agent tool怎么传递复杂参数

提问区
1

问题

from langchain.agents import initialize_agent, Tool
from langchain.agents import AgentType
from langchain.agents import tool
from langchain.memory import ConversationBufferMemory
from langchain_community.llms.ollama import Ollama
from langchain.pydantic_v1 import BaseModel, Field

memory = ConversationBufferMemory()
llm = Ollama(model="llama2")

@tool
def get_word_length(word: str) -> int:
    """Returns the length of a word."""
    return len(word)

@tool
def sum_two_number(a: int, b: int) -> int:
    """Returns the sum of two numbers."""
    print(f"输入的参数是{a},{b}")
    return a+b

class ReleaseInfo(BaseModel):
    release_date: str = Field(description="should be the date of the release happen, should be formatted as YYYY-MM-DD")
    release_content: str = Field(description="should be the content of the release")
    qa: list = Field(description="should be the qa of the release")
    be: list = Field(description="should be the backend developer of the release")
    fe: list = Field(description="should be the frontend developer of the release")
    services: list = Field(description="should be the services of the release")

@tool("release_notification", args_schema=ReleaseInfo, return_direct=True)
def release_notification(release_info: ReleaseInfo):
    """Call the api to send a notification of the release info"""
    print(f"notify: {release_info}")


tools = [
    Tool.from_function(
        name="GET_WORD_LENGTH",
        func=get_word_length,
        description="this tool require you to provide a word, and return the length of the word. "
    ),
    Tool.from_function(
        name="SUM_TWO_NUMBERS",
        func=sum_two_number,
        description="this tool require you to provide two number, and return the sum of the two number. "
    ),
    Tool.from_function(
        name="RELEASE_NOTIFICATION",
        func=release_notification,
        description="""this tool require you to provide release info and call the api to send a notification
            release info could be like this:
            ```
            3/14 Release
            Remove Market consent
            QA：@Mandy x(QA UTC+8) 
            FE：@Sans X+8 
            BE：@robin w 
            Services：accounts-ui、pwa-ui、could-account
            ```
            then we can know the release_date is 2024-03-14, release content is Remove Market consent,
            qa is [Mandy x(QA UTC+8)] , fe is [Sans X+8] , be is [robin w], services are [accounts-ui,pwa-ui,could-account]
         """
    ),
]

SUFFIX = """Provide the final answer and terminate the chain of thought once you have an answer. Begin! 

Question: {input}
Thought:{agent_scratchpad}"""

agent = initialize_agent(
    tools, llm, agent=AgentType.CONVERSATIONAL_REACT_DESCRIPTION, verbose=True,
    handle_parsing_errors=True, max_iterations=5,memory=memory, agent_kwargs={'suffix':SUFFIX})

if __name__ == '__main__':
    # agent.run("What is the inventory of sneakers?")
    while True:
        human_input = input("Human: ")
        response = agent.invoke(human_input)
        print(f"Assistant: {response}")

当我传入input：
3/13 Release
FIX some bug
QA: @zeven
BE: @ted w
BE Services: capital，capital-job

报错信息

/Users/user/anaconda3/envs/AI/bin/python -X pycache_prefix=/Users/user/Library/Caches/JetBrains/PyCharm2023.3/cpython-cache /Applications/PyCharm.app/Contents/plugins/python/helpers/pydev/pydevd.py --multiprocess --qt-support=auto --client 127.0.0.1 --port 59257 --file /Users/user/Documents/Learn/AI/demo/main2.py
Connected to pydev debugger (build 233.14475.56)
/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain_core/_api/deprecation.py:117: LangChainDeprecationWarning: The function initialize_agent was deprecated in LangChain 0.1.0 and will be removed in 0.2.0. Use Use new agent constructor methods like create_react_agent, create_json_agent, create_structured_chat_agent, etc. instead.
warn_deprecated(
Human: 3/13 Release
Capital add BCH network
QA: @zeven 【utc+8】
BE: @ted w
BE Services: capital，capital-job

Entering new AgentExecutor chain…
Thought: Do I need to use a tool? Yes
Action: RELEASE_NOTIFICATION
Action Input: release info as follows:

