How to add reasoning_content with CustomLLM?

[KyleZhang0536]

Hi team,
We have an upstream llm api, it's incoming looks like

{
    "model": "xxxx",
    "query": "hi",
    "history": [],
    "modelAdvance": {
        "topP":0.3,
        "temperature":0.6,
        "maxTokens":1000
    },
    "stream": true,
    "clientSid": "123",
    "deviceId": "123",
    "sceneType": "chat"
}

and output chunk

{
    "id": "d4c228debd99426eb451a48caf1fb4ac",
    "model": "xxxx",
    "created": 1741867320827,
    "choices": [
        {
            "index": 0,
            "delta": {
                "role": "assistant",
                "content": "",
                "reasonContent": "Okay",
                "contentSource": "llm"
            }
        }
    ],
    "usage": {
        "completion_tokens": 0,
        "prompt_tokens": 0,
        "total_tokens": 0
    }
}

So we want to use CustomLLM and trans it into openai compatible format.
According to this doc, my CustomLLM return Iterator[GenericStreamingChunk] as response, but GenericStreamingChunk don't have reasoning_content.
And the response will be processed here

litellm/litellm/litellm_core_utils/streaming_handler.py

Line 1015 in 122ee63

anthropic_response_obj: GChunk = cast(GChunk, chunk)

How to add reasoning_content with CustomLLM?