Function create_completion

llama_cpp/server/app.py:303–357

(
    request: Request,
    body: CreateCompletionRequest,
)

    tags=[openai_v1_tag],
)
async def create_completion(
    request: Request,
    body: CreateCompletionRequest,
) -> llama_cpp.Completion:
    if isinstance(body.prompt, list):
        assert len(body.prompt) <= 1
        body.prompt = body.prompt[0] if len(body.prompt) > 0 else ""

    body_model = (
        body.model
        if request.url.path != "/v1/engines/copilot-codex/completions"
        else "copilot-codex"
    )

    exclude = {
        "n",
        "best_of",
        "logit_bias_type",
        "user",
        "min_tokens",
    }
    kwargs = body.model_dump(exclude=exclude)

    # handle streaming request
    if kwargs.get("stream", False):
        send_chan, recv_chan = anyio.create_memory_object_stream(10)
        return EventSourceResponse(
            recv_chan,
            data_sender_callable=partial(  # type: ignore
                get_event_publisher,
                request=request,
                inner_send_chan=send_chan,
                body=body,
                body_model=body_model,
                llama_call=llama_cpp.Llama.__call__,
                kwargs=kwargs,
            ),
            sep="\n",
            ping_message_factory=_ping_message_factory,
        )

    # handle regular request
    async with contextlib.asynccontextmanager(get_llama_proxy)() as llama_proxy:
        llama = prepare_request_resources(body, llama_proxy, body_model, kwargs)

        if await request.is_disconnected():
            print(
                f"Disconnected from client (via refresh/close) before llm invoked {request.client}"
            )
            raise HTTPException(
                status_code=status.HTTP_400_BAD_REQUEST,
                detail="Client closed request",
            )

        return await run_in_threadpool(llama, **kwargs)


@router.post(