaDeployment
@@ -175,6 +185,7 @@ class AquaDeploymentApp(AquaApp):
                 tags[tag] = aqua_model.freeform_tags[tag]
 
         tags.update({Tags.AQUA_MODEL_NAME_TAG: aqua_model.display_name})
+        tags.update({Tags.TASK: aqua_model.freeform_tags.get(Tags.TASK, None)})
 
         # Set up info to get deployment config
         config_source_id = model_id
@@ -196,9 +207,11 @@ class AquaDeploymentApp(AquaApp):
                     f"from custom metadata for the model {config_source_id}"
                 ) from err
 
-        # set up env vars
+        # set up env and cmd var
         if not env_var:
             env_var = {}
+        if not cmd_var:
+            cmd_var = []
 
         try:
             model_path_prefix = aqua_model.custom_metadata_list.get(
@@ -231,24 +244,56 @@ class AquaDeploymentApp(AquaApp):
             env_var.update({"FT_MODEL": f"{fine_tune_output_path}"})
 
         container_type_key = self._get_container_type_key(
-            model=aqua_model,
-            container_family=container_family
+            model=aqua_model, container_family=container_family
         )
 
-        # fetch image name from config
-        container_image = get_container_image(container_type=container_type_key)
-
+        container_image_uri = container_image_uri or get_container_image(
+            container_type=container_type_key
+        )
+        if not container_image_uri:
+            try:
+                container_image_uri = aqua_model.custom_metadata_list.get(
+                    AQUA_DEPLOYMENT_CONTAINER_URI_METADATA_NAME
+                ).value
+            except ValueError as err:
+                raise AquaValueError(
+                    f"{AQUA_DEPLOYMENT_CONTAINER_URI_METADATA_NAME} key is not available in the custom metadata "
+                    f"field. Either re-register the model with custom container URI, or set container_image_uri "
+                    f"parameter when creating this deployment."
+                ) from err
         logging.info(
-            f"Aqua Image used for deploying {aqua_model.id} : {container_image}"
+            f"Aqua Image used for deploying {aqua_model.id} : {container_image_uri}"
         )
 
+        try:
+            cmd_var_string = aqua_model.custom_metadata_list.get(
+                AQUA_DEPLOYMENT_CONTAINER_CMD_VAR_METADATA_NAME
+            ).value
+            default_cmd_var = shlex.split(cmd_var_string)
+            if default_cmd_var:
+                cmd_var = validate_cmd_var(default_cmd_var, cmd_var)
+            logging.info(f"CMD used for deploying {aqua_model.id} :{cmd_var}")
+        except ValueError:
+            logging.debug(
+                f"CMD will be ignored for this deployment as {AQUA_DEPLOYMENT_CONTAINER_CMD_VAR_METADATA_NAME} "
+                f"key is not available in the custom metadata field for this model."
+            )
+        except Exception as e:
+            logging.error(
+                f"There was an issue processing CMD arguments. Error: {str(e)}"
+            )
+
         model_formats_str = aqua_model.freeform_tags.get(
             Tags.MODEL_FORMAT, ModelFormat.SAFETENSORS.value
         ).upper()
         model_format = model_formats_str.split(",")
 
         # Figure out a better way to handle this in future release
-        if ModelFormat.GGUF.value in model_format and container_type_key.lower() == InferenceContainerTypeFamily.AQUA_LLAMA_CPP_CONTAINER_FAMILY:
+        if (
+            ModelFormat.GGUF.value in model_format
+            and container_type_key.lower()
+            == InferenceContainerTypeFamily.AQUA_LLAMA_CPP_CONTAINER_FAMILY
+        ):
             if model_file is not None:
                 logger.info(
                     f"Overriding {model_file} as model_file for model {aqua_model.id}."
@@ -299,13 +344,13 @@ class AquaDeploymentApp(AquaApp):
         if user_params:
             # todo: remove this check in the future version, logic to be moved to container_index
             if (
-                    container_type_key.lower()
-                    == InferenceContainerTypeFamily.AQUA_LLAMA_CPP_CONTAINER_FAMILY
+                container_type_key.lower()
+                == InferenceContainerTypeFamily.AQUA_LLAMA_CPP_CONTAINER_FAMILY
             ):
                 # AQUA_LLAMA_CPP_CONTAINER_FAMILY container uses uvicorn that required model/server params
                 # to be set as env vars
                 raise AquaValueError(
-                    f"Currently, parameters cannot be overridden for the container: {container_image}. Please proceed "
+                    f"Currently, parameters cannot be overridden for the container: {container_image_uri}. Please proceed "
                     f"with deployment without parameter overrides."
                 )
 
