Skip to content

feat: GenAI SDK client(evals) - Add agent run in run_inference #5885

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
copybara-service merged 1 commit into from
Oct 20, 2025
Merged

feat: GenAI SDK client(evals) - Add agent run in run_inference #5885

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 12 additions & 0 deletions tests/unit/vertexai/genai/replays/test_evaluate_instances.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -238,6 +238,18 @@ def test_inference_with_prompt_template(client):
assert inference_result.gcs_source is None


def test_run_inference_with_agent(client):
test_df = pd.DataFrame(
{"prompt": ["agent prompt"], "session_inputs": ['{"user_id": "user_123"}']}
)
inference_result = client.evals.run_inference(
agent="projects/977012026409/locations/us-central1/reasoningEngines/7188347537655332864",
src=test_df,
)
assert inference_result.candidate_name == "agent"
assert inference_result.gcs_source is None


pytestmark = pytest_helper.setup(
file=__file__,
globals_for_file=globals(),
Expand Down
320 changes: 320 additions & 0 deletions tests/unit/vertexai/genai/test_evals.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -200,6 +200,9 @@ def setup_method(self):
importlib.reload(_evals_metric_handlers)
importlib.reload(_genai.evals)

if hasattr(_evals_common._thread_local_data, "agent_engine_instances"):
del _evals_common._thread_local_data.agent_engine_instances

vertexai.init(
project=_TEST_PROJECT,
location=_TEST_LOCATION,
Expand Down Expand Up @@ -967,6 +970,227 @@ def test_inference_with_multimodal_content(
assert inference_result.candidate_name == "gemini-pro"
assert inference_result.gcs_source is None

@mock.patch.object(_evals_utils, "EvalDatasetLoader")
@mock.patch("vertexai._genai._evals_common.vertexai.Client")
def test_run_inference_with_agent_engine_and_session_inputs_dict(
self,
mock_vertexai_client,
mock_eval_dataset_loader,
):
mock_df = pd.DataFrame(
{
"prompt": ["agent prompt"],
"session_inputs": [
{
"user_id": "123",
"state": {"a": "1"},
}
],
}
)
mock_eval_dataset_loader.return_value.load.return_value = mock_df.to_dict(
orient="records"
)

mock_agent_engine = mock.Mock()
mock_agent_engine.async_create_session = mock.AsyncMock(
return_value={"id": "session1"}
)
stream_query_return_value = [
{
"id": "1",
"content": {"parts": [{"text": "intermediate1"}]},
"timestamp": 123,
"author": "model",
},
{
"id": "2",
"content": {"parts": [{"text": "agent response"}]},
"timestamp": 124,
"author": "model",
},
]

async def _async_iterator(iterable):
for item in iterable:
yield item

mock_agent_engine.async_stream_query.return_value = _async_iterator(
stream_query_return_value
)
mock_vertexai_client.return_value.agent_engines.get.return_value = (
mock_agent_engine
)

inference_result = self.client.evals.run_inference(
agent="projects/test-project/locations/us-central1/reasoningEngines/123",
src=mock_df,
)

mock_eval_dataset_loader.return_value.load.assert_called_once_with(mock_df)
mock_vertexai_client.return_value.agent_engines.get.assert_called_once_with(
name="projects/test-project/locations/us-central1/reasoningEngines/123"
)
mock_agent_engine.async_create_session.assert_called_once_with(
user_id="123", state={"a": "1"}
)
mock_agent_engine.async_stream_query.assert_called_once_with(
user_id="123", session_id="session1", message="agent prompt"
)

pd.testing.assert_frame_equal(
inference_result.eval_dataset_df,
pd.DataFrame(
{
"prompt": ["agent prompt"],
"session_inputs": [
{
"user_id": "123",
"state": {"a": "1"},
}
],
"intermediate_events": [
[
{
"event_id": "1",
"content": {"parts": [{"text": "intermediate1"}]},
"creation_timestamp": 123,
"author": "model",
}
]
],
"response": ["agent response"],
}
),
)
assert inference_result.candidate_name == "agent"
assert inference_result.gcs_source is None

@mock.patch.object(_evals_utils, "EvalDatasetLoader")
@mock.patch("vertexai._genai._evals_common.vertexai.Client")
def test_run_inference_with_agent_engine_and_session_inputs_literal_string(
self,
mock_vertexai_client,
mock_eval_dataset_loader,
):
session_inputs_str = '{"user_id": "123", "state": {"a": "1"}}'
mock_df = pd.DataFrame(
{
"prompt": ["agent prompt"],
"session_inputs": [session_inputs_str],
}
)
mock_eval_dataset_loader.return_value.load.return_value = mock_df.to_dict(
orient="records"
)

mock_agent_engine = mock.Mock()
mock_agent_engine.async_create_session = mock.AsyncMock(
return_value={"id": "session1"}
)
stream_query_return_value = [
{
"id": "1",
"content": {"parts": [{"text": "intermediate1"}]},
"timestamp": 123,
"author": "model",
},
{
"id": "2",
"content": {"parts": [{"text": "agent response"}]},
"timestamp": 124,
"author": "model",
},
]

async def _async_iterator(iterable):
for item in iterable:
yield item

mock_agent_engine.async_stream_query.return_value = _async_iterator(
stream_query_return_value
)
mock_vertexai_client.return_value.agent_engines.get.return_value = (
mock_agent_engine
)

