How to Build AI Agents Using “Tool Use”?

Introduction

Earlier than speaking about AI Brokers, It’s crucial to grasp the lifespan of a complicated language mannequin like GPT. A big language mannequin comparable to GPT begins its lifespan with pretraining when it learns from an enormous corpus of textual knowledge to determine a fundamental grasp of the language. The subsequent step is supervised fine-tuning when the mannequin is improved for particular duties by utilizing specified datasets to refine it. Through the use of optimistic reinforcement to optimize the mannequin’s habits, reward modeling enhances efficiency usually and decision-making particularly. Lastly, the mannequin could be taught and alter dynamically by way of interactions due to reinforcement studying, honing its expertise to do varied duties extra precisely and adaptable. On this article, we can even be taught how one can construct AI Brokers utilizing “Instrument Use.”

Overview

• Language fashions like GPT are developed by way of pretraining, supervised fine-tuning, reward modeling, and reinforcement studying.
• Every section entails particular datasets, algorithms, mannequin changes, and evaluations to boost the mannequin’s capabilities.
• Static fashions battle with offering real-time data, requiring common fine-tuning, which is resource-intensive and infrequently impractical.
• Construct AI Brokers Utilizing “Instrument Use” in Agentic Workflow.
• AI brokers with entry to exterior instruments can collect real-time knowledge, execute duties, and preserve context, enhancing accuracy and responsiveness.

GPT Assistant Coaching Pipeline

Every section of the mannequin’s growth—pretraining, supervised fine-tuning, reward modeling, and reinforcement studying—progresses by way of 4 important elements: Dataset, Algorithm, Mannequin, and Analysis.

Pretraining Part

Within the preliminary pretraining section, the mannequin ingests huge portions of uncooked web knowledge, totaling trillions of phrases. Whereas the information’s high quality could fluctuate, its sheer quantity is substantial however nonetheless falls wanting satisfying the mannequin’s starvation for extra. This section calls for vital {hardware} assets, together with GPUs, and months of intensive coaching. The method begins with initializing weights from scratch and updating them as studying progresses. Algorithms like language modeling predict the following token, forming the premise of the mannequin’s early levels.

Supervised Superb-Tuning Part

Transferring to supervised fine-tuning, the main focus shifts to task-specific labeled datasets the place the mannequin refines its parameters to foretell correct labels for every enter. Right here, the datasets’ high quality is paramount, resulting in a discount in amount. Algorithms tailor coaching for duties comparable to token prediction, culminating in a Supervised Superb-Tuning (SFT) Mannequin. This section requires fewer GPUs and fewer time than pretraining because of enhanced dataset high quality.

Reward Modeling Part

Reward modeling follows, using algorithms like binary classification to boost mannequin efficiency primarily based on optimistic reinforcement indicators. The ensuing Reward Modeling (RM) Mannequin undergoes additional enhancement by way of human suggestions or analysis.

Reinforcement Studying Part

Reinforcement studying optimizes the mannequin’s responses by way of iterative interactions with its atmosphere, making certain adaptability to new data and prompts. Nonetheless, integrating real-world knowledge to maintain the mannequin up to date stays a problem.

The Problem of Actual-Time Information

Addressing this problem entails bridging the hole between educated knowledge and real-world data. It necessitates methods to repeatedly replace and combine new knowledge into the mannequin’s data base, making certain it might probably reply precisely to the newest queries and prompts.

Nonetheless, a important query arises: Whereas we’ve educated our LLM on the information offered, how will we equip it to entry and reply to real-world data, particularly to handle the newest queries and prompts?

As an example, the mannequin struggled to offer responses grounded in real-world knowledge when testing ChatGPT 3.5 with particular questions, as proven within the picture beneath:

Superb-tune the Mannequin

One strategy is to fine-tune the mannequin, maybe scheduling day by day periods often. Nonetheless, because of useful resource limitations, the viability of this method is at present beneath doubt. Common fine-tuning comes with a number of difficulties:

1. Inadequate Information: A scarcity of recent knowledge continuously makes it not possible to justify quite a few fine-tuning periods.
2. Excessive Necessities for Computation: Superb-tuning normally requires vital processing energy, which could not be possible for normal duties.
3. Time Intensiveness: Retraining the mannequin may take an extended interval, which is a giant impediment.

In mild of those difficulties, it’s clear that including new knowledge to the mannequin requires overcoming a number of limitations and isn’t a easy operation.

