From fc1a9b8dbc34d48f74c0cc2adb5e3f4056421853 Mon Sep 17 00:00:00 2001
From: Vamsi-klu <nrvamsi13@gmail.com>
Date: Sat, 11 Jul 2026 06:52:08 +0000
Subject: [PATCH 01/12] Make Vertex AI evaluation optional to avoid litellm
 bloat

Base dependency on google-cloud-aiplatform[evaluation] forced litellm,
scikit-learn, tokenizers for all users. Move evaluation extra behind
optional provider extra 'evaluation'. Add lazy import guard with
AirflowOptionalProviderFeatureException in GenerativeModelHook.

Users needing RunEvaluationOperator should install
apache-airflow-providers-google[evaluation].

Fixes: #69323
---
 providers/google/README.rst                   |  2 +-
 providers/google/docs/changelog.rst           | 13 +++++++++++
 providers/google/docs/index.rst               |  5 +++--
 providers/google/pyproject.toml               |  5 ++++-
 .../cloud/hooks/vertex_ai/generative_model.py | 22 ++++++++++++++++++-
 .../hooks/vertex_ai/test_generative_model.py  |  1 +
 .../vertex_ai/test_generative_model.py        |  1 +
 7 files changed, 44 insertions(+), 5 deletions(-)

diff --git a/providers/google/README.rst b/providers/google/README.rst
index bd1c048e1a7e1..9bf97e7a945c5 100644
--- a/providers/google/README.rst
+++ b/providers/google/README.rst
@@ -77,7 +77,7 @@ PIP package                                 Version required
 ``google-auth``                             ``>=2.29.0``
 ``google-auth-httplib2``                    ``>=0.0.1``
 ``google-genai``                            ``>=2.8.0``
-``google-cloud-aiplatform[evaluation]``     ``>=1.155.0``
+``google-cloud-aiplatform``                 ``>=1.155.0``
 ``ray[default]``                            ``>=2.42.0; python_version < "3.13"``
 ``ray[default]``                            ``>=2.49.0; python_version >= "3.13" and python_version < "3.14"``
 ``ray[default]``                            ``>=2.55.0; python_version >= "3.14" and python_version < "3.15"``
diff --git a/providers/google/docs/changelog.rst b/providers/google/docs/changelog.rst
index 8d3603296bd97..2a946acc78967 100644
--- a/providers/google/docs/changelog.rst
+++ b/providers/google/docs/changelog.rst
@@ -44,6 +44,16 @@ Changelog
     because a team name may itself contain ``--`` and the resulting name would be ambiguous.
     See :doc:`/secrets-backends/google-cloud-secret-manager-backend` for the full convention.
 