@@ -342,6 +387,7 @@ class AquaDeploymentApp(AquaApp):
             .with_bandwidth_mbps(bandwidth_mbps)
             .with_replica(instance_count)
             .with_web_concurrency(web_concurrency)
+            .with_private_endpoint_id(private_endpoint_id)
             .with_access_log(
                 log_group_id=log_group_id,
                 log_id=access_log_id,
@@ -359,7 +405,7 @@ class AquaDeploymentApp(AquaApp):
         # configure model deployment runtime
         container_runtime = (
             ModelDeploymentContainerRuntime()
-            .with_image(container_image)
+            .with_image(container_image_uri)
             .with_server_port(server_port)
             .with_health_check_port(health_check_port)
             .with_env(env_var)
@@ -369,6 +415,8 @@ class AquaDeploymentApp(AquaApp):
             .with_overwrite_existing_artifact(True)
             .with_remove_existing_artifact(True)
         )
+        if cmd_var:
+            container_runtime.with_cmd(cmd_var)
 
         # configure model deployment and deploy model on container runtime
         deployment = (
@@ -422,9 +470,8 @@ class AquaDeploymentApp(AquaApp):
                     f"for model {model.id}. For unverified Aqua models, {AQUA_DEPLOYMENT_CONTAINER_METADATA_NAME} should be"
                     f"set and value can be one of {', '.join(InferenceContainerTypeFamily.values())}."
                 ) from err
-            
+
         return container_type_key
-        
 
     @telemetry(entry_point="plugin=deployment&action=list", name="aqua")
     def list(self, **kwargs) -> List["AquaDeployment"]:
@@ -453,8 +500,8 @@ class AquaDeploymentApp(AquaApp):
         for model_deployment in model_deployments:
             oci_aqua = (
                 (
-                        Tags.AQUA_TAG in model_deployment.freeform_tags
-                        or Tags.AQUA_TAG.lower() in model_deployment.freeform_tags
+                    Tags.AQUA_TAG in model_deployment.freeform_tags
+                    or Tags.AQUA_TAG.lower() in model_deployment.freeform_tags
                 )
                 if model_deployment.freeform_tags
                 else False
@@ -508,8 +555,8 @@ class AquaDeploymentApp(AquaApp):
 
         oci_aqua = (
             (
-                    Tags.AQUA_TAG in model_deployment.freeform_tags
-                    or Tags.AQUA_TAG.lower() in model_deployment.freeform_tags
+                Tags.AQUA_TAG in model_deployment.freeform_tags
+                or Tags.AQUA_TAG.lower() in model_deployment.freeform_tags
             )
             if model_deployment.freeform_tags
             else False
@@ -526,8 +573,8 @@ class AquaDeploymentApp(AquaApp):
         log_group_name = ""
 
         logs = (
-                model_deployment.category_log_details.access
-                or model_deployment.category_log_details.predict
+            model_deployment.category_log_details.access
+            or model_deployment.category_log_details.predict
         )
         if logs:
             log_id = logs.log_id
@@ -574,17 +621,15 @@ class AquaDeploymentApp(AquaApp):
         """
         config = self.get_config(model_id, AQUA_MODEL_DEPLOYMENT_CONFIG)
         if not config:
-            logger.info(f"Fetching default deployment config for model: {model_id}")
-            config = load_config(
-                AQUA_CONFIG_FOLDER,
-                config_file_name=AQUA_MODEL_DEPLOYMENT_CONFIG_DEFAULTS,
+            logger.debug(
+                f"Deployment config for custom model: {model_id} is not available."
             )
         return config
 
     def get_deployment_default_params(
-            self,
-            model_id: str,
-            instance_shape: str,
+        self,
+        model_id: str,
+        instance_shape: str,
     ) -> List[str]:
         """Gets the default params set in the deployment configs for the given model and instance shape.
 
