Added support to create multi model deployment. (#1072)

Author: lu-ohai

ads/aqua/app.py

@@ -6,9 +6,11 @@
 import os
 import traceback
 from dataclasses import fields
+from datetime import datetime, timedelta
 from typing import Dict, Optional, Union
 
 import oci
+from cachetools import TTLCache, cached
 from oci.data_science.models import UpdateModelDetails, UpdateModelProvenanceDetails
 
 from ads import set_auth
@@ -268,6 +270,7 @@ def if_artifact_exist(self, model_id: str, **kwargs) -> bool:
                 logger.info(f"Artifact not found in model {model_id}.")
                 return False
 
+    @cached(cache=TTLCache(maxsize=1, ttl=timedelta(minutes=1), timer=datetime.now))
     def get_config(
         self,
         model_id: str,

ads/aqua/common/entities.py

@@ -4,6 +4,8 @@
 
 from typing import Optional
 
+from pydantic import Field
+
 from ads.aqua.config.utils.serializer import Serializable
 
 
@@ -42,16 +44,22 @@ class AquaMultiModelRef(Serializable):
     ----------
     model_id : str
         The unique identifier of the model.
+    model_name : Optional[str]
+        The name of the model.
     gpu_count : Optional[int]
         Number of GPUs required for deployment.
     env_var : Optional[Dict[str, Any]]
         Optional environment variables to override during deployment.
     """
 
-    model_id: str
-    model_name: Optional[str] = None
-    gpu_count: Optional[int] = None
-    env_var: Optional[dict] = None
+    model_id: str = Field(..., description="The model OCID to deploy.")
+    model_name: Optional[str] = Field(None, description="The name of model.")
+    gpu_count: Optional[int] = Field(
+        None, description="The gpu count allocation for the model."
+    )
+    env_var: Optional[dict] = Field(
+        default_factory=dict, description="The environment variables of the model."
+    )
 
     class Config:
         extra = "ignore"

ads/aqua/common/enums.py

@@ -31,6 +31,7 @@ class Tags(ExtendedEnum):
     AQUA_TAG = "OCI_AQUA"
     AQUA_SERVICE_MODEL_TAG = "aqua_service_model"
     AQUA_FINE_TUNED_MODEL_TAG = "aqua_fine_tuned_model"
+    AQUA_MODEL_ID_TAG = "aqua_model_id"
     AQUA_MODEL_NAME_TAG = "aqua_model_name"
     AQUA_EVALUATION = "aqua_evaluation"
     AQUA_FINE_TUNING = "aqua_finetuning"

ads/aqua/constants.py

@@ -1,5 +1,5 @@
 #!/usr/bin/env python
-# Copyright (c) 2024 Oracle and/or its affiliates.
+# Copyright (c) 2024, 2025 Oracle and/or its affiliates.
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
 """This module defines constants used in ads.aqua module."""
 
@@ -30,14 +30,17 @@
 READY_TO_FINE_TUNE_STATUS = "TRUE"
 PRIVATE_ENDPOINT_TYPE = "MODEL_DEPLOYMENT"
 AQUA_GA_LIST = ["id19sfcrra6z"]
+AQUA_MULTI_MODEL_CONFIG = "MULTI_MODEL_CONFIG"
 AQUA_MODEL_TYPE_SERVICE = "service"
 AQUA_MODEL_TYPE_CUSTOM = "custom"
+AQUA_MODEL_TYPE_MULTI = "multi_model"
 AQUA_MODEL_ARTIFACT_CONFIG = "config.json"
 AQUA_MODEL_ARTIFACT_CONFIG_MODEL_NAME = "_name_or_path"
 AQUA_MODEL_ARTIFACT_CONFIG_MODEL_TYPE = "model_type"
 AQUA_MODEL_ARTIFACT_FILE = "model_file"
 HF_METADATA_FOLDER = ".cache/"
 HF_LOGIN_DEFAULT_TIMEOUT = 2
+MODEL_NAME_DELIMITER = ";"
 
 TRAINING_METRICS_FINAL = "training_metrics_final"
 VALIDATION_METRICS_FINAL = "validation_metrics_final"

ads/aqua/model/model.py

@@ -65,6 +65,7 @@
     FineTuningCustomMetadata,
     FineTuningMetricCategories,
     ModelCustomMetadataFields,
+    ModelTask,
     ModelType,
 )
 from ads.aqua.model.entities import (
@@ -254,12 +255,31 @@ def create_multi(
         artifact_list = []
         display_name_list = []
         model_custom_metadata = ModelCustomMetadata()
-        default_deployment_container = None
+        # TODO: update it when more deployment containers are supported
+        default_deployment_container = (
+            InferenceContainerTypeFamily.AQUA_VLLM_CONTAINER_FAMILY
+        )
 
         # Process each model
         for idx, model in enumerate(models):
             source_model = DataScienceModel.from_id(model.model_id)
             display_name = source_model.display_name
+
+            if not source_model.freeform_tags.get(Tags.AQUA_SERVICE_MODEL_TAG, UNKNOWN):
+                raise AquaValueError(
+                    f"Invalid selected model {display_name}. "
+                    "Currently only service models are supported for multi model deployment."
+                )
+
+            if (
+                source_model.freeform_tags.get(Tags.TASK, UNKNOWN)
+                != ModelTask.TEXT_GENERATION
+            ):
+                raise AquaValueError(
+                    f"Invalid or missing {Tags.TASK} tag for selected model {display_name}. "
+                    f"Currently only {ModelTask.TEXT_GENERATION} models are support for multi model deployment."
+                )
+
             display_name_list.append(display_name)
 
             # Retrieve model artifact
@@ -280,12 +300,10 @@ def create_multi(
                 ),
             ).value
 
-            if idx == 0:
-                default_deployment_container = deployment_container
-            elif deployment_container != default_deployment_container:
+            if default_deployment_container != deployment_container:
                 raise AquaValueError(
-                    "Deployment container mismatch detected. "
-                    "All selected models must use the same deployment container."
+                    f"Unsupported deployment container '{deployment_container}' for model '{source_model.id}'. "
+                    f"Only '{InferenceContainerTypeFamily.AQUA_VLLM_CONTAINER_FAMILY}' is supported for multi-model deployments."
                 )
 
             # Add model-specific metadata