+Breaking changes
+~~~~~~~~~~~~~~~~
+
+* The Evaluation feature of Vertex AI is now optional. ``RunEvaluationOperator``
+  and ``GenerativeModelHook.run_evaluation`` require
+  ``apache-airflow-providers-google[evaluation]`` extra. Previously
+  ``google-cloud-aiplatform[evaluation]`` was installed unconditionally,
+  pulling ``litellm`` and ``scikit-learn`` for all provider users. To
+  restore old behavior, install with ``pip install apache-airflow-providers-google[evaluation]``.
+
 Features
 ~~~~~~~~
 
@@ -90,9 +100,12 @@ Doc-only
    appropriate section above if needed. Do not delete the lines(!):
    * ``Fix the google provider dataform system test (#70136)``
    * ``Format Dataflow wordcount system-test resource with gofmt (#70174)``
+<<<<<<< HEAD
    * ``Preserve Dataform workflow invocation config (#53843) (#69161)``
    * ``Revert Dataform workflow invocation dict normalization (#69161) (#70395)``
    * ``Prepare providers release 2026-07-22 (#70256)``
+=======
+>>>>>>> e44b736b76 (Make Vertex AI evaluation optional to avoid litellm bloat)
 
 22.2.2
 ......
diff --git a/providers/google/docs/index.rst b/providers/google/docs/index.rst
index 05c764f3c1cce..f8a9e91a5823b 100644
--- a/providers/google/docs/index.rst
+++ b/providers/google/docs/index.rst
@@ -130,7 +130,7 @@ PIP package                                 Version required
 ``google-auth``                             ``>=2.29.0``
 ``google-auth-httplib2``                    ``>=0.0.1``
 ``google-genai``                            ``>=2.8.0``
-``google-cloud-aiplatform[evaluation]``     ``>=1.155.0``
+``google-cloud-aiplatform``                 ``>=1.155.0``
 ``ray[default]``                            ``>=2.42.0; python_version < "3.13"``
 ``ray[default]``                            ``>=2.49.0; python_version >= "3.13" and python_version < "3.14"``
 ``ray[default]``                            ``>=2.55.0; python_version >= "3.14" and python_version < "3.15"``
@@ -238,12 +238,13 @@ Install them when installing from PyPI. For example:
 
 .. code-block:: bash
 
-    pip install apache-airflow-providers-google[cncf.kubernetes]
+    pip install apache-airflow-providers-google[evaluation]
 
 
 ====================  ====================================================
 Extra                 Dependencies
 ====================  ====================================================
+``evaluation``        ``google-cloud-aiplatform[evaluation]>=1.155.0``
 ``cncf.kubernetes``   ``apache-airflow-providers-cncf-kubernetes>=10.1.0``
 ``fab``               ``apache-airflow-providers-fab>=2.0.0``
 ``leveldb``           ``plyvel>=1.5.1; python_version < '3.13'``
diff --git a/providers/google/pyproject.toml b/providers/google/pyproject.toml
index e750891904d1a..43cb82f0729ad 100644
--- a/providers/google/pyproject.toml
+++ b/providers/google/pyproject.toml
@@ -84,7 +84,7 @@ dependencies = [
     # google-cloud-aiplatform doesn't install ray for python 3.12 (issue: https://github.com/googleapis/python-aiplatform/issues/5252).
     # Temporarily lock in ray 2.42.0 which is compatible with python 3.12 until linked issue is solved.
     # Remove the ray dependency as well as google-cloud-bigquery-storage once linked issue is fixed
-    "google-cloud-aiplatform[evaluation]>=1.155.0",
+    "google-cloud-aiplatform>=1.155.0",
     "ray[default]>=2.42.0;python_version<'3.13'",
     "ray[default]>=2.49.0;python_version>='3.13' and python_version <'3.14'",
     "ray[default]>=2.55.0;python_version>='3.14' and python_version <'3.15'",
@@ -165,6 +165,9 @@ dependencies = [
 # The optional dependencies should be modified in place in the generated file
 # Any change in the dependencies is preserved when the file is regenerated
 [project.optional-dependencies]
+"evaluation" = [
+    "google-cloud-aiplatform[evaluation]>=1.155.0",
+]
 "cncf.kubernetes" = [
     "apache-airflow-providers-cncf-kubernetes>=10.1.0",
 ]
diff --git a/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py b/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py
index 06854d313248a..6aae691ad9a67 100644
--- a/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py
+++ b/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py
@@ -26,10 +26,18 @@
 from vertexai.language_models import TextEmbeddingModel
 from vertexai.preview import generative_models as preview_generative_model
 from vertexai.preview.caching import CachedContent
-from vertexai.preview.evaluation import EvalResult, EvalTask
 
+from airflow.providers.common.compat.sdk import AirflowOptionalProviderFeatureException
 from airflow.providers.google.common.hooks.base_google import PROVIDE_PROJECT_ID, GoogleBaseHook
 
+try:
+    from vertexai.preview.evaluation import EvalResult, EvalTask
+except ImportError as e:
+    EvalResult = EvalTask = None  # type: ignore[assignment,misc]
+    _evaluation_import_error = e
+else:
+    _evaluation_import_error = None
+
 
 class GenerativeModelHook(GoogleBaseHook):
     """Hook for Google Cloud Vertex AI Generative Model APIs."""
@@ -64,6 +72,12 @@ def get_eval_task(
         experiment: str,
     ) -> EvalTask:
         """Return an EvalTask object."""
+        if _evaluation_import_error:
+            raise AirflowOptionalProviderFeatureException(
+                "The 'evaluation' extra is required for Vertex AI evaluation. "
+                f"Original error: {_evaluation_import_error}. "
+                "Install with: pip install apache-airflow-providers-google[evaluation]"
+            )
         eval_task = EvalTask(
             dataset=dataset,
             metrics=metrics,
@@ -115,6 +129,12 @@ def run_evaluation(
         :param system_instruction: Optional. An instruction given to the model to guide its behavior.
         :param tools: Optional. A list of tools available to the model during evaluation, such as a data store.
         """
+        if _evaluation_import_error:
+            raise AirflowOptionalProviderFeatureException(
+                "The 'evaluation' extra is required for Vertex AI evaluation. "
+                f"Original error: {_evaluation_import_error}. "
+                "Install with: pip install apache-airflow-providers-google[evaluation]"
+            )
         vertexai.init(project=project_id, location=location, credentials=self.get_credentials())
 
         model = self.get_generative_model(
diff --git a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py
index 3146992dba599..adbdf63b10faf 100644
--- a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py
+++ b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py
@@ -23,6 +23,7 @@
 
 # For no Pydantic environment, we need to skip the tests
 pytest.importorskip("google.cloud.aiplatform_v1")
+pytest.importorskip("vertexai.preview.evaluation")
 
 from vertexai.generative_models import HarmBlockThreshold, HarmCategory, Part, Tool, grounding
 from vertexai.preview.evaluation import MetricPromptTemplateExamples
diff --git a/providers/google/tests/unit/google/cloud/operators/vertex_ai/test_generative_model.py b/providers/google/tests/unit/google/cloud/operators/vertex_ai/test_generative_model.py
index e9bd014eac794..cc3b29b704129 100644
--- a/providers/google/tests/unit/google/cloud/operators/vertex_ai/test_generative_model.py
+++ b/providers/google/tests/unit/google/cloud/operators/vertex_ai/test_generative_model.py
@@ -23,6 +23,7 @@
 # For no Pydantic environment, we need to skip the tests
 pytest.importorskip("google.cloud.aiplatform_v1")
 pytest.importorskip("google.cloud.aiplatform_v1beta1")
+pytest.importorskip("vertexai.preview.evaluation")
 vertexai = pytest.importorskip("vertexai.generative_models")
 from vertexai.generative_models import HarmBlockThreshold, HarmCategory, Tool, grounding
 from vertexai.preview.evaluation import MetricPromptTemplateExamples

From cd4c65991d312ab5dc573cd23a796e86baaff913 Mon Sep 17 00:00:00 2001
From: Vamsi-klu <nrvamsi13@gmail.com>
Date: Sat, 11 Jul 2026 07:26:40 +0000
Subject: [PATCH 02/12] Fix mypy for optional evaluation import

Use TYPE_CHECKING guard and Any fallback to satisfy mypy when
evaluation extra not installed. Runtime guard still raises
AirflowOptionalProviderFeatureException before using None types.

Fixes mypy failure in CI for #69323
---
 .../google/cloud/hooks/vertex_ai/generative_model.py   | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py b/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py
index 6aae691ad9a67..d1a67e89de3b8 100644
--- a/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py
+++ b/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py
@@ -19,7 +19,7 @@
 
 from __future__ import annotations
 
-from typing import Any
+from typing import TYPE_CHECKING, Any
 
 import vertexai
 from vertexai.generative_models import GenerativeModel
@@ -30,11 +30,17 @@
 from airflow.providers.common.compat.sdk import AirflowOptionalProviderFeatureException
 from airflow.providers.google.common.hooks.base_google import PROVIDE_PROJECT_ID, GoogleBaseHook
 
+if TYPE_CHECKING:
+    from vertexai.preview.evaluation import EvalResult, EvalTask
+
 try:
     from vertexai.preview.evaluation import EvalResult, EvalTask
 except ImportError as e:
-    EvalResult = EvalTask = None  # type: ignore[assignment,misc]
     _evaluation_import_error = e
+    # Fallback for mypy: use Any so that calls like EvalTask(...) don't error at type-check time.
+    # At runtime, guard checks _evaluation_import_error and raises before using these.
+    EvalResult = Any  # type: ignore[no-redef]
+    EvalTask = Any  # type: ignore[no-redef]
 else:
     _evaluation_import_error = None
 

From a913bdc86965b8ab3691c41f12a9427b31389cec Mon Sep 17 00:00:00 2001
From: probe <probe@example.com>
Date: Sun, 12 Jul 2026 19:41:10 +0000
Subject: [PATCH 03/12] Document optional Vertex AI evaluation dependency

The provider now keeps evaluation dependencies out of the base install, so the missing-extra path needs regression coverage and operator docs that show users how to opt back in.
---
 .../google/docs/operators/cloud/vertex_ai.rst |  3 +
 ...st_generative_model_optional_evaluation.py | 91 +++++++++++++++++++
 uv.lock                                       | 10 +-
 3 files changed, 101 insertions(+), 3 deletions(-)
 create mode 100644 providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py

diff --git a/providers/google/docs/operators/cloud/vertex_ai.rst b/providers/google/docs/operators/cloud/vertex_ai.rst
index f11b8787ef684..41df2c3c4f4b8 100644
--- a/providers/google/docs/operators/cloud/vertex_ai.rst
+++ b/providers/google/docs/operators/cloud/vertex_ai.rst
@@ -627,6 +627,9 @@ To evaluate a model you can use
 :class:`~airflow.providers.google.cloud.operators.vertex_ai.generative_model.RunEvaluationOperator`.
 The operator returns the evaluation summary metrics in :ref:`XCom <concepts:xcom>` under ``summary_metrics`` key.
 
+Vertex AI evaluation requires the ``evaluation`` extra. Install it with
+``pip install apache-airflow-providers-google[evaluation]``.
+
 .. exampleinclude:: /../../google/tests/system/google/cloud/gen_ai/example_gen_ai_generative_model.py
     :language: python
     :dedent: 4
diff --git a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py
new file mode 100644
index 0000000000000..04608f10e4493
--- /dev/null