@@ -616,8 +661,8 @@ class AquaDeploymentApp(AquaApp):
             )
 
         if (
-                container_type_key
-                and container_type_key in InferenceContainerTypeFamily.values()
+            container_type_key
+            and container_type_key in InferenceContainerTypeFamily.values()
         ):
             deployment_config = self.get_deployment_config(model_id)
             config_params = (
@@ -640,10 +685,10 @@ class AquaDeploymentApp(AquaApp):
         return default_params
 
     def validate_deployment_params(
-            self,
-            model_id: str,
-            params: List[str] = None,
-            container_family: str = None,
+        self,
+        model_id: str,
+        params: List[str] = None,
+        container_family: str = None,
     ) -> Dict:
         """Validate if the deployment parameters passed by the user can be overridden. Parameter values are not
         validated, only param keys are validated.
@@ -666,8 +711,7 @@ class AquaDeploymentApp(AquaApp):
         if params:
             model = DataScienceModel.from_id(model_id)
             container_type_key = self._get_container_type_key(
-                model=model,
-                container_family=container_family
+                model=model, container_family=container_family
             )
 
             container_config = get_container_config()
@@ -689,9 +733,9 @@ class AquaDeploymentApp(AquaApp):
 
     @staticmethod
     def _find_restricted_params(
-            default_params: Union[str, List[str]],
-            user_params: Union[str, List[str]],
-            container_family: str,
+        default_params: Union[str, List[str]],
+        user_params: Union[str, List[str]],
+        container_family: str,
     ) -> List[str]:
         """Returns a list of restricted params that user chooses to override when creating an Aqua deployment.
         The default parameters coming from the container index json file cannot be overridden.

ads/aqua/modeldeployment/entities.py

@@ -3,7 +3,7 @@
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
 
 from dataclasses import dataclass, field
-from typing import Union
+from typing import List, Optional, Union
 
 from oci.data_science.models import (
     ModelDeployment,
@@ -47,11 +47,13 @@ class AquaDeployment(DataClassSerializable):
     created_on: str = None
     created_by: str = None
     endpoint: str = None
+    private_endpoint_id: str = None
     console_link: str = None
     lifecycle_details: str = None
-    shape_info: field(default_factory=ShapeInfo) = None
+    shape_info: Optional[ShapeInfo] = None
     tags: dict = None
     environment_variables: dict = None
+    cmd: List[str] = None
 
     @classmethod
     def from_oci_model_deployment(
@@ -80,6 +82,7 @@ class AquaDeployment(DataClassSerializable):
         )
         instance_count = oci_model_deployment.model_deployment_configuration_details.model_configuration_details.scaling_policy.instance_count
         environment_variables = oci_model_deployment.model_deployment_configuration_details.environment_configuration_details.environment_variables
+        cmd = oci_model_deployment.model_deployment_configuration_details.environment_configuration_details.cmd
         shape_info = ShapeInfo(
             instance_shape=instance_configuration.instance_shape_name,
             instance_count=instance_count,
@@ -98,6 +101,9 @@ class AquaDeployment(DataClassSerializable):
         freeform_tags = oci_model_deployment.freeform_tags or UNKNOWN_DICT
         aqua_service_model_tag = freeform_tags.get(Tags.AQUA_SERVICE_MODEL_TAG, None)
         aqua_model_name = freeform_tags.get(Tags.AQUA_MODEL_NAME_TAG, UNKNOWN)
+        private_endpoint_id = getattr(
+            instance_configuration, "private_endpoint_id", UNKNOWN
+        )
 
