artificial-intelligence - OpenAI函数调用错误----openai.error.InvalidRequestError :

Question

我正在创建一个聊天机器人，它可以根据用户查询查询数据库中的所有“ View ”。我尝试了很多其他方法但没有成功，所以现在我想我应该尝试OpenAI的函数调用。

我做了什么:我为其中一个 View 创建了一个函数。其中，我调用 GPT3 根据我在参数中提供的用户问题创建 SQL 查询。我已经为模型提供了说明和架构，以便它可以创建正确的查询。下面是该函数。

def get_rent_details(user_query):
    """Get the current weather in a given location"""
    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo-0613",
        prompt="""User will ask you the question regarding their properties, assets and finance. 

        Follow below steps to get correct answer:

        1. Understand the user question and prepare a syntactically correct SQL query to retrieve the correct data.
        2. If you don't find the data in the table, just type "No answer found".
        3. Do not make up any answer by your own.
        4. Instead of '=', always use 'LIKE' statement with 'WHERE' statement.
        5. The user will mention either property name or tenant name. So to make sure the query is correct, use both columns 'TenantName' and 'PropertyName' with 'WHERE' statement. For example: SELECT PropertyCode FROM viewRentRoll WHERE PropertyName LIKE 'Younger, 3003' OR TenantName LIKE 'Younger, 3003'.
        6. DO NOT create any DML query like UPDATE, INSERT, DELETE, ADD.
        7. Below is the table schema to run query on:

        CREATE TABLE [dbo].[viewRentRoll] (            
       [PropertyPKId] [bigint]
      ,[PropertyCode] [nvarchar]
      ,[PropertyName] [nvarchar]
      ,[PropertyList] [nvarchar]
      ,[LeaseCode] [nvarchar]
      ,[TenantName] [nvarchar]
      ,[SnP Rating] [nvarchar]
      ,[Unit Number] [nvarchar]
      ,[Lease Status] [nvarchar]
      ,[Lease Start Date] [datetime]
      ,[Lease Expiration Date] [datetime]
      ,[Unit Square Feet] [bigint]
      ,[Remaining Lease Term] [bigint]
      ,[Currently Monthly Base Rent] [bigint]
      ,[Rent PSF] [bigint]
      ,[ABR] [bigint]
      ,[local tenant] [nvarchar]
      ,[Current Annualized Base Rent PSF] [bigint]
      ,[CreatedLeaseExpirationDate] [datetime]
      ,[TenantCategory] [nvarchar]
  )
    """ + user_query,
        max_tokens=200,
        temperature=0,
    )
    return (response['choices'][0]['text'])

我正在考虑为每个 View 创建这样的函数。之后，我从 OpenAI 函数调用文档中获取了代码，并根据我的需要对其进行了修改。下面是“函数调用”函数:

def run_conversation(user_query):
    # Step 1: send the conversation and available functions to GPT

    print("Running run_conversion............\n\n")

    messages = [{"role": "user", "content": user_query}]
    functions = [
        {
            "name": "get_rent_details",
            "description": "Get the details of rent of tenants or properties",
            "parameters": {
                "type": "object",
                "user_query" : {
                    "type" : "string",
                    "description" : "User's question regarding the rent of Tenant or properties"
                }
            }
        }
    ]
    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo-0613",
        messages=messages,
        functions=functions,
        function_call="auto",  # auto is default, but we'll be explicit
    )
    response_message = response["choices"][0]["message"]

    # Step 2: check if GPT wanted to call a function
    if response_message.get("function_call"):
        # Step 3: call the function
        # Note: the JSON response may not always be valid; be sure to handle errors
        available_functions = {
            "get_rent_details": get_rent_details,
        }  # only one function in this example, but you can have multiple
        function_name = response_message["function_call"]["name"]
        fuction_to_call = available_functions[function_name]
        function_args = json.loads(response_message["function_call"]["arguments"])
        function_response = fuction_to_call(
            user_query=function_args.get("user_query"),
        )

        # Step 4: send the info on the function call and function response to GPT
        messages.append(response_message)  # extend conversation with assistant's reply
        messages.append(
            {
                "role": "function",
                "name": function_name,
                "content": function_response,
            }
        )  # extend conversation with function response
        second_response = openai.ChatCompletion.create(
            model="gpt-3.5-turbo-0613",
            messages=messages,
        )  # get a new response from GPT where it can see the function response
        return second_response

这是我第一次尝试函数调用，所以我不能百分百确定这是否有效。当我运行此代码时，我收到此错误:openai.error.InvalidRequestError: <exception str() failed>对于 response = openai.ChatCompletion.create()在run_conversation(user_query)功能。

任何人都可以指导我哪里犯了错误吗？

我在下面提供完整的代码:

import openai
import json
import os

user_query = "What is the monthly rent of Good Neighbor Homes, Inc."

openai.api_key=os.environ['OPENAI_API_KEY']

def run_conversation(user_query):
    # Step 1: send the conversation and available functions to GPT

    print("Running run_conversion............\n\n")

    messages = [{"role": "user", "content": user_query}]
    functions = [
        {
            "name": "get_rent_details",
            "description": "Get the details of rent of tenants or properties",
            "parameters": {
                "type": "object",
                "user_query" : {
                    "type" : "string",
                    "description" : "User's question regarding the rent of Tenant or properties"
                }
            }
        }
    ]
    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo-0613",
        messages=messages,
        functions=functions,
        function_call="auto",  # auto is default, but we'll be explicit
    )
    response_message = response["choices"][0]["message"]

