fix(convert_dict_to_response): name the non-list choices type in the converter error

When a provider returns choices as null, an object, a string or a number, the
converter said the response had no 'choices' even though the key was present in
the raw keys it listed. A shared message now keeps the old wording for a missing
key and names the offending type otherwise.

The cached-stream regression test also pins the chunk count so a leaked extra
chunk fails it.

litellm/litellm_core_utils/llm_response_utils/convert_dict_to_response.py

@@ -3,7 +3,7 @@ import json
 import re
 import time
 import traceback
-from collections.abc import Sequence
+from collections.abc import Mapping, Sequence
 from typing import Final, Literal, cast
 
 import litellm
@@ -151,6 +151,16 @@ def _clear_later_replay_slice_metadata(choice: StreamingChoices) -> None:
         del choice.enhancements
 
 
+def _invalid_choices_message(response_object: Mapping[str, object]) -> str:
+    raw_keys: Final = list(response_object.keys())
+    if "choices" not in response_object:
+        return f"LiteLLM: provider returned a response with no 'choices'. Raw keys: {raw_keys}"
+    return (
+        f"LiteLLM: provider returned 'choices' that is not a list ({type(response_object['choices']).__name__}). "
+        f"Raw keys: {raw_keys}"
+    )
+
+
 async def convert_to_streaming_response_async(
     response_object: dict | None = None,
 ):
@@ -184,9 +194,7 @@ async def convert_to_streaming_response_async(
 
         raise APIError(
             status_code=500,
-            message=(
-                f"LiteLLM: provider returned a response with no 'choices'. Raw keys: {list(response_object.keys())}"
-            ),
+            message=_invalid_choices_message(response_object),
             llm_provider="",
             model="",
         )
@@ -292,9 +300,7 @@ def convert_to_streaming_response(
 
         raise APIError(
             status_code=500,
-            message=(
-                f"LiteLLM: provider returned a response with no 'choices'. Raw keys: {list(response_object.keys())}"
-            ),
+            message=_invalid_choices_message(response_object),
             llm_provider="",
             model="",
         )
@@ -628,10 +634,7 @@ def convert_to_model_response_object(
 
                 raise APIError(
                     status_code=500,
-                    message=(
-                        "LiteLLM: provider returned a response with no 'choices'. "
-                        f"Raw keys: {list(response_object.keys())}"
-                    ),
+                    message=_invalid_choices_message(response_object),
                     llm_provider="",
                     model="",
                 )

tests/test_litellm/litellm_core_utils/llm_response_utils/test_convert_dict_to_response.py

@@ -167,9 +167,9 @@ def test_convert_missing_choices_raises_api_error() -> None:
     assert "no 'choices'" in str(exc_info.value)
 
 
-@pytest.mark.parametrize("choices", [{}, "", None, 0])
+@pytest.mark.parametrize(("choices", "type_name"), [({}, "dict"), ("", "str"), (None, "NoneType"), (0, "int")])
 @pytest.mark.asyncio
-async def test_convert_non_list_choices_raises_api_error(choices: object) -> None:
+async def test_convert_non_list_choices_raises_api_error(choices: object, type_name: str) -> None:
     from litellm.exceptions import APIError
     from litellm.litellm_core_utils.llm_response_utils.convert_dict_to_response import (
         convert_to_streaming_response,
@@ -183,14 +183,15 @@ async def test_convert_non_list_choices_raises_api_error(choices: object) -> Non
         "object": "chat.completion",
         "choices": choices,
     }
-    with pytest.raises(APIError, match="no 'choices'"):
+    expected: Final = f"'choices' that is not a list \\({type_name}\\)"
+    with pytest.raises(APIError, match=expected):
         convert_to_model_response_object(
             response_object=resp,
             model_response_object=ModelResponse(),
             response_type="completion",
         )
-    with pytest.raises(APIError, match="no 'choices'"):
+    with pytest.raises(APIError, match=expected):
         list(convert_to_streaming_response(response_object=resp))
-    with pytest.raises(APIError, match="no 'choices'"):
+    with pytest.raises(APIError, match=expected):
         async for _ in convert_to_streaming_response_async(response_object=resp):
             pass

tests/test_litellm/litellm_core_utils/test_streaming_handler.py

@@ -2670,6 +2670,7 @@ async def test_cached_response_without_choices_streams_a_single_stop_chunk(
 
     chunks: Final = tuple([chunk async for chunk in wrapper])
 
+    assert len(chunks) == 1
     assert tuple(choice.finish_reason for chunk in chunks for choice in chunk.choices) == ("stop",)
     assert all(choice.delta.content in (None, "") for chunk in chunks for choice in chunk.choices)