Validation error while using multi-input tools in ReAct Agent - relating to number of inputs

[hepbc]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

@tool
def get_financial_info(co: str, fundamental: str, yr: str) -> float:
    """useful for getting information on a fundamental for a given company and year"""
 <code for returning relevant financial information>

#get_financial_info.invoke({"co":"Tesla", "fundamental":"sales", "yr":"FY22"}) - WORKS OK

tools = [get_financial_info]

template = """Answer the following questions as best you can. You have access to the following tools:\
{tools}\

Use the following format:\

Question: the input question you must answer\
Thought: you should always think about what to do\
Action: the action to take, should be one of [{tool_names}]\
Action Input: the input to the action\
If there's an error, re run the tool with corrected input\
Observation: the result of the action\
... (this Thought/Action/Action Input/Observation can repeat N times)\
Thought: I now know the final answer\
Final Answer: the final answer to the original input question\

Begin!\

Question: {input}\
Thought:{agent_scratchpad}
"""
prompt_template = PromptTemplate.from_template(template)

react_agent = create_react_agent(llm, tools=tools, prompt=prompt_template)
# executes the logical steps we created
react_agent_executor = AgentExecutor(
    agent=react_agent, tools=tools, verbose=True, handle_parsing_errors=True
)

query = "What is the sales for Tesla in FY22?"
react_agent_executor.invoke({"input": query})

> Entering new AgentExecutor chain...
I need to use the get_financial_info tool to get the sales information for Tesla in FY22.
Action: get_financial_info
Action Input: {"co": "Tesla", "fundamental": "sales", "yr": "FY22"}
---------------------------------------------------------------------------
ValidationError                           Traceback (most recent call last)
<ipython-input-36-38d1c873087b> in <cell line: 2>()
      1 query = "What is the sales for Tesla in FY22?"
----> 2 react_agent_executor.invoke({"input": query})

11 frames
/usr/local/lib/python3.10/dist-packages/pydantic/v1/main.py in __init__(__pydantic_self__, **data)
    339         values, fields_set, validation_error = validate_model(__pydantic_self__.__class__, data)
    340         if validation_error:
--> 341             raise validation_error
    342         try:
    343             object_setattr(__pydantic_self__, '__dict__', values)

ValidationError: 2 validation errors for get_financial_infoSchema
fundamental
  field required (type=value_error.missing)
yr
  field required (type=value_error.missing)

Description

Hi folks: I am using a function to get financial information for a company and yr and coupling it with a react agent. The Action Input is generating the correct input: {"co": "Tesla", "fundamental": "sales", "yr": "FY22"} but the agent still gives the Validation error for inputs "fundamental" and "yr".

When I run get_financial_info.invoke({"co":"Tesla"}) it gives the same error which means the other 2 arguments are not getting passed correctly.

Looks like an output parser issue. Can someone please help? Am a newbie to Langchain Agents. Appreciate it.
Thanks

System Info

Google Colab
%pip install openai langchain pypdf faiss-gpu tiktoken
model is GPT-4

[dosubot[bot]]

🤖

Hey @hepbc, great to see you diving into LangChain Agents! Always here to help.

From the error message you provided, it seems like the get_financial_info tool is not receiving the "fundamental" and "yr" inputs correctly. This could be due to the way the ReAct Agent is parsing the output of the language model.

In LangChain, the ReAct Agent uses the ReActSingleInputOutputParser to parse the output of the language model. This parser is designed to handle single input-output pairs. It extracts the action, action input, and observation from the output of the language model. If the output does not contain these elements in the expected format, the parser will not be able to extract them correctly.

Here is the relevant code snippet from the ReActSingleInputOutputParser:

class ReActSingleInputOutputParser(Runnable):
    def run(self, text: str) -> dict:
        lines = text.split("\n")
        action = lines[0].split(": ")[1]
        action_input = json.loads(lines[1].split(": ")[1])
        observation = lines[2].split(": ")[1]
        return {"action": action, "action_input": action_input, "observation": observation}

Source: libs/langchain/langchain/agents/react/parsers.py

From your code, it seems like the action input is being generated correctly: {"co": "Tesla", "fundamental": "sales", "yr": "FY22"}. However, the get_financial_info tool is not receiving these inputs. This suggests that the issue might be with the way the ReAct Agent is passing the inputs to the tool.

One possible solution could be to modify the get_financial_info tool to accept a single dictionary input instead of multiple inputs. This would make it compatible with the ReActSingleInputOutputParser. Here is an example of how you could modify the tool:

@tool
def get_financial_info(inputs: dict) -> float:
    """useful for getting information on a fundamental for a given company and year"""
    co = inputs["co"]
    fundamental = inputs["fundamental"]
    yr = inputs["yr"]
    # <code for returning relevant financial information>

Then, you would need to modify the action input in the ReAct Agent to pass a single dictionary input to the tool:

Action Input: {"inputs": {"co": "Tesla", "fundamental": "sales", "yr": "FY22"}}

Please try this solution and let me know if it resolves the issue. If not, we can explore other potential solutions.

Sources

• libs/langchain/langchain/agents/react/agent.py

About Dosu

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[ccurme]

The agent type you're using doesn't currently support multi-input tools. See more alternatives here: https://python.langchain.com/docs/modules/agents/agent_types/

[hepbc]

Thanks C Curme!