         return AquaDeployment(
             id=oci_model_deployment.id,
@@ -113,6 +119,7 @@ class AquaDeployment(DataClassSerializable):
             created_on=str(oci_model_deployment.time_created),
             created_by=oci_model_deployment.created_by,
             endpoint=oci_model_deployment.model_deployment_url,
+            private_endpoint_id=private_endpoint_id,
             console_link=get_console_link(
                 resource="model-deployments",
                 ocid=oci_model_deployment.id,
@@ -120,6 +127,7 @@ class AquaDeployment(DataClassSerializable):
             ),
             tags=freeform_tags,
             environment_variables=environment_variables,
+            cmd=cmd,
         )
 
 

ads/aqua/ui.py

@@ -17,14 +17,13 @@ from ads.aqua.app import AquaApp
 from ads.aqua.common.entities import ContainerSpec
 from ads.aqua.common.enums import Tags
 from ads.aqua.common.errors import AquaResourceAccessError, AquaValueError
-from ads.aqua.common.utils import get_container_config, load_config, sanitize_response
+from ads.aqua.common.utils import get_container_config, sanitize_response
+from ads.aqua.constants import PRIVATE_ENDPOINT_TYPE
 from ads.common import oci_client as oc
 from ads.common.auth import default_signer
 from ads.common.object_storage_details import ObjectStorageDetails
 from ads.common.serializer import DataClassSerializable
 from ads.config import (
-    AQUA_CONFIG_FOLDER,
-    AQUA_RESOURCE_LIMIT_NAMES_CONFIG,
     COMPARTMENT_OCID,
     DATA_SCIENCE_SERVICE_NAME,
     TENANCY_OCID,
@@ -549,6 +548,32 @@ class AquaUIApp(AquaApp):
 
         return sanitize_response(oci_client=vcn_client, response=res)
 
+    @telemetry(entry_point="plugin=ui&action=list_private_endpoints", name="aqua")
+    def list_private_endpoints(self, **kwargs) -> list:
+        """Lists the private endpoints in the specified compartment.
+        Data seicne private endpoints have two types: `NOTEBOOK_SESSION` and `MODEL_DEPLOYMENT`.
+        This api will by default list `MODEL_DEPLOYMENT` type as needed by AQUA model deployment.
+
+        Parameters
+        ----------
+        **kwargs
+            Addtional arguments, such as `compartment_id`,
+            for `list_data_science_private_endpoints <https://docs.oracle.com/en-us/iaas/tools/python/latest/api/data_science/client/oci.data_science.DataScienceClient.html#oci.data_science.DataScienceClient.list_data_science_private_endpoints>`_
+
+        Returns
+        -------
+            json representation of `oci.data_science.models.DataSciencePrivateEndpointSummary`.
+        """
+        compartment_id = kwargs.pop("compartment_id", COMPARTMENT_OCID)
+        resource_type = kwargs.pop("resource_type", PRIVATE_ENDPOINT_TYPE)
+        logger.info(f"Loading private endpoints from compartment: {compartment_id}")
+
+        res = self.ds_client.list_data_science_private_endpoints(
+            compartment_id=compartment_id, data_science_resource_type=resource_type
+        ).data
+
+        return sanitize_response(oci_client=self.ds_client, response=res)
+
     @telemetry(entry_point="plugin=ui&action=get_shape_availability", name="aqua")
     def get_shape_availability(self, **kwargs):
         """
@@ -572,25 +597,13 @@ class AquaUIApp(AquaApp):
         """
         compartment_id = kwargs.pop("compartment_id", COMPARTMENT_OCID)
         instance_shape = kwargs.pop("instance_shape", None)
+        limit_name = kwargs.pop("limit_name", None)
 
         if not instance_shape:
             raise AquaValueError("instance_shape argument is required.")
 
