BerriAI/litellm · error · ValueError

safety_settings must be a list of dicts

Error message

safety_settings must be a list of dicts

What it means

Type guard on safety_settings elements: the value is a list but its items are not dicts (the Vertex SDK expects [{'category': ..., 'threshold': ...}] entries), so the malformed setting is refused before the API call.

Source

Thrown at litellm/llms/vertex_ai/vertex_ai_non_gemini.py:167

                project=vertex_project,
                location=vertex_location,
                credentials=cast(Credentials, creds),
            )

        ## Load Config
        config: Final = litellm.VertexAIConfig.get_config()
        for k, v in config.items():
            if k not in optional_params:
                optional_params[k] = v

        ## Process safety settings into format expected by vertex AI
        safety_settings = None
        if "safety_settings" in optional_params:
            safety_settings = optional_params.pop("safety_settings")
            if not isinstance(safety_settings, list):
                raise ValueError("safety_settings must be a list")
            if len(safety_settings) > 0 and not isinstance(safety_settings[0], dict):
                raise ValueError("safety_settings must be a list of dicts")
            safety_settings = [gapic_content_types.SafetySetting(x) for x in safety_settings]

        # vertexai does not use an API key, it looks for credentials.json in the environment

        prompt: Final = " ".join(
            [message.get("content") for message in messages if isinstance(message.get("content", None), str)]
        )

        mode = ""

        request_str = ""
        response_obj: Final = None
        instances = None
        client_options: Final = {"api_endpoint": f"{vertex_location}-aiplatform.googleapis.com"}
        fake_stream = False
        if model in litellm.vertex_language_models or model in litellm.vertex_vision_models:
            llm_model: Any = _vertex_llm_model_object or GenerativeModel(model)
            mode = "vision"

View on GitHub (pinned to 77b7c6c40c)

Solutions

1. Ensure every entry in safety_settings is a dict with 'category' and 'threshold' keys.
2. Remove or convert non-dict entries (strings, None) in the safety_settings list.

When it happens

Trigger: Thrown at litellm/llms/vertex_ai/vertex_ai_non_gemini.py:167 when the library encounters an invalid state.

Common situations: See trigger scenarios.

AI-assisted analysis of BerriAI/litellm@77b7c6c40c (2026-08-18). Data as JSON: /api/errors/5aa1d124305234f8. Report an issue: GitHub.