Exclude the streaming environment variable when creating model deployments. (#1180)

mrDzurb · web-flow · commit dfe2197aeaf2 · 2025-06-05T13:00:57.000-04:00
diff --git a/ads/aqua/config/container_config.py b/ads/aqua/config/container_config.py
@@ -196,7 +196,6 @@ def from_service_config(
                 config_keys = {
                     "MODEL_DEPLOY_PREDICT_ENDPOINT": UNKNOWN,
                     "MODEL_DEPLOY_HEALTH_ENDPOINT": UNKNOWN,
-                    "MODEL_DEPLOY_ENABLE_STREAMING": UNKNOWN,
                     "PORT": UNKNOWN,
                     "HEALTH_CHECK_PORT": UNKNOWN,
                     "VLLM_USE_V1": UNKNOWN,
diff --git a/tests/unitary/with_extras/aqua/test_data/deployment/aqua_create_deployment.yaml b/tests/unitary/with_extras/aqua/test_data/deployment/aqua_create_deployment.yaml
@@ -24,7 +24,6 @@ spec:
     spec:
       env:
         BASE_MODEL: service_models/model-name/artifact
-        MODEL_DEPLOY_ENABLE_STREAMING: 'true'
         MODEL_DEPLOY_PREDICT_ENDPOINT: /v1/completions
         PARAMS: --served-model-name odsc-llm --seed 42 --trust-remote-code --max-model-len 4096
       healthCheckPort: 8080
diff --git a/tests/unitary/with_extras/aqua/test_data/deployment/aqua_create_gguf_deployment.yaml b/tests/unitary/with_extras/aqua/test_data/deployment/aqua_create_gguf_deployment.yaml
@@ -28,7 +28,6 @@ spec:
       env:
         BASE_MODEL: service_models/model-name/artifact
         BASE_MODEL_FILE: model-name.gguf
-        MODEL_DEPLOY_ENABLE_STREAMING: 'true'
         MODEL_DEPLOY_PREDICT_ENDPOINT: /v1/completions
         MODEL_DEPLOY_HEALTH_ENDPOINT: /v1/models
       healthCheckPort: 8080
diff --git a/tests/unitary/with_extras/aqua/test_data/ui/container_index.json b/tests/unitary/with_extras/aqua/test_data/ui/container_index.json
@@ -9,9 +9,6 @@
         {
           "MODEL_DEPLOY_HEALTH_ENDPOINT": "/v1/models"
         },
-        {
-          "MODEL_DEPLOY_ENABLE_STREAMING": "true"
-        },
         {
           "PORT": "8080"
         },
@@ -33,9 +30,6 @@
         {
           "MODEL_DEPLOY_PREDICT_ENDPOINT": "/v1/completions"
         },
-        {
-          "MODEL_DEPLOY_ENABLE_STREAMING": "true"
-        },
         {
           "PORT": "8080"
         },
@@ -59,9 +53,6 @@
         {
           "MODEL_DEPLOY_PREDICT_ENDPOINT": "/v1/completions"
         },
-        {
-          "MODEL_DEPLOY_ENABLE_STREAMING": "true"
-        },
         {
           "PORT": "8080"
         },
diff --git a/tests/unitary/with_extras/aqua/test_deployment.py b/tests/unitary/with_extras/aqua/test_deployment.py
@@ -118,7 +118,6 @@ class TestDataset:
                         "health_check_port": 8080,
                         "additional_configurations": {
                             "HEALTH_CHECK_PORT": "8080",
-                            "MODEL_DEPLOY_ENABLE_STREAMING": "true",
                             "MODEL_DEPLOY_PREDICT_ENDPOINT": "/v1/completions",
                             "PORT": "8080",
                             "modelFormats": "SAFETENSORS",
@@ -150,7 +149,6 @@ class TestDataset:
                     {
                         "MODEL_DEPLOY_PREDICT_ENDPOINT": "/v1/completions",
                         "MODEL_DEPLOY_HEALTH_ENDPOINT": "",
-                        "MODEL_DEPLOY_ENABLE_STREAMING": "true",
                         "PORT": "8080",
                         "HEALTH_CHECK_PORT": "8080",
                     }
@@ -202,7 +200,6 @@ class TestDataset:
                             "environment_configuration_type": "OCIR_CONTAINER",
                             "environment_variables": {
                                 "BASE_MODEL": "service_models/model-name/artifact",
-                                "MODEL_DEPLOY_ENABLE_STREAMING": "true",
                                 "MODEL_DEPLOY_PREDICT_ENDPOINT": "/v1/completions",
                                 "PARAMS": "--served-model-name odsc-llm --seed 42",
                             },
@@ -326,7 +323,6 @@ class TestDataset:
                             "environment_variables": {
                                 "BASE_MODEL": "service_models/model-name/artifact",
                                 "BASE_MODEL_FILE": "model-name.gguf",
-                                "MODEL_DEPLOY_ENABLE_STREAMING": "true",
                                 "MODEL_DEPLOY_PREDICT_ENDPOINT": "/v1/completions",
                                 "MODEL_DEPLOY_HEALTH_ENDPOINT": "/v1/models",
                             },
@@ -451,7 +447,6 @@ class TestDataset:
         "model_id": "ocid1.datasciencemodel.oc1.<region>.<OCID>",
         "environment_variables": {
             "BASE_MODEL": "service_models/model-name/artifact",
-            "MODEL_DEPLOY_ENABLE_STREAMING": "true",
             "MODEL_DEPLOY_PREDICT_ENDPOINT": "/v1/completions",
             "PARAMS": "--served-model-name odsc-llm --seed 42",
         },
@@ -531,7 +526,6 @@ class TestDataset:
     aqua_deployment_gguf_env_vars = {
         "BASE_MODEL": "service_models/model-name/artifact",
         "BASE_MODEL_FILE": "model-name.gguf",
-        "MODEL_DEPLOY_ENABLE_STREAMING": "true",
         "MODEL_DEPLOY_HEALTH_ENDPOINT": "/v1/models",
         "MODEL_DEPLOY_PREDICT_ENDPOINT": "/v1/completions",
     }
diff --git a/tests/unitary/with_extras/aqua/test_ui.py b/tests/unitary/with_extras/aqua/test_ui.py
@@ -15,18 +15,17 @@
 import pytest
 from oci.data_science.models import (
     ContainerSummary,
-    WorkloadConfigurationDetails,
-    ModelDeployWorkloadConfigurationDetails,
     JobRunWorkloadConfigurationDetails,
+    ModelDeployWorkloadConfigurationDetails,
+    WorkloadConfigurationDetails,
 )
-
-from ads.aqua.app import AquaApp
-from ads.aqua.extension.base_handler import AquaAPIhandler
 from parameterized import parameterized
 