+++ b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py
@@ -0,0 +1,91 @@
+#
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+from __future__ import annotations
+
+import importlib
+import sys
+from collections.abc import Iterator
+from types import ModuleType
+from unittest import mock
+
+import pytest
+
+from airflow.providers.common.compat.sdk import AirflowOptionalProviderFeatureException
+
+from unit.google.cloud.utils.base_gcp_mock import mock_base_gcp_hook_default_project_id
+
+pytest.importorskip("google.cloud.aiplatform_v1")
+
+EVALUATION_MODULE = "vertexai.preview.evaluation"
+HOOK_MODULE = "airflow.providers.google.cloud.hooks.vertex_ai.generative_model"
+BASE_HOOK_INIT = "airflow.providers.google.common.hooks.base_google.GoogleBaseHook.__init__"
+INSTALL_EXTRA_REGEX = r"apache-airflow-providers-google\[evaluation\]"
+
+_MISSING = object()
+
+
+@pytest.fixture
+def generative_model_hook_module_without_evaluation() -> Iterator[ModuleType]:
+    original_evaluation_module = sys.modules.get(EVALUATION_MODULE, _MISSING)
+    original_hook_module = sys.modules.get(HOOK_MODULE, _MISSING)
+
+    sys.modules[EVALUATION_MODULE] = None
+    hook_module = importlib.import_module(HOOK_MODULE)
+    hook_module = importlib.reload(hook_module)
+
+    try:
+        yield hook_module
+    finally:
+        if original_evaluation_module is _MISSING:
+            sys.modules.pop(EVALUATION_MODULE, None)
+        else:
+            sys.modules[EVALUATION_MODULE] = original_evaluation_module
+
+        if original_hook_module is _MISSING:
+            sys.modules.pop(HOOK_MODULE, None)
+        else:
+            importlib.reload(hook_module)
+
+
+def test_get_eval_task_raises_optional_provider_feature_exception_without_evaluation_extra(
+    generative_model_hook_module_without_evaluation: ModuleType,
+):
+    with mock.patch(BASE_HOOK_INIT, new=mock_base_gcp_hook_default_project_id):
+        hook = generative_model_hook_module_without_evaluation.GenerativeModelHook()
+
+    with pytest.raises(AirflowOptionalProviderFeatureException, match=INSTALL_EXTRA_REGEX):
+        hook.get_eval_task(dataset={}, metrics=[], experiment="test-experiment")
+
+
+def test_run_evaluation_raises_optional_provider_feature_exception_without_evaluation_extra(
+    generative_model_hook_module_without_evaluation: ModuleType,
+):
+    with mock.patch(BASE_HOOK_INIT, new=mock_base_gcp_hook_default_project_id):
+        hook = generative_model_hook_module_without_evaluation.GenerativeModelHook()
+
+    with pytest.raises(AirflowOptionalProviderFeatureException, match=INSTALL_EXTRA_REGEX):
+        hook.run_evaluation(
+            project_id="test-project",
+            location="us-central1",
+            pretrained_model="gemini-pro",
+            eval_dataset={},
+            metrics=[],
+            experiment_name="test-experiment",
+            experiment_run_name="test-run",
+            prompt_template="{prompt}",
+        )
diff --git a/uv.lock b/uv.lock
index 3b52d8d26c2f2..09fc6ab614252 100644
--- a/uv.lock
+++ b/uv.lock
@@ -5520,7 +5520,7 @@ dependencies = [
     { name = "google-api-python-client" },
     { name = "google-auth" },
     { name = "google-auth-httplib2" },
-    { name = "google-cloud-aiplatform", extra = ["evaluation"] },
+    { name = "google-cloud-aiplatform" },
     { name = "google-cloud-alloydb" },
     { name = "google-cloud-automl" },
     { name = "google-cloud-batch" },
@@ -5591,6 +5591,9 @@ cncf-kubernetes = [
 common-messaging = [
     { name = "apache-airflow-providers-common-messaging" },
 ]
+evaluation = [
+    { name = "google-cloud-aiplatform", extra = ["evaluation"] },
+]
 fab = [
     { name = "apache-airflow-providers-fab" },
 ]
@@ -5713,7 +5716,8 @@ requires-dist = [
     { name = "google-api-python-client", specifier = ">=2.0.2" },
     { name = "google-auth", specifier = ">=2.29.0" },
     { name = "google-auth-httplib2", specifier = ">=0.0.1" },
-    { name = "google-cloud-aiplatform", extras = ["evaluation"], specifier = ">=1.155.0" },
+    { name = "google-cloud-aiplatform", specifier = ">=1.155.0" },
+    { name = "google-cloud-aiplatform", extras = ["evaluation"], marker = "extra == 'evaluation'", specifier = ">=1.155.0" },
     { name = "google-cloud-alloydb", specifier = ">=0.4.0" },
     { name = "google-cloud-automl", specifier = ">=2.12.0" },
     { name = "google-cloud-batch", specifier = ">=0.13.0" },
@@ -5777,7 +5781,7 @@ requires-dist = [
     { name = "tenacity", specifier = ">=8.3.0" },
     { name = "types-protobuf", specifier = ">=5.27.0,!=5.29.1.20250402" },
 ]
-provides-extras = ["cncf-kubernetes", "fab", "leveldb", "oracle", "facebook", "amazon", "apache-cassandra", "microsoft-azure", "microsoft-mssql", "mongo", "mysql", "openlineage", "postgres", "presto", "salesforce", "sftp", "ssh", "trino", "http", "standard", "common-messaging"]
+provides-extras = ["evaluation", "cncf-kubernetes", "fab", "leveldb", "oracle", "facebook", "amazon", "apache-cassandra", "microsoft-azure", "microsoft-mssql", "mongo", "mysql", "openlineage", "postgres", "presto", "salesforce", "sftp", "ssh", "trino", "http", "standard", "common-messaging"]
 
 [package.metadata.requires-dev]
 dev = [

From ac4c0bf547f416a9730efa35540110b00952195a Mon Sep 17 00:00:00 2001
From: Ramachandra Nalam <nalamvamsi13@gmail.com>
Date: Sun, 12 Jul 2026 16:54:02 -0700
Subject: [PATCH 04/12] Fix CI failures after optional Vertex AI evaluation
 change

---
 .../cloud/hooks/vertex_ai/generative_model.py | 20 +++++++++----------
 .../gen_ai/example_gen_ai_generative_model.py | 16 ++++++++-------
 2 files changed, 19 insertions(+), 17 deletions(-)

diff --git a/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py b/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py
index d1a67e89de3b8..506afab210e71 100644
--- a/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py
+++ b/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py
@@ -33,16 +33,16 @@
 if TYPE_CHECKING:
     from vertexai.preview.evaluation import EvalResult, EvalTask
 
-try:
-    from vertexai.preview.evaluation import EvalResult, EvalTask
-except ImportError as e:
-    _evaluation_import_error = e
-    # Fallback for mypy: use Any so that calls like EvalTask(...) don't error at type-check time.
-    # At runtime, guard checks _evaluation_import_error and raises before using these.
-    EvalResult = Any  # type: ignore[no-redef]
-    EvalTask = Any  # type: ignore[no-redef]
-else:
-    _evaluation_import_error = None
+_evaluation_import_error: ImportError | None = None
+
+if not TYPE_CHECKING:
+    try:
+        from vertexai.preview.evaluation import EvalResult, EvalTask
+    except ImportError as e:
+        _evaluation_import_error = e
+        # Runtime fallback: guard checks _evaluation_import_error and raises before using these.
+        EvalResult = Any
+        EvalTask = Any
 
 
 class GenerativeModelHook(GoogleBaseHook):
diff --git a/providers/google/tests/system/google/cloud/gen_ai/example_gen_ai_generative_model.py b/providers/google/tests/system/google/cloud/gen_ai/example_gen_ai_generative_model.py
index c9a6f9ffcafa4..31764d07327cb 100644
--- a/providers/google/tests/system/google/cloud/gen_ai/example_gen_ai_generative_model.py
+++ b/providers/google/tests/system/google/cloud/gen_ai/example_gen_ai_generative_model.py
@@ -42,6 +42,7 @@
 )
 
 from airflow.models.dag import DAG
+from airflow.providers.common.compat.sdk import AirflowOptionalProviderFeatureException
 from airflow.providers.google.cloud.operators.gen_ai import (
     GenAICountTokensOperator,
     GenAICreateCachedContentOperator,
@@ -169,13 +170,14 @@ def _get_actual_models(key) -> dict[str, str]:
 
 
 def _get_metrics():
-    """
-    Lazily import and return the metrics list.
-
-    This avoids slow imports during DAG parsing by deferring the import
-    until the operator is actually created.
-    """
-    from vertexai.preview.evaluation import MetricPromptTemplateExamples
+    """Return metrics without importing optional evaluation dependencies during module import."""
+    try:
+        from vertexai.preview.evaluation import MetricPromptTemplateExamples
+    except ImportError as e:
+        raise AirflowOptionalProviderFeatureException(
+            "The 'evaluation' extra is required for Vertex AI evaluation. "
+            "Install with: pip install apache-airflow-providers-google[evaluation]"
+        ) from e
 
     return [
         MetricPromptTemplateExamples.Pointwise.SUMMARIZATION_QUALITY,

From 32e72121eda4f9bf010a86c1d5e04b66634f9c36 Mon Sep 17 00:00:00 2001
From: Ramachandra Nalam <nalamvamsi13@gmail.com>
Date: Sun, 12 Jul 2026 18:02:19 -0700
Subject: [PATCH 05/12] Fix optional evaluation test typing

---
 .../test_generative_model_optional_evaluation.py   | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py
index 04608f10e4493..cc3204886b487 100644
--- a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py
+++ b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py
@@ -21,6 +21,7 @@
 import sys
 from collections.abc import Iterator
 from types import ModuleType
+from typing import cast
 from unittest import mock
 
 import pytest
@@ -41,10 +42,11 @@
 
 @pytest.fixture
 def generative_model_hook_module_without_evaluation() -> Iterator[ModuleType]:
-    original_evaluation_module = sys.modules.get(EVALUATION_MODULE, _MISSING)
-    original_hook_module = sys.modules.get(HOOK_MODULE, _MISSING)
+    modules = cast("dict[str, object]", sys.modules)
+    original_evaluation_module = modules.get(EVALUATION_MODULE, _MISSING)
+    original_hook_module = modules.get(HOOK_MODULE, _MISSING)
 
-    sys.modules[EVALUATION_MODULE] = None
+    modules[EVALUATION_MODULE] = None
     hook_module = importlib.import_module(HOOK_MODULE)
     hook_module = importlib.reload(hook_module)
 
@@ -52,12 +54,12 @@ def generative_model_hook_module_without_evaluation() -> Iterator[ModuleType]:
         yield hook_module
     finally:
         if original_evaluation_module is _MISSING:
-            sys.modules.pop(EVALUATION_MODULE, None)
+            modules.pop(EVALUATION_MODULE, None)
         else:
-            sys.modules[EVALUATION_MODULE] = original_evaluation_module
+            modules[EVALUATION_MODULE] = original_evaluation_module
 
         if original_hook_module is _MISSING:
-            sys.modules.pop(HOOK_MODULE, None)
+            modules.pop(HOOK_MODULE, None)
         else:
             importlib.reload(hook_module)
 

From 1141e4fe138e548350d84c10e46364e76019c6ff Mon Sep 17 00:00:00 2001
From: Vamsi-klu <nrvamsi13@gmail.com>
Date: Tue, 4 Aug 2026 06:55:44 +0000
Subject: [PATCH 06/12] Stop optional evaluation tests leaking into other
 Vertex AI tests

Reloading the hook module rebound GenerativeModelHook, so tests that
imported the class at collection time patched a different class object
and hit the real EvalTask. Patching the import-guard variable tests the
same behavior without touching module state.
---
 ...st_generative_model_optional_evaluation.py | 50 ++++++-------------
 1 file changed, 14 insertions(+), 36 deletions(-)

diff --git a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py
index cc3204886b487..139760c91845c 100644
--- a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py
+++ b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py
