大语言模型中的回调函数

By Shaoyang 2024/05/06 Llm

为什么要在大语言模型中使用回调函数

Function Calling的使用是为了克服大语言模型的下述缺点

  • 回答的一致性。在函数调用之前，LLMs 的回复是非结构化且不一致的。如果我们能够更好地控制响应格式，我们就可以更轻松地将响应下游集成到其他系统。（这一点通过调整 prompt 也可以做到）
  • 外部数据。 能够在聊天上下文中使用应用程序其他来源的数据，比如可以让 LLMs 回答诸如“北京现在的天气怎么样的”的问题

架构

LLMs 使用回调函数，本质上是对用户输入 prompt 的自动拆解，利用拆解出来的参数和自动识别出的函数名称，调用外部资源，然后将返回的外部资源添加到用户的 prompt 之中，再次调用 LLMs，从而得到返回结果

image-20240506153033235

实现

Step1 声明回调函数

# 用户输入
messages= [ {"role": "user", "content": "Find me a good course for a beginner student to learn Azure."} ]
messages= [ {"role": "user", "content": "你好，一个高级工程师想学习 Azure，他已经有 10 年的开发经验，可以用哪些资料做参考"} ]

# 这里你可以定义多个函数
# 这里只是函数的声明，大语言模型会按照这里的定义的 schema 去拆分用户的 prompt，并填充进对应的参数中
functions = [
   {
      "name":"search_courses",
      "description":"Retrieves courses from the search index based on the parameters provided",
      "parameters":{
         "type":"object",
         "properties":{
            "role":{
               "type":"string",
               "description":"The role of the learner (i.e. developer, data scientist, student, etc.)"
            },
            "product":{
               "type":"string",
               "description":"The product that the lesson is covering (i.e. Azure, Power BI, etc.)"
            },
            "level":{
               "type":"string",
               "description":"The level of experience the learner has prior to taking the course (i.e. beginner, intermediate, advanced)"
            }
         },
         "required":[
            "role"
         ]
      }
   }
]


response = client.chat.completions.create(model="gpt-3.5-turbo",
                                        messages=messages,
                                        functions=functions,
                                        function_call="auto")

Step2 定义回调函数的具体实现

# 这里是函数的具体实现
def search_courses(role, product, level="beginner"):
  url = "https://learn.microsoft.com/api/catalog/"
  params = {
     "role": role,
     "product": product,
     "level": level
  }
  response = requests.get(url, params=params)
  modules = response.json()["modules"]
  results = []
  for module in modules[:5]:
     title = module["title"]
     url = module["url"]
     results.append({"title": title, "url": url})
  return str(results)

# Check if the model wants to call a function
if getattr(getattr(response_message, 'function_call', None), 'name', None):

    if response_message.function_call.name:
        print("Recommended Function call:")
        print(response_message.function_call.name)
        print()

        # Call the function.
        function_name = response_message.function_call.name

        available_functions = {
                "search_courses": search_courses,
        }
        # 文本和具体函数名称的映射
        function_to_call = available_functions[function_name]

        function_args = json.loads(response_message.function_call.arguments)
        function_response = function_to_call(**function_args)

        print("Output of function call:")
        print(function_response)
        print(type(function_response))


        # Add the assistant response and function response to the messages
        messages.append( # adding assistant response to messages
            {
                "role": response_message.role,
                "function_call": {
                    "name": function_name,
                    "arguments": response_message.function_call.arguments,
                },
                "content": None
            }
        )
        messages.append( # adding function response to messages
            {
                "role": "function",
                "name": function_name,
                "content":function_response,
            }
        )

print("Messages in next request:")
print(messages)
print()

Step3 再次调用大语言模型，返回标准化的输出

second_response = client.chat.completions.create(
   messages=messages,
   model="gpt-3.5-turbo",
   function_call="auto",
   functions=functions,
   temperature=0
      )  # get a new response from GPT where it can see the function response


print(second_response.choices[0].message)

结果示例

对于一个有10年开发经验且级别为高级的学生来说，以下资料可以作为参考：

1. [Advanced PyBryt for Classroom Auto-assessment](https://learn.microsoft.com/en-us/training/modules/advanced-pybryt/?WT.mc_id=api_CatalogApi)
2. [Enrich your data with Azure AI Language](https://learn.microsoft.com/en-us/training/modules/enrich-search-index-using-language-studio/?WT.mc_id=api_CatalogApi)
3. [Implement Azure App Service on Kubernetes with Arc](https://learn.microsoft.com/en-us/training/modules/configure-application-kubernetes-arc/?WT.mc_id=api_CatalogApi)
4. [Implement Datacenter Firewall and Software Load Balancer on Azure Stack HCI](https://learn.microsoft.com/en-us/training/modules/implement-firewall-load-balancer/?WT.mc_id=api_CatalogApi)

这些资料将帮助学生深入学习 Azure 相关的知识。

其他

微软提供的 Learn API 还是挺不错的，作为 GET 接口的测试可以以备不时之需

该 API 是基于 REST 的 Web API，返回 JSON 编码的响应。 若要请求完整目录，请将 GET 请求发送到：https://learn.microsoft.com/api/catalog/，限制的参数格式：{“level”: “advanced”,“product”: “Azure”,“role”: “student”}

.