         limits_client = oc.OCIClientFactory(**default_signer()).limits
 
-        artifact_path = AQUA_CONFIG_FOLDER
-        config = load_config(
-            artifact_path,
-            config_file_name=AQUA_RESOURCE_LIMIT_NAMES_CONFIG,
-        )
-
-        if instance_shape not in config:
-            logger.error(
-                f"{instance_shape} does not have mapping details in {AQUA_RESOURCE_LIMIT_NAMES_CONFIG}"
-            )
-            return {}
-
-        limit_name = config[instance_shape]
         try:
             res = limits_client.get_resource_availability(
                 DATA_SCIENCE_SERVICE_NAME, limit_name, compartment_id, **kwargs

ads/config.py

@@ -1,5 +1,4 @@
 #!/usr/bin/env python
-# -*- coding: utf-8; -*-
 
 # Copyright (c) 2020, 2024 Oracle and/or its affiliates.
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
@@ -57,15 +56,11 @@ AQUA_MODEL_FINETUNING_CONFIG = os.environ.get(
 AQUA_CONTAINER_INDEX_CONFIG = os.environ.get(
     "AQUA_CONTAINER_INDEX_CONFIG", "container_index.json"
 )
-AQUA_MODEL_DEPLOYMENT_CONFIG_DEFAULTS = os.environ.get(
-    "AQUA_MODEL_DEPLOYMENT_CONFIG_DEFAULTS", "deployment_config_defaults.json"
-)
-AQUA_RESOURCE_LIMIT_NAMES_CONFIG = os.environ.get(
-    "AQUA_RESOURCE_LIMIT_NAMES_CONFIG", "resource_limit_names.json"
-)
 AQUA_DEPLOYMENT_CONTAINER_METADATA_NAME = "deployment-container"
 AQUA_FINETUNING_CONTAINER_METADATA_NAME = "finetune-container"
 AQUA_EVALUATION_CONTAINER_METADATA_NAME = "evaluation-container"
+AQUA_DEPLOYMENT_CONTAINER_CMD_VAR_METADATA_NAME = "container-cmd-var"
+AQUA_DEPLOYMENT_CONTAINER_URI_METADATA_NAME = "deployment-container-uri"
 AQUA_DEPLOYMENT_CONTAINER_OVERRIDE_FLAG_METADATA_NAME = "deployment-container-custom"
 AQUA_FINETUNING_CONTAINER_OVERRIDE_FLAG_METADATA_NAME = "finetune-container-custom"
 AQUA_MODEL_DEPLOYMENT_FOLDER = "/opt/ds/model/deployed_model/"
@@ -212,7 +207,7 @@ def open(
         frame.f_globals.pop("config", None)
 
         # Restores original globals
-        for key in defined_globals.keys():
+        for key in defined_globals:
             frame.f_globals[key] = defined_globals[key]
 
         # Saving config if it necessary

ads/llm/deploy.py

@@ -19,13 +19,19 @@ from ads.llm.serialize import dump, load_from_yaml
 
 
 class ChainDeployment(GenericModel):
+    """Represents a model deployment with LangChain.
+    """
     def __init__(self, chain, **kwargs):
         self.chain = chain
+        if "model_input_serializer" not in kwargs:
+            kwargs["model_input_serializer"] = self.model_input_serializer_type.JSON
         super().__init__(**kwargs)
 
     def prepare(self, **kwargs) -> GenericModel:
         """Prepares the model artifact."""
         chain_yaml_uri = os.path.join(self.artifact_dir, "chain.yaml")
+        if not os.path.exists(self.artifact_dir):
+            os.makedirs(self.artifact_dir)
         with open(chain_yaml_uri, "w", encoding="utf-8") as f:
             f.write(yaml.safe_dump(dump(self.chain)))
 

ads/llm/guardrails/base.py

@@ -156,7 +156,6 @@ class Guardrail(BaseTool):
 
     class Config:
         arbitrary_types_allowed = True
-        underscore_attrs_are_private = True
 
     name: str = ""
     description: str = "Guardrail"