    # Step 2: check if GPT wanted to call a function
    if response_message.get("function_call"):
        # Step 3: call the function
        # Note: the JSON response may not always be valid; be sure to handle errors
        available_functions = {
            "get_rent_details": get_rent_details,
        }
        function_name = response_message["function_call"]["name"]
        fuction_to_call = available_functions[function_name]
        function_args = json.loads(response_message["function_call"]["arguments"])
        function_response = fuction_to_call(
            user_query=function_args.get("user_query"),
        )

        # Step 4: send the info on the function call and function response to GPT
        messages.append(response_message)  # extend conversation with assistant's reply
        messages.append(
            {
                "role": "function",
                "name": function_name,
                "content": function_response,
            }
        )  # extend conversation with function response
        second_response = openai.ChatCompletion.create(
            model="gpt-3.5-turbo-0613",
            messages=messages,
        )  # get a new response from GPT where it can see the function response
        return second_response


def get_rent_details(user_query):
    """Get the current weather in a given location"""
    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo-0613",
        prompt="""User will ask you the question regarding their properties, assets and finance. 

        Follow below steps to get correct answer:

        1. Understand the user question and prepare a syntactically correct SQL query to retrieve the correct data.
        2. If you don't find the data in the table, just type "No answer found".
        3. Do not make up any answer by your own.
        4. Instead of '=', always use 'LIKE' statement with 'WHERE' statement.
        5. The user will mention either property name or tenant name. So to make sure the query is correct, use both columns 'TenantName' and 'PropertyName' with 'WHERE' statement. For example: SELECT PropertyCode FROM viewRentRoll WHERE PropertyName LIKE 'Younger, 3003' OR TenantName LIKE 'Younger, 3003'.
        6. DO NOT create any DML query like UPDATE, INSERT, DELETE, ADD.
        7. Below is the table schema to run query on:

        CREATE TABLE [dbo].[viewRentRoll] (            
       [PropertyPKId] [bigint]
      ,[PropertyCode] [nvarchar]
      ,[PropertyName] [nvarchar]
      ,[PropertyList] [nvarchar]
      ,[LeaseCode] [nvarchar]
      ,[TenantName] [nvarchar]
      ,[SnP Rating] [nvarchar]
      ,[Unit Number] [nvarchar]
      ,[Lease Status] [nvarchar]
      ,[Lease Start Date] [datetime]
      ,[Lease Expiration Date] [datetime]
      ,[Unit Square Feet] [bigint]
      ,[Remaining Lease Term] [bigint]
      ,[Currently Monthly Base Rent] [bigint]
      ,[Rent PSF] [bigint]
      ,[ABR] [bigint]
      ,[local tenant] [nvarchar]
      ,[Current Annualized Base Rent PSF] [bigint]
      ,[CreatedLeaseExpirationDate] [datetime]
      ,[TenantCategory] [nvarchar]
  )

"""+user_query+"?",
        max_tokens=200,
        temperature=0,
    )
    print(response['choices'][0]['text'])
    return (response['choices'][0]['text'])

run_conversation(user_query)

Answer 1

尝试将函数修改为如下所示:

{
    "name": "get_rent_details",
    "description": "Get the details of rent of tenants or properties",
    "parameters": {
        "type": "object",
        "properties": {
            "user_query": {
                "type": "string",
                "description": "User's question regarding the rent of Tenant or properties"
            }
        },
        "required": ["user_query"]
    }
}

即添加属性和必需。

我面临着类似的问题，对我有用的是从我验证过的官方 Open AI 功能开始，然后我逐行修改它以验证没有任何更改会破坏它。我错过了必需的属性之一。

我还为自己构建了一个验证函数，以检查我将来是否仅将有效函数传递给 Open AI。它并不完美，但已经帮助我发现了一些错误。

def validate_function(function):
    # example func
    """
    function = {
        "name": "get_current_weather",
        "description": "Get the current weather in a given location",
        "parameters": {
            "type": "object",
            "properties": {
                "location": {
                    "type": "string",
                    "description": "The city and state, e.g. San Francisco, CA",
                },
                "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]},
            },
            "required": ["location"],
        },
    }
    """

    # Check the presence of required keys and their types
    assert "name" in function and isinstance(
        function["name"], str
    ), "'name' must be a string."
    assert "description" in function and isinstance(
        function["description"], str
    ), "'description' must be a string."
    assert "parameters" in function and isinstance(
        function["parameters"], dict
    ), "'parameters' must be a dictionary."

    # Check the structure of 'parameters' key
    params = function["parameters"]

    assert (
        "type" in params and params["type"] == "object"
    ), "'type' must be 'object' in parameters."
    assert "properties" in params and isinstance(
        params["properties"], dict
    ), "'properties' must be a dictionary."
    assert "required" in params and isinstance(
        params["required"], list
    ), "'required' must be a list."

    # Check the structure of 'properties' in 'parameters'
    for key, prop in params["properties"].items():
        assert "type" in prop and isinstance(
            prop["type"], str
        ), f"'type' must be a string in properties of {key}."

        if prop["type"] == "array":
            assert (
                "items" in prop
            ), f"'items' must be present in properties of {key} when type is 'array'."

        # Enum check only if it exists
        if "enum" in prop:
            assert isinstance(
                prop["enum"], list
            ), f"'enum' must be a list in properties of {key}."

    # Check 'required' properties are in 'properties'
    for key in params["required"]:
        assert (
            key in params["properties"]
        ), f"'{key}' mentioned in 'required' must exist in 'properties'."