So right here comes AI Brokers 

Right here, we current AI brokers, basically LLMs, with built-in entry to exterior instruments. These brokers can accumulate and course of data, perform duties, and preserve observe of previous encounters of their working reminiscence. Though acquainted LLM-based programs are able to working programming and conducting net searches, AI brokers go one step additional:

• Exterior Instrument Use: AI brokers can interface with and make the most of exterior instruments.
• Information Gathering and Manipulation: They’ll accumulate and course of knowledge to assist them with their duties.
• Activity Planning: They’ll plan and perform duties delegated to those brokers.
• Working Reminiscence: They preserve particulars from earlier exchanges, which improves dialogue circulation and context.
• Function Enhancements: The vary of what LLMs can accomplish is elevated by this characteristic enhancement, which fits past fundamental questions and solutions to actively manipulating and leveraging exterior assets

Utilizing AI Brokers for Actual-Time Info Retrieval

If prompted with “What’s the present temperature and climate in Delhi, India?” a web based LLM-based chat system may provoke an online search to assemble related data. Early on, builders of LLMs acknowledged that relying solely on pre-trained transformers to generate output is limiting. By integrating an online search software, LLMs can carry out extra complete duties. On this state of affairs, the LLM may very well be fine-tuned or prompted (doubtlessly with few-shot studying) to generate a particular command like {software: web-search, question: “present temperature and climate in Delhi, India”} to provoke a search engine question.

A subsequent step identifies such instructions, triggers the online search perform with the suitable parameters, retrieves the climate data, and integrates it again into the LLM’s enter context for additional processing.

Dealing with Advanced Queries with Computational Instruments

In case you pose a query comparable to, “If a product-based firm sells an merchandise at a 20% loss, what could be the ultimate revenue or loss?” an LLM outfitted with a code execution software may deal with this by executing a Python command to compute the consequence precisely. As an example, it would generate a command like {software: python-interpreter, code: “cost_price * (1 – 0.20)”}, the place “cost_price” represents the preliminary value of the merchandise. This strategy ensures that the LLM leverages computational instruments successfully to offer the right revenue or loss calculation reasonably than making an attempt to generate the reply immediately by way of its language processing capabilities, which could not yield correct outcomes. In addition to that, with the assistance of exterior instruments, the customers can even e-book a ticket, which is planning an execution, i.e., Activity Planning – Agentic Workflow.

So, AI brokers may also help ChatGPT with the issue of not having any details about the newest knowledge in the actual world. We will present entry to the Web, the place it might probably Google search and retrieve the highest matches. So right here, on this case, the software is the Web search.

When the AI identifies the need for present climate data in responding to a person’s question, it features a checklist of accessible instruments in its API request, indicating its entry to such features. Upon recognizing the necessity to use get_current_weather, it generates a particular perform name with a chosen location, comparable to “London,” because the parameter. Subsequently, the system executes this perform name, fetching the newest climate particulars for London. The retrieved climate knowledge is then seamlessly built-in into the AI’s response, enhancing the accuracy and relevance of the data offered to the person.

Now, let’s implement and inculcate the Instrument Use to grasp the Agentic workflow!

We’re going to Use AI brokers, a software, to get data on present climate. As we noticed within the above instance, it can not generate a response to the real-world query utilizing the newest knowledge. 

So, we are going to now start with the Implementation.

Let’s start:

Putting in dependencies and Libraries 

Let’s set up dependencies first:

langchain
langchain-community>=0.0.36
langchainhub>=0.1.15
llama_cpp_python  # please set up the right construct primarily based in your {hardware} and OS
pandas
loguru
googlesearch-python
transformers
Openai

Importing Libraries 

Now, we are going to import libraries:

from openai import OpenAI
import json
from wealthy import print


import dotenv
dotenv.load_dotenv()

Preserve your OpenAI API key in an env file, or you’ll be able to put the important thing in a variable 

OPENAI_API_KEY= "your_open_api_key"

shopper = OpenAI(api_key= OPENAI_API_KEY)

Work together with the GPT mannequin utilizing code and never interface : 

messages = [{"role": "user", "content": "What's the weather like in London?"}]
response = shopper.chat.completions.create(
   mannequin="gpt-4o",
   messages=messages,
)
print(response)

This code units up a easy interplay with an AI mannequin, asking in regards to the climate in London. The API would course of this request and return a response, which you’d have to parse to get the precise reply.

It’s price noting that this code doesn’t fetch real-time climate knowledge. As an alternative, it asks an AI mannequin to generate a response primarily based on its coaching knowledge, which can not mirror the present climate in London.

On this case, the AI acknowledged it couldn’t present real-time data and advised checking a climate web site or app for present London climate.

This construction permits straightforward parsing and extracting related data from the API response. The extra metadata (like token utilization) will be helpful for monitoring and optimizing API utilization.

