from typing import Any, List, Sequence, Tuple, Union
from langchain_core._api import deprecated
from langchain_core.agents import AgentAction, AgentFinish
from langchain_core.callbacks import Callbacks
from langchain_core.language_models import BaseLanguageModel
from langchain_core.prompts.base import BasePromptTemplate
from langchain_core.prompts.chat import AIMessagePromptTemplate, ChatPromptTemplate
from langchain_core.runnables import Runnable, RunnablePassthrough
from langchain_core.tools import BaseTool
from langchain.agents.agent import BaseSingleActionAgent
from langchain.agents.format_scratchpad import format_xml
from langchain.agents.output_parsers import XMLAgentOutputParser
from langchain.agents.xml.prompt import agent_instructions
from langchain.chains.llm import LLMChain
from langchain.tools.render import ToolsRenderer, render_text_description
[docs]@deprecated("0.1.0", alternative="create_xml_agent", removal="0.3.0")
class XMLAgent(BaseSingleActionAgent):
"""代理程序使用XML标记。
参数：
工具：代理程序可以选择的工具列表
llm_chain：调用以预测下一步动作的LLMChain
示例：
.. code-block:: python
from langchain.agents import XMLAgent
from langchain
工具 = ...
模型 ="""
tools: List[BaseTool]
"""代理程序可以访问的工具列表。"""
llm_chain: LLMChain
"""用于预测动作的链。"""
@property
def input_keys(self) -> List[str]:
return ["input"]
[docs] @staticmethod
def get_default_prompt() -> ChatPromptTemplate:
base_prompt = ChatPromptTemplate.from_template(agent_instructions)
return base_prompt + AIMessagePromptTemplate.from_template(
"{intermediate_steps}"
)
[docs] @staticmethod
def get_default_output_parser() -> XMLAgentOutputParser:
return XMLAgentOutputParser()
[docs] def plan(
self,
intermediate_steps: List[Tuple[AgentAction, str]],
callbacks: Callbacks = None,
**kwargs: Any,
) -> Union[AgentAction, AgentFinish]:
log = ""
for action, observation in intermediate_steps:
log += (
f"<tool>{action.tool}</tool><tool_input>{action.tool_input}"
f"</tool_input><observation>{observation}</observation>"
)
tools = ""
for tool in self.tools:
tools += f"{tool.name}: {tool.description}\n"
inputs = {
"intermediate_steps": log,
"tools": tools,
"question": kwargs["input"],
"stop": ["</tool_input>", "</final_answer>"],
}
response = self.llm_chain(inputs, callbacks=callbacks)
return response[self.llm_chain.output_key]
[docs] async def aplan(
self,
intermediate_steps: List[Tuple[AgentAction, str]],
callbacks: Callbacks = None,
**kwargs: Any,
) -> Union[AgentAction, AgentFinish]:
log = ""
for action, observation in intermediate_steps:
log += (
f"<tool>{action.tool}</tool><tool_input>{action.tool_input}"
f"</tool_input><observation>{observation}</observation>"
)
tools = ""
for tool in self.tools:
tools += f"{tool.name}: {tool.description}\n"
inputs = {
"intermediate_steps": log,
"tools": tools,
"question": kwargs["input"],
"stop": ["</tool_input>", "</final_answer>"],
}
response = await self.llm_chain.acall(inputs, callbacks=callbacks)
return response[self.llm_chain.output_key]
[docs]def create_xml_agent(
llm: BaseLanguageModel,
tools: Sequence[BaseTool],
prompt: BasePromptTemplate,
tools_renderer: ToolsRenderer = render_text_description,
*,
stop_sequence: Union[bool, List[str]] = True,
) -> Runnable:
"""创建一个使用XML格式化其逻辑的代理。
参数：
llm：LLM，用作代理。
tools：此代理可以访问的工具。
prompt：要使用的提示，必须具有输入键
`tools`：包含每个工具的描述。
`agent_scratchpad`：包含先前代理操作和工具输出。
tools_renderer：控制工具如何转换为字符串，然后传递给LLM。默认为`render_text_description`。
stop_sequence：布尔值或字符串列表。
如果为True，则添加一个停止标记"</tool_input>"以避免产生幻觉。
如果为False，则不添加停止标记。
如果为字符串列表，则使用提供的列表作为停止标记。
默认为True。如果您使用的LLM不支持停止序列，则可以将其设置为False。
返回：
代表代理的可运行序列。它接受与传递的提示相同的所有输入变量。它返回一个AgentAction或AgentFinish。
示例：
.. code-block:: python
from langchain import hub
from langchain_community.chat_models import ChatAnthropic
from langchain.agents import AgentExecutor, create_xml_agent
prompt = hub.pull("hwchase17/xml-agent-convo")
model = ChatAnthropic(model="claude-3-haiku-20240307")
tools = ...
agent = create_xml_agent(model, tools, prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools)
agent_executor.invoke({"input": "hi"})
# 与聊天历史一起使用
from langchain_core.messages import AIMessage, HumanMessage
agent_executor.invoke(
{
"input": "what's my name?",
# 请注意，chat_history是一个字符串
# 因为此提示针对LLMs，而不是聊天模型
"chat_history": "Human: My name is Bob\nAI: Hello Bob!",
}
)
提示：
提示必须具有以下输入键：
* `tools`：包含每个工具的描述。
* `agent_scratchpad`：包含先前代理操作和工具输出的XML字符串。
这里有一个示例：
.. code-block:: python
from langchain_core.prompts import PromptTemplate
template = '''You are a helpful assistant. Help the user answer any questions.
You have access to the following tools:
{tools}
In order to use a tool, you can use <tool></tool> and <tool_input></tool_input> tags. You will then get back a response in the form <observation></observation>
For example, if you have a tool called 'search' that could run a google search, in order to search for the weather in SF you would respond:
<tool>search</tool><tool_input>weather in SF</tool_input>
<observation>64 degrees</observation>
When you are done, respond with a final answer between <final_answer></final_answer>. For example:
<final_answer>The weather in SF is 64 degrees</final_answer>
Begin!
Previous Conversation:
{chat_history}
Question: {input}
{agent_scratchpad}'''
prompt = PromptTemplate.from_template(template)
""" # noqa: E501
missing_vars = {"tools", "agent_scratchpad"}.difference(
prompt.input_variables + list(prompt.partial_variables)
)
if missing_vars:
raise ValueError(f"Prompt missing required variables: {missing_vars}")
prompt = prompt.partial(
tools=tools_renderer(list(tools)),
)
if stop_sequence:
stop = ["</tool_input>"] if stop_sequence is True else stop_sequence
llm_with_stop = llm.bind(stop=stop)
else:
llm_with_stop = llm
agent = (
RunnablePassthrough.assign(
agent_scratchpad=lambda x: format_xml(x["intermediate_steps"]),
)
| prompt
| llm_with_stop
| XMLAgentOutputParser()
)
return agent