Agents: explore the powerful emerging development of LLM as reasoning agents.

rutamstwt

Rutam Bhagat

Posted on March 25, 2024

Agents: explore the powerful emerging development of LLM as reasoning agents.

Have you ever thought of a large language model (LLMs) as more than just an information storage? What if I told you that LLMs can act as powerful reasoning engines, capable of using new information and their existing knowledge to solve problems, answer complex questions, and even make decisions? This is where LangChain's Agents framework comes in, opening up a whole new world of possibilities for your AI applications.

As a Machine Learning Engineer specializing in LLMs, I'm constantly exploring new ways to push the boundaries of what these models can do. LangChain's Agents have truly captured my imagination, and I believe they hold immense potential for developers and organizations looking to build intelligent and versatile AI systems.

In this blog post, we'll dive into the exciting world of LangChain Agents. I'll explore how they work, how you can equip them with powerful tools, and even how to create your own custom tools to connect them to any data source imaginable.

Built-in Tools: Supercharge Your Agents with Search Engines and Wikipedia

LangChain provides several built-in tools that can significantly enhance your agent's capabilities. Two popular examples are DuckDuckGo search and Wikipedia access. These tools allow your agent to access vast amounts of information on the fly, enabling it to answer questions about recent events, historical figures, and much more.

Let's see how you can utilize these tools in practice. First, you need to set up our environment and import the necessary libraries:



import os
from dotenv import load_dotenv, find_dotenv
_ = load_dotenv(find_dotenv()) # read local .env file

import warnings
warnings.filterwarnings("ignore")

# Account for LLM model deprecation
import datetime
current_date = datetime.datetime.now().date()
target_date = datetime.date(2024, 6, 12)
if current_date > target_date:
    llm_model = "gpt-3.5-turbo"
else:
    llm_model = "gpt-3.5-turbo-0301"

from langchain.agents.agent_toolkits import create_python_agent
from langchain.agents import load_tools, initialize_agent
from langchain.agents import AgentType
from langchain.chat_models import ChatOpenAI

llm = ChatOpenAI(temperature=0, model=llm_model)
tools = load_tools(["llm-math", "wikipedia"], llm=llm)

agent = initialize_agent(
    tools, 
    llm, 
    agent=AgentType.CHAT_ZERO_SHOT_REACT_DESCRIPTION,
    handle_parsing_errors=True,
    verbose=True
)



Enter fullscreen mode Exit fullscreen mode

Now, imagine you want your agent to answer a question about the 2022 World Cup. Since most LLMs are trained on data up to 2021, they wouldn't have this information readily available. However, by leveraging DuckDuckGo search, the agent can access the latest information and provide an accurate answer:



question = "Who won the 2022 World Cup?"
result = agent(question)
print(result)



Enter fullscreen mode Exit fullscreen mode

Image description

The best part? You can easily experiment with different inputs and explore the capabilities of these built-in tools to see what your agent can achieve.

Python Agent and Custom Tools: Connect to Anything

While built-in tools are powerful, the true magic of LangChain Agents lies in the ability to create your own custom tools. This opens up a world of possibilities, allowing you to connect your agent to any data source, API, or function imaginable.

Let's say you want your agent to be able to tell you the current date. You can easily create a custom "time" tool using Python:



from langchain.agents import tool
from datetime import date

@tool
def time(text: str) -> str:
    """Returns todays date, use this for any \
    questions related to knowing todays date. \
    The input should always be an empty string, \
    and this function will always return todays \
    date - any date mathmatics should occur \
    outside this function."""
    return str(date.today())



Enter fullscreen mode Exit fullscreen mode

This simple function, decorated with @tool, becomes a powerful tool for your agent. Now, we can add this tool to our agent's toolkit:



agent = initialize_agent(
    tools + [time], 
    llm, 
    agent=AgentType.CHAT_ZERO_SHOT_REACT_DESCRIPTION,
    handle_parsing_errors=True,
    verbose=True
)



Enter fullscreen mode Exit fullscreen mode

Now, if you ask your agent "What's the date today?", it will recognize the need to use the "time" tool and provide you with the correct answer:



result = agent("What's the date today?") 
print(result)



Enter fullscreen mode Exit fullscreen mode

Image description

This is just the tip of the iceberg. You can create custom tools to access databases, interact with APIs, perform calculations, and much more. The possibilities are truly endless!

Debugging and Troubleshooting: A Glimpse into the Agent's Mind

It's important to remember that LangChain Agents are still under development and may sometimes produce unexpected results. Thankfully, LangChain provides a handy debugging feature. By setting langchain.debug=True, you can view detailed outputs of the agent's chain of thought, allowing you to identify and address any issues that may arise.

For example, let's run the previous query with debug mode enabled:



import langchain
langchain.debug = True

result = agent("What's the date today?") 
print(result)

langchain.debug = False



Enter fullscreen mode Exit fullscreen mode

This will provide you with a detailed breakdown of the agent's reasoning process, including the tools it considered, the actions it took, and the observations it made before arriving at the final answer.

Conclusion:

LangChain Agents represent a significant leap forward in the field of AI. They empower LLMs to go beyond simple information retrieval and become true reasoning engines, capable of interacting with the world in a meaningful way. By leveraging built-in tools, creating custom tools, and utilizing debugging features, you can build intelligent and versatile AI applications that can tackle a wide range of tasks.

💖 💪 🙅 🚩
rutamstwt
Rutam Bhagat

Posted on March 25, 2024

Join Our Newsletter. No Spam, Only the good stuff.

Sign up to receive the latest update from our blog.

Related