inference_result = self.client.evals.run_inference(
agent="projects/test-project/locations/us-central1/reasoningEngines/123",
src=mock_df,
)

mock_eval_dataset_loader.return_value.load.assert_called_once_with(mock_df)
mock_vertexai_client.return_value.agent_engines.get.assert_called_once_with(
name="projects/test-project/locations/us-central1/reasoningEngines/123"
)
mock_agent_engine.async_create_session.assert_called_once_with(
user_id="123", state={"a": "1"}
)
mock_agent_engine.async_stream_query.assert_called_once_with(
user_id="123", session_id="session1", message="agent prompt"
)

pd.testing.assert_frame_equal(
inference_result.eval_dataset_df,
pd.DataFrame(
{
"prompt": ["agent prompt"],
"session_inputs": [session_inputs_str],
"intermediate_events": [
[
{
"event_id": "1",
"content": {"parts": [{"text": "intermediate1"}]},
"creation_timestamp": 123,
"author": "model",
}
]
],
"response": ["agent response"],
}
),
)
assert inference_result.candidate_name == "agent"
assert inference_result.gcs_source is None

@mock.patch.object(_evals_utils, "EvalDatasetLoader")
@mock.patch("vertexai._genai._evals_common.vertexai.Client")
def test_run_inference_with_agent_engine_with_response_column_raises_error(
self,
mock_vertexai_client,
mock_eval_dataset_loader,
):
mock_df = pd.DataFrame(
{
"prompt": ["agent prompt"],
"session_inputs": [
{
"user_id": "123",
"state": {"a": "1"},
}
],
"response": ["some response"],
}
)
mock_eval_dataset_loader.return_value.load.return_value = mock_df.to_dict(
orient="records"
)

mock_agent_engine = mock.Mock()
mock_vertexai_client.return_value.agent_engines.get.return_value = (
mock_agent_engine
)

with pytest.raises(ValueError) as excinfo:
self.client.evals.run_inference(
agent="projects/test-project/locations/us-central1/reasoningEngines/123",
src=mock_df,
)
assert (
"The eval dataset provided for agent run should not contain "
"'intermediate_events' or 'response' columns"
) in str(excinfo.value)

def test_run_inference_with_litellm_string_prompt_format(
self,
mock_api_client_fixture,
Expand Down Expand Up @@ -1229,6 +1453,102 @@ def test_run_inference_with_litellm_parsing(
pd.testing.assert_frame_equal(call_kwargs["prompt_dataset"], mock_df)


@pytest.mark.usefixtures("google_auth_mock")
class TestRunAgentInternal:
"""Unit tests for the _run_agent_internal function."""

def setup_method(self):
importlib.reload(vertexai_genai_types)
importlib.reload(_evals_common)

@mock.patch.object(_evals_common, "_run_agent")
def test_run_agent_internal_success(self, mock_run_agent):
mock_run_agent.return_value = [
[
{
"id": "1",
"content": {"parts": [{"text": "intermediate1"}]},
"timestamp": 123,
"author": "model",
},
{
"id": "2",
"content": {"parts": [{"text": "final response"}]},
"timestamp": 124,
"author": "model",
},
]
]
prompt_dataset = pd.DataFrame({"prompt": ["prompt1"]})
mock_agent_engine = mock.Mock()
mock_api_client = mock.Mock()
result_df = _evals_common._run_agent_internal(
api_client=mock_api_client,
agent_engine=mock_agent_engine,
prompt_dataset=prompt_dataset,
)

expected_df = pd.DataFrame(
{
"prompt": ["prompt1"],
"intermediate_events": [
[
{
"event_id": "1",
"content": {"parts": [{"text": "intermediate1"}]},
"creation_timestamp": 123,
"author": "model",
}
]
],
"response": ["final response"],
}
)
pd.testing.assert_frame_equal(result_df, expected_df)

@mock.patch.object(_evals_common, "_run_agent")
def test_run_agent_internal_error_response(self, mock_run_agent):
mock_run_agent.return_value = [{"error": "agent run failed"}]
prompt_dataset = pd.DataFrame({"prompt": ["prompt1"]})
mock_agent_engine = mock.Mock()
mock_api_client = mock.Mock()
result_df = _evals_common._run_agent_internal(
api_client=mock_api_client,
agent_engine=mock_agent_engine,
prompt_dataset=prompt_dataset,
)

assert "response" in result_df.columns
response_content = result_df["response"][0]
assert "Unexpected response type from agent run" in response_content
assert not result_df["intermediate_events"][0]

@mock.patch.object(_evals_common, "_run_agent")
def test_run_agent_internal_malformed_event(self, mock_run_agent):
mock_run_agent.return_value = [
[
{
"id": "1",
"content": {"parts1": [{"text123": "final response"}]},
"timestamp": 124,
"author": "model",
},
]
]
prompt_dataset = pd.DataFrame({"prompt": ["prompt1"]})
mock_agent_engine = mock.Mock()
mock_api_client = mock.Mock()
result_df = _evals_common._run_agent_internal(
api_client=mock_api_client,
agent_engine=mock_agent_engine,
prompt_dataset=prompt_dataset,
)
assert "response" in result_df.columns
response_content = result_df["response"][0]
assert "Failed to parse agent run response" in response_content
assert not result_df["intermediate_events"][0]


class TestMetricPromptBuilder:
"""Unit tests for the MetricPromptBuilder class."""

Expand Down
Loading
Loading