test: give mocked chat completion clients a real payload so the no-choices guard does not trip

tests/llm_translation/conftest.py

@@ -9,6 +9,7 @@
 import importlib
 
 import pytest
+from openai.types.chat import ChatCompletion
 
 
 import litellm  # noqa: E402
@@ -29,6 +30,26 @@
 from tests.fake_openai_endpoint import ensure_fake_openai_endpoint  # noqa: E402
 
 
+@pytest.fixture
+def chat_completion_response() -> ChatCompletion:
+    return ChatCompletion.model_validate(
+        {
+            "id": "chatcmpl-123",
+            "object": "chat.completion",
+            "created": 1677652288,
+            "model": "gpt-4o-mini",
+            "choices": [
+                {
+                    "index": 0,
+                    "message": {"role": "assistant", "content": "Hello there"},
+                    "finish_reason": "stop",
+                }
+            ],
+            "usage": {"prompt_tokens": 9, "completion_tokens": 12, "total_tokens": 21},
+        }
+    )
+
+
 @pytest.fixture(scope="session", autouse=True)
 def fake_openai_endpoint():
     ensure_fake_openai_endpoint()

tests/llm_translation/test_azure_o_series.py

@@ -156,10 +156,13 @@ def test_azure_o_series_routing():
 
 
 @patch("litellm.main.azure_o1_chat_completions._get_openai_client")
-def test_openai_o_series_max_retries_0(mock_get_openai_client):
+def test_openai_o_series_max_retries_0(mock_get_openai_client, chat_completion_response):
     import litellm
 
     litellm.set_verbose = True
+    mock_get_openai_client.return_value.chat.completions.with_raw_response.create.return_value.parse.return_value = (
+        chat_completion_response
+    )
     response = litellm.completion(
         model="azure/o1-preview",
         messages=[{"role": "user", "content": "hi"}],

tests/llm_translation/test_azure_openai.py

@@ -290,7 +290,7 @@ class ResponseFormat(BaseModel):
         # "2024-02-15-preview",
     ],
 )
-def test_azure_gpt_4o_with_tool_call_and_response_format(api_version):
+def test_azure_gpt_4o_with_tool_call_and_response_format(api_version, chat_completion_response):
     from litellm import completion
     from typing import Optional
     from pydantic import BaseModel
@@ -335,6 +335,7 @@ class InvestigationOutput(BaseModel):
     ]
 
     with patch.object(client.chat.completions.with_raw_response, "create") as mock_post:
+        mock_post.return_value.parse.return_value = chat_completion_response
         response = litellm.completion(
             model="azure/gpt-4.1-mini",
             messages=[

tests/llm_translation/test_openai.py

@@ -289,10 +289,13 @@ def test_prompt_caching(self):
 
 
 @patch("litellm.main.openai_chat_completions._get_openai_client")
-def test_openai_max_retries_0(mock_get_openai_client):
+def test_openai_max_retries_0(mock_get_openai_client, chat_completion_response):
     import litellm
 
     litellm.set_verbose = True
+    mock_get_openai_client.return_value.chat.completions.with_raw_response.create.return_value.parse.return_value = (
+        chat_completion_response
+    )
     response = litellm.completion(
         model="gpt-4o-mini",
         messages=[{"role": "user", "content": "hi"}],

tests/local_testing/test_completion.py

@@ -3999,7 +3999,7 @@ def test_completion_novita_ai():
     openai_client = OpenAI(api_key="fake-key")
 
     with patch.object(
-        openai_client.chat.completions, "create", new=MagicMock()
+        openai_client.chat.completions, "create", new=MagicMock(return_value=_openai_mock_response())
     ) as mock_call:
         try:
             completion(