 import ads.config
+from ads.aqua.app import AquaApp
 from ads.aqua.common.errors import AquaValueError
 from ads.aqua.common.utils import load_config
+from ads.aqua.extension.base_handler import AquaAPIhandler
 from ads.aqua.ui import AquaUIApp
 from tests.unitary.with_extras.aqua.utils import ServiceManagedContainers
 
@@ -607,7 +606,6 @@ def test_list_containers(self, mock_list_service_containers):
                         "env_vars": [
                             {"MODEL_DEPLOY_PREDICT_ENDPOINT": "/v1/completions"},
                             # {"MODEL_DEPLOY_HEALTH_ENDPOINT": ""},
-                            {"MODEL_DEPLOY_ENABLE_STREAMING": "true"},
                             {"PORT": "8080"},
                             {"HEALTH_CHECK_PORT": "8080"},
                         ],
diff --git a/tests/unitary/with_extras/aqua/utils.py b/tests/unitary/with_extras/aqua/utils.py
@@ -58,7 +58,6 @@ class ServiceManagedContainers:
                             "health_check_port": 8080,
                             "additional_configurations": {
                                 "HEALTH_CHECK_PORT": "8080",
-                                "MODEL_DEPLOY_ENABLE_STREAMING": "true",
                                 "MODEL_DEPLOY_PREDICT_ENDPOINT": "/v1/completions",
                                 "PORT": "8080",
                                 "modelFormats": "SAFETENSORS",
diff --git a/tests/unitary/with_extras/model/test_model_framework_lightgbm_model.py b/tests/unitary/with_extras/model/test_model_framework_lightgbm_model.py
@@ -1,6 +1,6 @@
 #!/usr/bin/env python
 
