## What happened

I called `litellm.completion()` (litellm 1.83.7, sync, non-streaming) against an OpenAI-compatible internal gateway (`api_base` set). The gateway rejected the request with its own error response form:

```json {"response": "Token is invalid [2]", "status": 400} ```

instead of the typical OpenAI-shaped one, like:

```json { "error": { "message": "Incorrect API key provided: sk-proj-********. You can find your API key at https://platform.openai.com/account/api-keys.", "type": "invalid_request_error", "param": null, "code": "invalid_api_key" } } ```

i.e. no top-level `error` object, `choices` absent/None. While that isn't great on the part of the server, litellm did still parse it as a _successful_ completion payload and then failed like this:

``` litellm.InternalServerError: InternalServerError: OpenAIException - Invalid response object Traceback (most recent call last): File ".../litellm/litellm_core_utils/llm_response_utils/convert_dict_to_response.py", line 539, in convert_to_model_response_object assert response_object["choices"] is not None and isinstance( ... AssertionError

received_args={'response_object': {'id': None…