如何将工具转换为 OpenAI 函数

本笔记本介绍了如何将 LangChain 工具用作 OpenAI 函数。

%pip install -qU langchain-community langchain-openai

<!--IMPORTS:[{"imported": "MoveFileTool", "source": "langchain_community.tools", "docs": "https://python.langchain.com/api_reference/community/tools/langchain_community.tools.file_management.move.MoveFileTool.html", "title": "How to convert tools to OpenAI Functions"}, {"imported": "HumanMessage", "source": "langchain_core.messages", "docs": "https://python.langchain.com/api_reference/core/messages/langchain_core.messages.human.HumanMessage.html", "title": "How to convert tools to OpenAI Functions"}, {"imported": "convert_to_openai_function", "source": "langchain_core.utils.function_calling", "docs": "https://python.langchain.com/api_reference/core/utils/langchain_core.utils.function_calling.convert_to_openai_function.html", "title": "How to convert tools to OpenAI Functions"}, {"imported": "ChatOpenAI", "source": "langchain_openai", "docs": "https://python.langchain.com/api_reference/openai/chat_models/langchain_openai.chat_models.base.ChatOpenAI.html", "title": "How to convert tools to OpenAI Functions"}]-->
from langchain_community.tools import MoveFileTool
from langchain_core.messages import HumanMessage
from langchain_core.utils.function_calling import convert_to_openai_function
from langchain_openai import ChatOpenAI

model = ChatOpenAI(model="gpt-3.5-turbo")

tools = [MoveFileTool()]
functions = [convert_to_openai_function(t) for t in tools]

functions[0]

{'name': 'move_file',
 'description': 'Move or rename a file from one location to another',
 'parameters': {'type': 'object',
  'properties': {'source_path': {'description': 'Path of the file to move',
    'type': 'string'},
   'destination_path': {'description': 'New path for the moved file',
    'type': 'string'}},
  'required': ['source_path', 'destination_path']}}

message = model.invoke(
    [HumanMessage(content="move file foo to bar")], functions=functions
)

message

AIMessage(content='', additional_kwargs={'function_call': {'arguments': '{\n  "source_path": "foo",\n  "destination_path": "bar"\n}', 'name': 'move_file'}})

message.additional_kwargs["function_call"]

{'name': 'move_file',
 'arguments': '{\n  "source_path": "foo",\n  "destination_path": "bar"\n}'}

使用 OpenAI 聊天模型，我们还可以通过 bind_functions 自动绑定和转换类似函数的对象

model_with_functions = model.bind_functions(tools)
model_with_functions.invoke([HumanMessage(content="move file foo to bar")])

AIMessage(content='', additional_kwargs={'function_call': {'arguments': '{\n  "source_path": "foo",\n  "destination_path": "bar"\n}', 'name': 'move_file'}})

或者我们可以使用更新的 OpenAI API，该 API 使用 tools 和 tool_choice 代替 functions 和 function_call，通过使用 ChatOpenAI.bind_tools：

model_with_tools = model.bind_tools(tools)
model_with_tools.invoke([HumanMessage(content="move file foo to bar")])

AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_btkY3xV71cEVAOHnNa5qwo44', 'function': {'arguments': '{\n  "source_path": "foo",\n  "destination_path": "bar"\n}', 'name': 'move_file'}, 'type': 'function'}]})