함수 호출과 API 통합

외부 도구 호출

LLM의 역할: 호출 제안

도구 정의

날씨 도구 정의

def get_current_weather(location: str, unit: str = "celsius"):
    """
    Get the current weather in a given location.
    Args:
        location (str): The city and state, e.g. "London, UK"
        unit (str): The unit of temperature, "celsius" or "fahrenheit"
    Returns:
        dict: Weather information.
    """
    # In a real app, this would call a weather API
    if location == "London, UK":
        return {"temperature": "15", "unit": unit, "forecast": "cloudy"}
    elif location == "New York, USA":
        return {"temperature": "20", "unit": unit, "forecast": "sunny"}
    else:
        return {"temperature": "N/A", "unit": unit, "forecast": "unknown"}

도구를 위한 프롬프트 작성

LLM의 호출 제안

{
  "tool_calls": [
    {
      "id": "call_abc123",
      "function": {
        "name": "get_current_weather",
        "arguments": "{\"location\": \"London, UK\", \"unit\": \"celsius\"}"
      },
      "type": "function"
    }
  ]
}

앱에서 도구 실행

LLM의 도구 출력 처리

처음부터 끝까지의 상호작용

def get_current_weather(location, unit="celsius"):
    if location == "London, UK":
        return {"temp": "15", "unit": unit, "forecast": "cloudy"}
    return {"temp": "N/A", "unit": unit, "forecast": "unknown"}

def simulate_llm_response(user_query):
    # LLM 'thinks' and decides to call a tool
    if "weather in London" in user_query:
        return {
            "type": "tool_call",
            "function": "get_current_weather",
            "args": {"location": "London, UK"}
        }
    return {"type": "text", "content": "I cannot help with that."}

def simulate_llm_final_response(tool_output):
    # LLM processes tool output and gives natural language
    temp = tool_output.get("temp", "N/A")
    forecast = tool_output.get("forecast", "unknown")
    return f"The weather in London is {temp} degrees and {forecast}."

def main():
    user_input = "What is the weather in London, UK?"
    print(f"User: {user_input}")

    llm_output = simulate_llm_response(user_input)

    if llm_output["type"] == "tool_call":
        func_name = llm_output["function"]
        func_args = llm_output["args"]
        print(f"LLM suggested calling: {func_name}({func_args})")

        # Execute the function
        tool_result = get_current_weather(**func_args)
        print(f"Tool output: {tool_result}")

        # Feed result back to LLM
        final_answer = simulate_llm_final_response(tool_result)
        print(f"LLM's final answer: {final_answer}")
    else:
        print(f"LLM: {llm_output['content']}")

if __name__ == "__main__":
    main()

함수 호출 확인

복습: 도구의 실제 활용