@@ -17,11 +17,6 @@
 # under the License.
 from __future__ import annotations
 
-import importlib
-import sys
-from collections.abc import Iterator
-from types import ModuleType
-from typing import cast
 from unittest import mock
 
 import pytest
@@ -32,54 +27,37 @@
 
 pytest.importorskip("google.cloud.aiplatform_v1")
 
-EVALUATION_MODULE = "vertexai.preview.evaluation"
+from airflow.providers.google.cloud.hooks.vertex_ai.generative_model import (
+    GenerativeModelHook,
+)
+
 HOOK_MODULE = "airflow.providers.google.cloud.hooks.vertex_ai.generative_model"
 BASE_HOOK_INIT = "airflow.providers.google.common.hooks.base_google.GoogleBaseHook.__init__"
 INSTALL_EXTRA_REGEX = r"apache-airflow-providers-google\[evaluation\]"
 
-_MISSING = object()
+# Patching the guard variable (instead of reloading the hook module with the evaluation import
+# blocked) keeps the module and class objects stable for other test files in the same session.
+MISSING_EVALUATION_IMPORT = ImportError("No module named 'vertexai.preview.evaluation'")
 
 
 @pytest.fixture
-def generative_model_hook_module_without_evaluation() -> Iterator[ModuleType]:
-    modules = cast("dict[str, object]", sys.modules)
-    original_evaluation_module = modules.get(EVALUATION_MODULE, _MISSING)
-    original_hook_module = modules.get(HOOK_MODULE, _MISSING)
-
-    modules[EVALUATION_MODULE] = None
-    hook_module = importlib.import_module(HOOK_MODULE)
-    hook_module = importlib.reload(hook_module)
-
-    try:
-        yield hook_module
-    finally:
-        if original_evaluation_module is _MISSING:
-            modules.pop(EVALUATION_MODULE, None)
-        else:
-            modules[EVALUATION_MODULE] = original_evaluation_module
-
-        if original_hook_module is _MISSING:
-            modules.pop(HOOK_MODULE, None)
-        else:
-            importlib.reload(hook_module)
+def hook() -> GenerativeModelHook:
+    with mock.patch(BASE_HOOK_INIT, new=mock_base_gcp_hook_default_project_id):
+        return GenerativeModelHook()
 
 
+@mock.patch(f"{HOOK_MODULE}._evaluation_import_error", MISSING_EVALUATION_IMPORT)
 def test_get_eval_task_raises_optional_provider_feature_exception_without_evaluation_extra(
-    generative_model_hook_module_without_evaluation: ModuleType,
+    hook: GenerativeModelHook,
 ):
-    with mock.patch(BASE_HOOK_INIT, new=mock_base_gcp_hook_default_project_id):
-        hook = generative_model_hook_module_without_evaluation.GenerativeModelHook()
-
     with pytest.raises(AirflowOptionalProviderFeatureException, match=INSTALL_EXTRA_REGEX):
         hook.get_eval_task(dataset={}, metrics=[], experiment="test-experiment")
 
 
+@mock.patch(f"{HOOK_MODULE}._evaluation_import_error", MISSING_EVALUATION_IMPORT)
 def test_run_evaluation_raises_optional_provider_feature_exception_without_evaluation_extra(
-    generative_model_hook_module_without_evaluation: ModuleType,
+    hook: GenerativeModelHook,
 ):
-    with mock.patch(BASE_HOOK_INIT, new=mock_base_gcp_hook_default_project_id):
-        hook = generative_model_hook_module_without_evaluation.GenerativeModelHook()
-
     with pytest.raises(AirflowOptionalProviderFeatureException, match=INSTALL_EXTRA_REGEX):
         hook.run_evaluation(
             project_id="test-project",

From 4c25fa3cbe3a48569269c6fcf12fb83848acd789 Mon Sep 17 00:00:00 2001
From: Vamsi-klu <nrvamsi13@gmail.com>
Date: Thu, 6 Aug 2026 03:04:22 +0000
Subject: [PATCH 07/12] Make Vertex AI evaluation tests runnable in CI without
 extra

Address review comment that pytest.importorskip('vertexai.preview.evaluation')
caused the only tests in two files to be skipped in the default CI image
which installs via uv sync --all-packages --group ci-image without extras.

Replace the skip with a try/except import that falls back to a MagicMock
providing MetricPromptTemplateExamples.Pointwise attributes. Tests now
execute with mocks in default CI, while the missing-extra path remains
covered by test_generative_model_optional_evaluation.

Also fix changelog merge markers left from previous rebase.
---
 providers/google/docs/changelog.rst                   |  3 ---
 .../cloud/hooks/vertex_ai/test_generative_model.py    | 11 +++++++++--
 .../operators/vertex_ai/test_generative_model.py      | 11 +++++++++--
 3 files changed, 18 insertions(+), 7 deletions(-)

diff --git a/providers/google/docs/changelog.rst b/providers/google/docs/changelog.rst
index 2a946acc78967..d77f0e09a2bd9 100644
--- a/providers/google/docs/changelog.rst
+++ b/providers/google/docs/changelog.rst
@@ -100,12 +100,9 @@ Doc-only
    appropriate section above if needed. Do not delete the lines(!):
    * ``Fix the google provider dataform system test (#70136)``
    * ``Format Dataflow wordcount system-test resource with gofmt (#70174)``
-<<<<<<< HEAD
    * ``Preserve Dataform workflow invocation config (#53843) (#69161)``
    * ``Revert Dataform workflow invocation dict normalization (#69161) (#70395)``
    * ``Prepare providers release 2026-07-22 (#70256)``
-=======
->>>>>>> e44b736b76 (Make Vertex AI evaluation optional to avoid litellm bloat)
 
 22.2.2
 ......
diff --git a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py
index adbdf63b10faf..752f4413b2603 100644
--- a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py
+++ b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py
@@ -23,10 +23,17 @@
 
 # For no Pydantic environment, we need to skip the tests
 pytest.importorskip("google.cloud.aiplatform_v1")
-pytest.importorskip("vertexai.preview.evaluation")
 
 from vertexai.generative_models import HarmBlockThreshold, HarmCategory, Part, Tool, grounding
-from vertexai.preview.evaluation import MetricPromptTemplateExamples
+
+try:
+    from vertexai.preview.evaluation import MetricPromptTemplateExamples
+except ImportError:
+    MetricPromptTemplateExamples = mock.MagicMock()
+    MetricPromptTemplateExamples.Pointwise.SUMMARIZATION_QUALITY = "summarization_quality"
+    MetricPromptTemplateExamples.Pointwise.GROUNDEDNESS = "groundedness"
+    MetricPromptTemplateExamples.Pointwise.VERBOSITY = "verbosity"
+    MetricPromptTemplateExamples.Pointwise.INSTRUCTION_FOLLOWING = "instruction_following"
 
 from airflow.providers.google.cloud.hooks.vertex_ai.generative_model import (
     GenerativeModelHook,
diff --git a/providers/google/tests/unit/google/cloud/operators/vertex_ai/test_generative_model.py b/providers/google/tests/unit/google/cloud/operators/vertex_ai/test_generative_model.py
index cc3b29b704129..5cd5eebe6124d 100644
--- a/providers/google/tests/unit/google/cloud/operators/vertex_ai/test_generative_model.py
+++ b/providers/google/tests/unit/google/cloud/operators/vertex_ai/test_generative_model.py
@@ -23,10 +23,17 @@
 # For no Pydantic environment, we need to skip the tests
 pytest.importorskip("google.cloud.aiplatform_v1")
 pytest.importorskip("google.cloud.aiplatform_v1beta1")
-pytest.importorskip("vertexai.preview.evaluation")
 vertexai = pytest.importorskip("vertexai.generative_models")
 from vertexai.generative_models import HarmBlockThreshold, HarmCategory, Tool, grounding
-from vertexai.preview.evaluation import MetricPromptTemplateExamples
+
+try:
+    from vertexai.preview.evaluation import MetricPromptTemplateExamples
+except ImportError:
+    MetricPromptTemplateExamples = mock.MagicMock()
+    MetricPromptTemplateExamples.Pointwise.SUMMARIZATION_QUALITY = "summarization_quality"
+    MetricPromptTemplateExamples.Pointwise.GROUNDEDNESS = "groundedness"
+    MetricPromptTemplateExamples.Pointwise.VERBOSITY = "verbosity"
+    MetricPromptTemplateExamples.Pointwise.INSTRUCTION_FOLLOWING = "instruction_following"
 
 from airflow.providers.google.cloud.operators.vertex_ai.generative_model import (
     RunEvaluationOperator,

From 74776c0f17cd1563a074884de8918cfbd8b34b17 Mon Sep 17 00:00:00 2001
From: Vamsi-klu <nrvamsi13@gmail.com>
Date: Tue, 11 Aug 2026 05:01:58 +0000
Subject: [PATCH 08/12] Move the evaluation breaking change to 23.0.0 and make
 the guard mypy-clean

22.3.0 shipped on 2026-08-08, so the breaking-change note moves into a new
23.0.0 section (major bump, next release) instead of sitting inside an
already-released section.

Collapses the two copies of the "install the evaluation extra" raise into a
single _raise_if_evaluation_unavailable() helper, so the message and the
install hint live in one place, and chains the original ImportError as the
cause. ImportError (not ModuleNotFoundError) stays the caught type, so a
partially installed evaluation stack gets the same hint rather than a raw
traceback. EvalResult is only ever an annotation, so it is now imported under
TYPE_CHECKING alone; the dead Any fallbacks are gone.

RunEvaluationOperator has no evaluation-specific code of its own -- it calls
GenerativeModelHook.run_evaluation -- so the single hook-side guard is what
surfaces the exception from the operator too.

Fixes the MyPy providers failure: assigning MagicMock over the imported
MetricPromptTemplateExamples is "Cannot assign to a type" [misc]. The fallback
is now built in one expression with a targeted ignore, and still engages only
on ImportError, so environments with the real extra keep testing the real
metric values.
---
 providers/google/docs/changelog.rst           | 25 +++++++------
 .../cloud/hooks/vertex_ai/generative_model.py | 35 ++++++++++---------
 .../hooks/vertex_ai/test_generative_model.py  | 16 ++++++---
 .../vertex_ai/test_generative_model.py        | 16 ++++++---
 4 files changed, 56 insertions(+), 36 deletions(-)

diff --git a/providers/google/docs/changelog.rst b/providers/google/docs/changelog.rst
index 5801a45a36c63..cb14ca4f35d58 100644
--- a/providers/google/docs/changelog.rst
+++ b/providers/google/docs/changelog.rst
@@ -27,6 +27,21 @@
 Changelog
 ---------
 
+23.0.0
+......
+
+Breaking changes
+~~~~~~~~~~~~~~~~
+
+* The Vertex AI evaluation feature is now optional. ``RunEvaluationOperator`` and
+  ``GenerativeModelHook.run_evaluation`` / ``GenerativeModelHook.get_eval_task`` now require the
+  ``evaluation`` extra. Previously ``google-cloud-aiplatform[evaluation]`` was installed
+  unconditionally, pulling ``litellm``, ``scikit-learn`` and their transitive dependencies into every
+  Google provider install. To restore the previous behaviour, install
+  ``pip install 'apache-airflow-providers-google[evaluation]'``. Without the extra the provider
+  imports and all non-evaluation features keep working; the evaluation code paths raise
+  ``AirflowOptionalProviderFeatureException`` with this install hint.
+
 22.3.0
 ......
 
@@ -44,16 +59,6 @@ Changelog
     because a team name may itself contain ``--`` and the resulting name would be ambiguous.
     See :doc:`/secrets-backends/google-cloud-secret-manager-backend` for the full convention.
 
-Breaking changes
-~~~~~~~~~~~~~~~~
-
-* The Evaluation feature of Vertex AI is now optional. ``RunEvaluationOperator``
-  and ``GenerativeModelHook.run_evaluation`` require
-  ``apache-airflow-providers-google[evaluation]`` extra. Previously
-  ``google-cloud-aiplatform[evaluation]`` was installed unconditionally,
-  pulling ``litellm`` and ``scikit-learn`` for all provider users. To
-  restore old behavior, install with ``pip install apache-airflow-providers-google[evaluation]``.
-
 Features
 ~~~~~~~~
 
