OpenAI的Function calling 和 LangChain的Search Agent

OpenAI的Function calling

         openai最近发布的gpt-3.5-turbo-0613 和 gpt-4-0613版本模型增加了function calling的功能，该功能通过定义功能函数，gpt通过分析问题和函数功能描述来决定是否调用函数，并且生成函数对应的入参。函数调用的功能可以弥补gpt的一些缺点，比如实时信息的缺乏、特定领域能力，使得能够进一步利用gpt的逻辑推理能力，可以将问题进行分解处理，解决问题能力更加强大。

gpt的函数调用功能步骤如下：
    1.使用问句和函数定义调用gpt
         2.gpt选择是否调用函数，并输出参数
         3.解析参数 调用函数
         4.将函数返回作为追加信息再次调用gpt

下面是一个通过调用search api的例子

1.定义+描述函数

        下面代码介绍了一个搜索函数，可以通过GoogleSerperAPI实时搜索网络上的信息。

###定义functions，用于描述函数作用和参数介绍。
functions = [
    {
        "name": "get_info_from_web",
        "description": "get more informations from internet use google search",
        "parameters": {
            "type": "object",
            "properties": {
                "query": {
                    "type": "string",
                    "description": "all the questions or information you want search from internet",
                }
            },
            "required": ["query"],
        },
    }
]

###函数定义
def get_info_from_web(query):
    search = GoogleSerperAPIWrapper(serper_api_key="xxxxx")
    return search.run(query)

2.调用gpt，决定是否调用函数以及函数参数

        当用户问句为"今天杭州天气怎么样？"时，gpt做出了进行调用get_info_from_web函数的决定，并且调用的参数为"query": "杭州天气"。

messages = []
messages.append({"role": "system", "content": "Don't make assumptions about what values to plug into functions. Ask for clarification if a user request is ambiguous. "})
messages.append({"role": "user", "content": "今天杭州天气怎么样？"})
chat_response = chat_completion_request(
    messages, functions=functions
)
assistant_message = chat_response.json()["choices"][0]["message"]
messages.append(assistant_message)
print(assistant_message)

>>>
{
	'role': 'assistant',
	'content': None,
	'function_call': {
		'name': 'get_info_from_web',
		'arguments': '{\n  "query": "杭州天气"\n}'
	}
}

3.执行gpt的决定，获得回答问题的中间结果

        调用第2步中gpt输出的参数执行相应的函数，获得中间结果。

assistant_message = chat_response.json()["choices"][0]["message"]
if assistant_message.get("function_call"):
    if assistant_message["function_call"]["name"] == "get_info_from_web":
        query = json.loads(assistant_message["function_call"]["arguments"])["query"]
        results = get_info_from_web(query)
    else:
        results = f"Error: function {assistant_message['function_call']['name']} does not exist"
print(results)

>>>
81°F

4.函数结果和原始问题再次询问gpt，获得最终结果

messages.append({"role": "function", "name": assistant_message["function_call"]["name"], "content": results})
second_response = openai.ChatCompletion.create(
            model= GPT_MODEL,
            messages=messages
        )
print(second_response["choices"][0]["message"]["content"])

>>>
今天杭州的天气是81°F。

LangChain的Search Agent        

        在openai的function calling发布之前，LangChain的Agent就可以实现类似功能。Agent接口是LangChain中一个重要的模块，一些应用程序需要根据用户输入灵活地调用LLM和其他工具。Agent接口为此类应用程序提供了灵活性。Agent可以访问一套工具，并根据用户输入确定要使用哪些工具。Agent可以使用多个工具，并将一个工具的输出用作下一个工具的输入。

        以下是search agent的例子。定义GoogleSerperApi工具作为LLM可用的tool，帮助解决相关问题。

from langchain.utilities import GoogleSerperAPIWrapper
from langchain.llms.openai import OpenAI
from langchain.agents import initialize_agent, Tool
from langchain.agents import AgentType

llm = OpenAI(temperature=0)
search = GoogleSerperAPIWrapper(serper_api_key="xxxxxx")
tools = [
    Tool(
        name="Intermediate Answer",
        func=search.run,
        description="useful for when you need to ask with search",
    )
]

self_ask_with_search = initialize_agent(
    tools, llm, agent=AgentType.SELF_ASK_WITH_SEARCH, verbose=True
)

self_ask_with_search.run(
    "今天杭州天气怎么样？"
)

>>>
> Entering new AgentExecutor chain...
 Yes.
Follow up: 今天是几号？
Intermediate answer: Sunday, July 16, 2023
Follow up: 杭州今天的天气情况？
Intermediate answer: 88°F
So the final answer is: 88°F

> Finished chain.
88°F

       agent功能通过设计prompt实现，search agent的prompt设计如下：

"""Question: Who lived longer, Muhammad Ali or Alan Turing?
Are follow up questions needed here: Yes.
Follow up: How old was Muhammad Ali when he died?
Intermediate answer: Muhammad Ali was 74 years old when he died.
Follow up: How old was Alan Turing when he died?
Intermediate answer: Alan Turing was 41 years old when he died.
So the final answer is: Muhammad Ali

Question: When was the founder of craigslist born?
Are follow up questions needed here: Yes.
Follow up: Who was the founder of craigslist?
Intermediate answer: Craigslist was founded by Craig Newmark.
Follow up: When was Craig Newmark born?
Intermediate answer: Craig Newmark was born on December 6, 1952.
So the final answer is: December 6, 1952

Question: Who was the maternal grandfather of George Washington?
Are follow up questions needed here: Yes.
Follow up: Who was the mother of George Washington?
Intermediate answer: The mother of George Washington was Mary Ball Washington.
Follow up: Who was the father of Mary Ball Washington?
Intermediate answer: The father of Mary Ball Washington was Joseph Ball.
So the final answer is: Joseph Ball

Question: Are both the directors of Jaws and Casino Royale from the same country?
Are follow up questions needed here: Yes.
Follow up: Who is the director of Jaws?
Intermediate answer: The director of Jaws is Steven Spielberg.
Follow up: Where is Steven Spielberg from?
Intermediate answer: The United States.
Follow up: Who is the director of Casino Royale?
Intermediate answer: The director of Casino Royale is Martin Campbell.
Follow up: Where is Martin Campbell from?
Intermediate answer: New Zealand.
So the final answer is: No

Question: {input}
Are followup questions needed here:{agent_scratchpad}"""

 可以从prompt看出，通过四个例子提出了解决问题的方式，即通过follow up + Intermediate answer 分解问题并解决子问题。follow up是gpt的输出，表示需要search tool搜索的问题， Intermediate answer 则为search tool的答案，循环多次之后得到最终答案。