-# Copyright (c) 2021, 2023 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/
 
 """Unit tests for model frameworks. Includes tests for:
@@ -10,19 +10,22 @@
 import base64
 import os
 import shutil
+import sys
+import tempfile
 from io import BytesIO
 
 import joblib
 import lightgbm as lgb
+import mock
 import numpy as np
 import onnx
 import onnxruntime as rt
 import pandas as pd
-import pytest, mock, sys
-import tempfile
+import pytest
+from sklearn.datasets import load_iris, make_regression
+
 from ads.model.framework.lightgbm_model import LightGBMModel
 from ads.model.serde.model_serializer import LightGBMOnnxModelSerializer
-from sklearn.datasets import load_iris, make_regression
 
 tmp_model_dir = tempfile.mkdtemp()
 
diff --git a/tests/unitary/with_extras/model/test_model_framework_spark_pipeline_model.py b/tests/unitary/with_extras/model/test_model_framework_spark_pipeline_model.py
@@ -1,6 +1,6 @@
 #!/usr/bin/env python
 
-# Copyright (c) 2021, 2023 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/
 
 """Unit tests for model frameworks. Includes tests for:
@@ -10,16 +10,17 @@
 import os
 import shutil
 import tempfile
-import pytest
+
 import numpy as np
+import pytest
 from packaging import version
-from ads.model.framework.spark_model import SparkPipelineModel
+from pyspark.ml.classification import LogisticRegression
+from pyspark.ml.feature import HashingTF, Tokenizer
 from pyspark.ml.linalg import Vectors
 from pyspark.ml.pipeline import Pipeline, PipelineModel
 from pyspark.sql import SparkSession
-from pyspark.ml.classification import LogisticRegression
-from pyspark.ml.feature import HashingTF, Tokenizer
 
+from ads.model.framework.spark_model import SparkPipelineModel
 
 spark = SparkSession.builder.appName("Python Spark SQL basic example").getOrCreate()
 artifact_dir1 = tempfile.mkdtemp()

Original file line number	Diff line number	Diff line change
`@@ -9,9 +9,6 @@`
`9`	`9`	`{`
`10`	`10`	`"MODEL_DEPLOY_HEALTH_ENDPOINT": "/v1/models"`
`11`	`11`	`},`
`12`		`- {`
`13`		`- "MODEL_DEPLOY_ENABLE_STREAMING": "true"`
`14`		`- },`
`15`	`12`	`{`
`16`	`13`	`"PORT": "8080"`
`17`	`14`	`},`
`@@ -33,9 +30,6 @@`
`33`	`30`	`{`
`34`	`31`	`"MODEL_DEPLOY_PREDICT_ENDPOINT": "/v1/completions"`
`35`	`32`	`},`
`36`		`- {`
`37`		`- "MODEL_DEPLOY_ENABLE_STREAMING": "true"`
`38`		`- },`
`39`	`33`	`{`
`40`	`34`	`"PORT": "8080"`
`41`	`35`	`},`
`@@ -59,9 +53,6 @@`
`59`	`53`	`{`
`60`	`54`	`"MODEL_DEPLOY_PREDICT_ENDPOINT": "/v1/completions"`
`61`	`55`	`},`
`62`		`- {`
`63`		`- "MODEL_DEPLOY_ENABLE_STREAMING": "true"`
`64`		`- },`
`65`	`56`	`{`
`66`	`57`	`"PORT": "8080"`
`67`	`58`	`},`