diff --git a/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py b/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py
index 506afab210e71..0dd8277e1b040 100644
--- a/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py
+++ b/providers/google/src/airflow/providers/google/cloud/hooks/vertex_ai/generative_model.py
@@ -33,16 +33,29 @@
 if TYPE_CHECKING:
     from vertexai.preview.evaluation import EvalResult, EvalTask
 
+# ``vertexai.preview.evaluation`` pulls in litellm and scikit-learn, so it ships in the optional
+# "evaluation" extra rather than in the base provider. Importing this module must stay safe without
+# it: the failure is remembered here and only re-raised from the evaluation code paths.
 _evaluation_import_error: ImportError | None = None
 
 if not TYPE_CHECKING:
+    # ImportError rather than ModuleNotFoundError: a partially installed evaluation stack should get
+    # the same install hint instead of a raw traceback.
     try:
-        from vertexai.preview.evaluation import EvalResult, EvalTask
+        from vertexai.preview.evaluation import EvalTask
     except ImportError as e:
         _evaluation_import_error = e
-        # Runtime fallback: guard checks _evaluation_import_error and raises before using these.
-        EvalResult = Any
-        EvalTask = Any
+
+
+def _raise_if_evaluation_unavailable() -> None:
+    """Raise with an install hint if the optional ``evaluation`` extra is not installed."""
+    if _evaluation_import_error is None:
+        return
+    raise AirflowOptionalProviderFeatureException(
+        "The 'evaluation' extra is required for Vertex AI evaluation. "
+        f"Original error: {_evaluation_import_error}. "
+        "Install with: pip install apache-airflow-providers-google[evaluation]"
+    ) from _evaluation_import_error
 
