zylon-ai/private-gpt · error · ValueError

INVALID_REQUEST_ERROR

INVALID_REQUEST_ERROR

Error message

Configured model does not support function calling

What it means

Identical guard in the synchronous ChatEngine.initialize_run: llm = self._llm_component.get_llm(request.system.model) must be a FunctionCallingLLM or the run is aborted with ValueError before the context-stack requirement and sampling-params validation are checked. Tagged INVALID_REQUEST_ERROR, so it is treated as caller error (bad model choice in the request), not an internal fault.

Source

Thrown at private_gpt/components/engines/chat/chat_engine.py:869

                    thinking="",
                    signature=f"sig_{uuid4().hex}",
                ),
            )
        stream_delta_state.active_block.index = run.block_count
        run.block_count += 1
        stream_delta_state.active_block_kind = target_kind
        handler.emit(stream_delta_state.active_block)

    def initialize_run(
        self,
        request: ChatRequest,
        context_stack: ContextStack | None = None,
        hooks: list[ToolExecutionHook] | None = None,
    ) -> _LoopRun:
        """Build initial llm and state for one run."""
        llm = self._llm_component.get_llm(request.system.model)
        if not isinstance(llm, FunctionCallingLLM):
            raise ValueError("Configured model does not support function calling")

        if not isinstance(request, ResolvedChatRequest) and context_stack is None:
            raise ValueError("Configured context stack is required")

        llm_kwargs = ChatLLMParameters.model_validate(request.sampling_params)
        if request.thinking.enabled and request.thinking.type:
            llm_kwargs = llm_kwargs.model_copy(
                update={
                    "reasoning_effort": ReasoningEffort.from_str(request.thinking.type)
                }
            )
        if request.response_format and request.response_format.output_cls:
            structured = StructuredOutputsParams.from_optional(
                output_cls=request.response_format.output_cls,
            )
            if structured is not None:
                llm_kwargs = llm_kwargs.model_copy(
                    update={"structured_outputs": structured}

View on GitHub (pinned to 4a030776a3)

Solutions

1. Set system.model to a function-calling-capable model in the request
2. Make custom LLM classes subclass FunctionCallingLLM
3. Validate the resolved LLM type in a pre-flight check before submitting the request
4. Use a FunctionCallingLLM-based mock in tests

Example fix

# before
request.system.model = 'my-completion-model'  # ValueError

# after
request.system.model = 'my-tool-model'  # resolves to FunctionCallingLLM

llm = llm_component.get_llm(request.system.model)
if not isinstance(llm, FunctionCallingLLM):
    raise InvalidRequestError(f'model {request.system.model!r} lacks function calling')

from llama_index.core.llms import FunctionCallingLLM

def is_function_calling_llm(llm) -> TypeGuard[FunctionCallingLLM]:
    return isinstance(llm, FunctionCallingLLM)

try:
    run = engine.initialize_run(request, context_stack)
except ValueError as e:
    if 'function calling' in str(e):
        return JSONResponse({'error': {'code': 'INVALID_REQUEST_ERROR', 'message': str(e)}}, 400)
    raise

When it happens

Trigger: Calling initialize_run / streaming chat APIs with request.system.model pointing at a non-function-calling LLM — completion-only models, generic wrappers, or mocks that don't subclass FunctionCallingLLM.

Common situations: Requesting a small/cheap model without tool support while the engine's features (tools, structured output via response_format) require function calling; custom LLM integrations not inheriting FunctionCallingLLM; model id resolving to an unexpected LLM class after provider config changes.

Related errors

• Configured model does not support function calling
• Expected tool calls in response but found none. Message: {re
• OVERLOADED_CONDENSATION_ERROR
• Audio blocks found but no audio-capable LLM provided.
• Failed to describe audio in the message.

AI-assisted analysis of zylon-ai/private-gpt@4a030776a3 (2026-08-15). Data as JSON: /api/errors/54525af4a8357f8e. Report an issue: GitHub.