3/13 Release
Remove Market consent
QA：@Mandy x(QA UTC+8) 
FE：@Sans X+8 
BE：@robin w 
Services：accounts-ui、pwa-ui、could-account
```python-BaseException
Traceback (most recent call last):
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain/chains/base.py", line 163, in invoke
    raise e
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain/chains/base.py", line 153, in invoke
    self._call(inputs, run_manager=run_manager)
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain/agents/agent.py", line 1432, in _call
    next_step_output = self._take_next_step(
                       ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain/agents/agent.py", line 1138, in _take_next_step
    [
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain/agents/agent.py", line 1138, in <listcomp>
    [
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain/agents/agent.py", line 1223, in _iter_next_step
    yield self._perform_agent_action(
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain/agents/agent.py", line 1245, in _perform_agent_action
    observation = tool.run(
                  ^^^^^^^^^
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain_core/tools.py", line 382, in run
    raise e
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain_core/tools.py", line 376, in run
    self._run(*tool_args, run_manager=run_manager, **tool_kwargs)
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain_core/tools.py", line 574, in _run
    self.func(
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain_core/tools.py", line 515, in __call__
    return self.run(tool_input, callbacks=callbacks)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain_core/tools.py", line 382, in run
    raise e
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain_core/tools.py", line 373, in run
    parsed_input = self._parse_input(tool_input)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/langchain_core/tools.py", line 276, in _parse_input
    input_args.validate({key_: tool_input})
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/pydantic/v1/main.py", line 711, in validate
    return cls(**value)
           ^^^^^^^^^^^^
  File "/Users/user/anaconda3/envs/AI/lib/python3.11/site-packages/pydantic/v1/main.py", line 341, in __init__
    raise validation_error
pydantic.v1.error_wrappers.ValidationError: 5 validation errors for ReleaseInfo
release_content
  field required (type=value_error.missing)
qa
  field required (type=value_error.missing)
be
  field required (type=value_error.missing)
fe
  field required (type=value_error.missing)
services
  field required (type=value_error.missing)

Process finished with exit code 1


### 环境
本地运行了Ollama，模型选择的是llama2:7b
2

使用这个方法，这个agent支持多输入参数。

3

我修改了一下代码，用大佬说的这个函数，但效果也不太对

from langchain.agents import initialize_agent, Tool, create_structured_chat_agent, AgentExecutor
from langchain.agents import AgentType
from langchain.agents import tool
from langchain.memory import ConversationBufferMemory
from langchain_community.llms.ollama import Ollama
from langchain.pydantic_v1 import BaseModel, Field
from langchain import hub
from langchain_core.utils.function_calling import format_tool_to_openai_function

prompt = hub.pull("hwchase17/structured-chat-agent")



memory = ConversationBufferMemory()
llm = Ollama(model="llama2")

@tool
def get_word_length(word: str) -> int:
    """Returns the length of a word."""
    return len(word)

@tool
def sum_two_number(a: int, b: int) -> int:
    """Returns the sum of two numbers."""
    print(f"输入的参数是{a},{b}")
    return a+b

class ReleaseInfo(BaseModel):
    release_date: str = Field(description="should be the date of the release happen, should be formatted as YYYY-MM-DD")
    release_content: str = Field(description="should be the content of the release")
    qa: list = Field(description="should be the qa of the release")
    be: list = Field(description="should be the backend developer of the release")
    fe: list = Field(description="should be the frontend developer of the release")
    services: list = Field(description="should be the services of the release")

@tool("release_notification", args_schema=ReleaseInfo, return_direct=True)
def release_notification(release_info: ReleaseInfo):
    """this tool require you to provide release info and call the api to send a notification
            release info could be like this:
            ```
            3/14 Release
            Remove Market consent
            QA：@Mandy x(QA UTC+8)
            FE：@Sans X+8
            BE：@robin w
            Services：accounts-ui、pwa-ui、could-account
            ```
            then we can know the release_date is 2024-03-14, release content is Remove Market consent,
            qa is [Mandy x(QA UTC+8)] , fe is [Sans X+8] , be is [robin w], services are [accounts-ui,pwa-ui,could-account]
         """
    print(f"notify: {release_info}")
    return {"result": release_info}


tools = [sum_two_number,release_notification]


agent = create_structured_chat_agent(
    tools=tools, llm=llm, prompt=prompt)

agent_executor = AgentExecutor(
    agent=agent, tools=tools, verbose=True, handle_parsing_errors=True
)

if __name__ == '__main__':
    # agent.run("What is the inventory of sneakers?")
    while True:
        human_input = input("Human: ")
        response = agent_executor.invoke({"input": human_input})
        print(f"Assistant: {response}")

报错信息：

Human: 3/13 Release
Capital add BCH network
QA: @zeven 【utc+8】 
BE: @ted w 
BE Services: capital，capital-job

> Entering new AgentExecutor chain...
Could not parse LLM output: Action: sum_two_number
Action Input: { 'a': 5, 'b': 7 }

Invalid or incomplete responseCould not parse LLM output: Action: release_notification
Action Input: {
"release_date": "2024-03-14",
"release_content": "Remove Market consent",
"qa": ["Mandy x(QA UTC+8)", "Sans X+8"],
"be": ["robin w"],
"fe": [],
"services": ["accounts-ui", "pwa-ui", "could-account"]
}Invalid or incomplete responseCould not parse LLM output: Action: sum_two_number
Action Input: { 'a': 5, 'b': 7 }
4

我看参数传递也是对的。是不是其他逻辑导致的。你多加点log分析下吧