 
 class GenerativeModelHook(GoogleBaseHook):
@@ -78,12 +91,7 @@ def get_eval_task(
         experiment: str,
     ) -> EvalTask:
         """Return an EvalTask object."""
-        if _evaluation_import_error:
-            raise AirflowOptionalProviderFeatureException(
-                "The 'evaluation' extra is required for Vertex AI evaluation. "
-                f"Original error: {_evaluation_import_error}. "
-                "Install with: pip install apache-airflow-providers-google[evaluation]"
-            )
+        _raise_if_evaluation_unavailable()
         eval_task = EvalTask(
             dataset=dataset,
             metrics=metrics,
@@ -135,12 +143,7 @@ def run_evaluation(
         :param system_instruction: Optional. An instruction given to the model to guide its behavior.
         :param tools: Optional. A list of tools available to the model during evaluation, such as a data store.
         """
-        if _evaluation_import_error:
-            raise AirflowOptionalProviderFeatureException(
-                "The 'evaluation' extra is required for Vertex AI evaluation. "
-                f"Original error: {_evaluation_import_error}. "
-                "Install with: pip install apache-airflow-providers-google[evaluation]"
-            )
+        _raise_if_evaluation_unavailable()
         vertexai.init(project=project_id, location=location, credentials=self.get_credentials())
 
