How to pass extra_headers argument to litellm model

[Akshay Bindlish (akshaybndlsh3)]

How can we pass extra_headers argument to litellm model which is mentioned below in _get_kwargs function from settings.yaml or otherwise ?
I'm using graphrag installation from pip.

graphrag/language_model/providers/litellm/chat_model.py

[Juyel (juyelhushen)]

If you're using the pip installation, you shouldn't need to modify chat_model.py directly.

GraphRAG exposes call_args in the model configuration as a dictionary of arguments that are passed to the underlying model provider. So you can try passing extra_headers there:

completion_models:
  default_completion_model:
    model_provider: openai
    model: <your-model>
    auth_method: api_key
    api_key: ${API_KEY}
    call_args:
      extra_headers:
        X-Custom-Header: ${CUSTOM_HEADER}

If you need multiple headers:

call_args:
  extra_headers:
    X-Custom-Header: ${CUSTOM_HEADER}
    X-Another-Header: ${ANOTHER_HEADER}

GraphRAG documents call_args as the place for additional keyword arguments sent with each model request, so this should allow LiteLLM-supported arguments such as extra_headers to be passed without editing the installed package.

If this doesn't work with the specific GraphRAG version you're using, I'd check the installed version first because the configuration format has changed between GraphRAG releases.