react agent can't invoke custom tools

[olevanss]

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

import os
from langchain.tools import tool
from langchain.agents import AgentExecutor, create_react_agent
from langchain_openai import ChatOpenAI
from dotenv import load_dotenv
from langchain import hub
import httpx
from langchain.pydantic_v1 import BaseModel, Field
import numpy as np

load_dotenv()
prompt = hub.pull("hwchase17/react")

class Metrics(BaseModel):
    num1: float = Field(description="Value 1")
    num2: float = Field(description="Value 2")


@tool(args_schema=Metrics)
def perc_diff(num1: float, num2: float) -> float:
    """Calculates the percentage difference between two numbers"""
    return (np.abs(num1 - num2) / ((num1 + num2) / 2)) * 100

tools = [perc_diff]


llm = ChatOpenAI(model_name="gpt-4o-mini")

agent = create_react_agent(llm, tools, prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)

agent_executor.invoke({"input": "Calculate the percentage difference between 5 and 7"})

Description

The problem

I am trying to build react agent with custom tools. I encounter problems every time I try to use tool with more than 1 input parameter.
Agent tries to provide input for custom tool, but it is parsed incorrectly (as a single string object, as far as I tested).
I would appreciate a piece of advice here.

The exact error I get is the following:

pydantic.v1.error_wrappers.ValidationError: 2 validation errors for Metrics
To calculate the percentage difference between 5 and 7, I will use the perc_diff function.
Action: perc_diff
Action Input: (5, 7)num1
value is not a valid float (type=type_error.float)
num2
field required (type=value_error.missing)

System Info

System Information

OS: Darwin
OS Version: Darwin Kernel Version 23.5.0: Wed May 1 20:12:58 PDT 2024; root:xnu-10063.121.3~5/RELEASE_ARM64_T6000
Python Version: 3.11.8 | packaged by conda-forge | (main, Feb 16 2024, 20:49:36) [Clang 16.0.6 ]

Package Information

langchain_core: 0.2.27
langchain: 0.2.12
langchain_community: 0.0.28
langsmith: 0.1.95
langchain_cli: Installed. No version info available.
langchain_experimental: 0.0.52
langchain_openai: 0.0.8
langchain_text_splitters: 0.2.2
langchainhub: 0.1.15
langserve: Installed. No version info available.

Packages not installed (Not Necessarily a Problem)

The following packages were not found:

langgraph

[dosubot[bot]]

Hello there, @olevanss! I'm here to help with any bugs, questions, or contributions you might have. Let's squash those bugs together! 😄

To resolve the issue of incorrect parsing of input parameters when using a custom tool with more than one input parameter in a ReAct agent with LangChain, you need to ensure that the prompt template correctly handles multiple input parameters for the tools. Here's how you can modify your code to handle this:

1. Modify the Prompt Template:
   Ensure that the Action Input is specified to be in the format of a dictionary with keys as parameter names.

   from langchain_core.prompts import PromptTemplate
   
   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, should be in the format of a dictionary with keys as parameter names
   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 = PromptTemplate.from_template(template)

2. Ensure the Tool Handles Dictionary Input:
   Define your tool to handle the dictionary format for input parameters.

   from langchain_core.tools import BaseTool
   
   class CustomTool(BaseTool):
       name = "perc_diff"
       description = "Calculates the percentage difference between two numbers."
   
       def run(self, input_params: dict):
           num1 = input_params.get("num1")
           num2 = input_params.get("num2")
           return (np.abs(num1 - num2) / ((num1 + num2) / 2)) * 100
   
   tools = [CustomTool()]

3. Update the Agent and Agent Executor:
   Create the agent and agent executor with the updated prompt and tools.

   from langchain.agents import AgentExecutor, create_react_agent
   from langchain_openai import ChatOpenAI
   from dotenv import load_dotenv
   import numpy as np
   
   load_dotenv()
   
   llm = ChatOpenAI(model_name="gpt-4o-mini")
   
   agent = create_react_agent(llm, tools, prompt)
   agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)
   
   agent_executor.invoke({"input": "Calculate the percentage difference between 5 and 7"})

By ensuring that the prompt template, tools, and agent executor are correctly set up to handle multiple input parameters, you can resolve the issue of incorrect parsing of input parameters in a ReAct agent with LangChain ^[1][2].

To continue talking to Dosu, mention @dosu.

[olevanss]

UPD:
I run a little through this code with debugger and found out that the mistake rises due to the BaseTool class passing the input as a str instead of dict, finding a way to solve this

UPD2:
RunnableAgent parses input as a string despite it being dict

[itssidhere]

any solution for this? The tool calling agent works fine but the react agent cannot call custom tools in javascript