         model = self.get_generative_model(
diff --git a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py
index 752f4413b2603..a68cd9bb557f0 100644
--- a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py
+++ b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py
@@ -29,11 +29,17 @@
 try:
     from vertexai.preview.evaluation import MetricPromptTemplateExamples
 except ImportError:
-    MetricPromptTemplateExamples = mock.MagicMock()
-    MetricPromptTemplateExamples.Pointwise.SUMMARIZATION_QUALITY = "summarization_quality"
-    MetricPromptTemplateExamples.Pointwise.GROUNDEDNESS = "groundedness"
-    MetricPromptTemplateExamples.Pointwise.VERBOSITY = "verbosity"
-    MetricPromptTemplateExamples.Pointwise.INSTRUCTION_FOLLOWING = "instruction_following"
+    # Only when the optional "evaluation" extra is absent: stand in for the metric constants so the
+    # tests below still run against the real hook/operator. With the extra installed the real values
+    # are used. Built in one expression so a single ignore covers it -- mypy sees a class here.
+    MetricPromptTemplateExamples = mock.MagicMock(  # type: ignore[assignment,misc]
+        Pointwise=mock.MagicMock(
+            SUMMARIZATION_QUALITY="summarization_quality",
+            GROUNDEDNESS="groundedness",
+            VERBOSITY="verbosity",
+            INSTRUCTION_FOLLOWING="instruction_following",
+        )
+    )
 
 from airflow.providers.google.cloud.hooks.vertex_ai.generative_model import (
     GenerativeModelHook,
diff --git a/providers/google/tests/unit/google/cloud/operators/vertex_ai/test_generative_model.py b/providers/google/tests/unit/google/cloud/operators/vertex_ai/test_generative_model.py
index 5cd5eebe6124d..c17bbe09e1ba5 100644
--- a/providers/google/tests/unit/google/cloud/operators/vertex_ai/test_generative_model.py
+++ b/providers/google/tests/unit/google/cloud/operators/vertex_ai/test_generative_model.py
@@ -29,11 +29,17 @@
 try:
     from vertexai.preview.evaluation import MetricPromptTemplateExamples
 except ImportError:
-    MetricPromptTemplateExamples = mock.MagicMock()
-    MetricPromptTemplateExamples.Pointwise.SUMMARIZATION_QUALITY = "summarization_quality"
-    MetricPromptTemplateExamples.Pointwise.GROUNDEDNESS = "groundedness"
-    MetricPromptTemplateExamples.Pointwise.VERBOSITY = "verbosity"
-    MetricPromptTemplateExamples.Pointwise.INSTRUCTION_FOLLOWING = "instruction_following"
+    # Only when the optional "evaluation" extra is absent: stand in for the metric constants so the
+    # tests below still run against the real hook/operator. With the extra installed the real values
+    # are used. Built in one expression so a single ignore covers it -- mypy sees a class here.
+    MetricPromptTemplateExamples = mock.MagicMock(  # type: ignore[assignment,misc]
+        Pointwise=mock.MagicMock(
+            SUMMARIZATION_QUALITY="summarization_quality",
+            GROUNDEDNESS="groundedness",
+            VERBOSITY="verbosity",
+            INSTRUCTION_FOLLOWING="instruction_following",
+        )
+    )
 
 from airflow.providers.google.cloud.operators.vertex_ai.generative_model import (
     RunEvaluationOperator,

From 8bbc35961c50aa862ac1c0fdef78e81f9ba30a21 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Wed, 19 Aug 2026 01:06:34 +0000
Subject: [PATCH 09/12] Keep Vertex AI evaluation tests running without the
 extra

CI images do not install the Google provider evaluation extra, so the
module-level import guard is set before the happy-path test mocks apply
and the suite fails on a missing sklearn install.
---
 .../unit/google/cloud/hooks/vertex_ai/test_generative_model.py   | 1 +
 1 file changed, 1 insertion(+)

diff --git a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py
index a68cd9bb557f0..3f350f943974c 100644
--- a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py
+++ b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model.py
@@ -157,6 +157,7 @@ def setup_method(self):
             self.hook = GenerativeModelHook(gcp_conn_id=TEST_GCP_CONN_ID)
             self.hook.get_credentials = self.dummy_get_credentials
 
+    @mock.patch(GENERATIVE_MODEL_STRING.format("_evaluation_import_error"), None)
     @mock.patch(GENERATIVE_MODEL_STRING.format("GenerativeModelHook.get_generative_model"))
     @mock.patch(GENERATIVE_MODEL_STRING.format("GenerativeModelHook.get_eval_task"))
     def test_run_evaluation(self, mock_eval_task, mock_model) -> None:

From 1b73602a7d1fb33ca001f2b84755bb4a0f1ea68d Mon Sep 17 00:00:00 2001
From: Vamsi-klu <nrvamsi13@gmail.com>
Date: Wed, 19 Aug 2026 06:43:12 +0000
Subject: [PATCH 10/12] Cover sklearn-missing Vertex AI evaluation extra in
 hook tests

CI images do not install the Google evaluation extra, so the happy-path
evaluation tests fail on a missing sklearn import unless the optional
extra guard is exercised for that exact ImportError.
---
 ...st_generative_model_optional_evaluation.py | 45 ++++++++++++-------
 1 file changed, 29 insertions(+), 16 deletions(-)

diff --git a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py
index 139760c91845c..26f4ad6c44b74 100644
--- a/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py
+++ b/providers/google/tests/unit/google/cloud/hooks/vertex_ai/test_generative_model_optional_evaluation.py
@@ -37,7 +37,16 @@
 
 # Patching the guard variable (instead of reloading the hook module with the evaluation import
 # blocked) keeps the module and class objects stable for other test files in the same session.
-MISSING_EVALUATION_IMPORT = ImportError("No module named 'vertexai.preview.evaluation'")
+MISSING_EVALUATION_ERRORS = (
+    pytest.param(
+        ImportError("No module named 'vertexai.preview.evaluation'"),
+        id="missing-evaluation-module",
+    ),
+    pytest.param(
+        ImportError("No module named 'sklearn'"),
+        id="missing-sklearn",
+    ),
+)
 
 
 @pytest.fixture
@@ -46,26 +55,30 @@ def hook() -> GenerativeModelHook:
         return GenerativeModelHook()
 
 
-@mock.patch(f"{HOOK_MODULE}._evaluation_import_error", MISSING_EVALUATION_IMPORT)
+@pytest.mark.parametrize("import_error", MISSING_EVALUATION_ERRORS)
 def test_get_eval_task_raises_optional_provider_feature_exception_without_evaluation_extra(
     hook: GenerativeModelHook,
+    import_error: ImportError,
 ):
-    with pytest.raises(AirflowOptionalProviderFeatureException, match=INSTALL_EXTRA_REGEX):
-        hook.get_eval_task(dataset={}, metrics=[], experiment="test-experiment")
+    with mock.patch(f"{HOOK_MODULE}._evaluation_import_error", import_error):
+        with pytest.raises(AirflowOptionalProviderFeatureException, match=INSTALL_EXTRA_REGEX):
+            hook.get_eval_task(dataset={}, metrics=[], experiment="test-experiment")
 
 
-@mock.patch(f"{HOOK_MODULE}._evaluation_import_error", MISSING_EVALUATION_IMPORT)
+@pytest.mark.parametrize("import_error", MISSING_EVALUATION_ERRORS)
 def test_run_evaluation_raises_optional_provider_feature_exception_without_evaluation_extra(
     hook: GenerativeModelHook,
+    import_error: ImportError,
 ):
-    with pytest.raises(AirflowOptionalProviderFeatureException, match=INSTALL_EXTRA_REGEX):
-        hook.run_evaluation(
-            project_id="test-project",
-            location="us-central1",
-            pretrained_model="gemini-pro",
-            eval_dataset={},
-            metrics=[],
-            experiment_name="test-experiment",
-            experiment_run_name="test-run",
-            prompt_template="{prompt}",
-        )
+    with mock.patch(f"{HOOK_MODULE}._evaluation_import_error", import_error):
+        with pytest.raises(AirflowOptionalProviderFeatureException, match=INSTALL_EXTRA_REGEX):
+            hook.run_evaluation(
+                project_id="test-project",
+                location="us-central1",
+                pretrained_model="gemini-pro",
+                eval_dataset={},
+                metrics=[],
+                experiment_name="test-experiment",
+                experiment_run_name="test-run",
+                prompt_template="{prompt}",
+            )

From 27fc3fcf7f887bb4389c1ad9c0794c9e9bdf39e5 Mon Sep 17 00:00:00 2001
From: Vamsi-klu <nrvamsi13@gmail.com>
Date: Wed, 19 Aug 2026 06:49:31 +0000
Subject: [PATCH 11/12] Re-trigger Apache CI for Vertex AI evaluation tests


From 2acc942c46f7921d3aecbe2f7b4df3ff8c934aea Mon Sep 17 00:00:00 2001
From: Vamsi-klu <nrvamsi13@gmail.com>
Date: Wed, 19 Aug 2026 07:57:30 +0000
Subject: [PATCH 12/12] Run Apache CI from the contributor account

Fork workflows were waiting on approval after a bot push.