Defining the Perform

Now, let’s outline a perform for getting climate data and arrange the construction for utilizing it as a software in an AI dialog:

def get_current_weather(location):
   """Get the present climate in a given metropolis"""
   if "london" in location.decrease():
       return json.dumps({"temperature": "20 C"})
   elif "san francisco" in location.decrease():
       return json.dumps({"temperature": "15 C"})
   elif "paris" in location.decrease():
       return json.dumps({"temperature": "22 C"})
   else:
       return json.dumps({"temperature": "unknown"})

messages = [{"role": "user", "content": "What's the weather like in London?"}]
instruments = [
   {
       "type": "function",
       "function": {
           "name": "get_current_weather",
           "description": "Get the current weather in a given location",
           "parameters": {
               "type": "object",
               "properties": {
                   "location": {
                       "type": "string",
                       "description": "The city and state, e.g. San Francisco",
                   },
               },
               "required": ["location"],
           },
       },
   }
]

Code Clarification

This code snippet defines a perform for getting climate data and units up the construction for utilizing it as a software in an AI dialog. Let’s break it down:

• get_current_weather perform:
  • Takes a location parameter.
  • Returns simulated climate knowledge for London, San Francisco, and Paris.
  • For every other location, it returns “unknown”.
  • The climate knowledge is returned as a JSON string.
• messages checklist:
  • Incorporates a single message from the person asking in regards to the climate in London.
  • This is similar as within the earlier instance.
• instruments checklist:
  • Defines a single software (perform) that the AI can use.
  • The software is of sort “perform”.
  • It describes the get_current_weather perform:
    • title: The title of the perform to be referred to as.
    • description: A short description of what the perform does.
    • parameters: Describes the anticipated enter for the perform:
      • It expects an object with a location property.
      • location needs to be a string describing a metropolis.
      • The placement parameter is required.

response = shopper.chat.completions.create(
   mannequin="gpt-4o",
   messages=messages,
   instruments=instruments,
)
print(response)

Additionally learn: Agentic AI Demystified: The Final Information to Autonomous Brokers

Right here, we use three exterior Scripts named LLMs, instruments, and tool_executor, which act as helper features.

fromllms import OpenAIChatCompletion
from instruments import get_current_weather
from tool_executor import need_tool_use

Earlier than going additional with the code circulation, let’s perceive the scripts.

llms.py script

It manages interactions with OpenAI’s chat completion API, enabling the usage of exterior instruments inside the chat context:

from typing import Record, Elective, Any, Dict

import logging
from brokers.specs import ChatCompletion
from brokers.tool_executor import ToolRegistry
from langchain_core.instruments import StructuredTool
from llama_cpp import ChatCompletionRequestMessage
from openai import OpenAI

logger = logging.getLogger(__name__)

class OpenAIChatCompletion:
   def __init__(self, mannequin: str = "gpt-4o"):
       self.mannequin = mannequin
       self.shopper = OpenAI()
       self.tool_registry = ToolRegistry()

   def bind_tools(self, instruments: Elective[List[StructuredTool]] = None):
       for software in instruments:
           self.tool_registry.register_tool(software)


   def chat_completion(
       self, messages: Record[ChatCompletionRequestMessage], **kwargs
   ) -> ChatCompletion:
       instruments = self.tool_registry.openai_tools
       output = self.shopper.chat.completions.create(
           mannequin=self.mannequin, messages=messages, instruments=instruments
       )
       logger.debug(output)
       return output


   def run_tools(self, chat_completion: ChatCompletion) -> Record[Dict[str, Any]]:
       return self.tool_registry.call_tools(chat_completion)

This code defines a category OpenAIChatCompletion that encapsulates the performance for interacting with OpenAI’s chat completion API and managing instruments. Let’s break it down:

Imports

Numerous typing annotations and mandatory modules are imported.

Class Definition

pythonCopyclass OpenAIChatCompletion:

This class serves as a wrapper for OpenAI’s chat completion performance.

Constructor

pythonCopydef __init__(self, mannequin: str = “gpt-4o”):

Initializes the category with a specified mannequin (default is “gpt-4o”).

Creates an OpenAI shopper and a ToolRegistry occasion.

bind_tools methodology

pythonCopydef bind_tools(self, instruments: Elective[List[StructuredTool]] = None):

Registers offered instruments with the ToolRegistry.

This permits the chat completion to make use of these instruments when wanted.

chat_completion methodology:

pythonCopydef chat_completion(

    self, messages: Record[ChatCompletionRequestMessage], **kwargs

) ->

ChatCompletion

Sends a request to the OpenAI API for chat completion.

Contains the registered instruments within the request.

Returns the API response as a ChatCompletion object.

run_tools methodology

pythonCopydef run_tools(self, chat_completion: ChatCompletion) -> Record[Dict[str, Any]]:

Executes the instruments referred to as within the chat completion response.

Returns the outcomes of the software executions.

instruments.py

It defines particular person instruments or features, comparable to fetching real-time climate knowledge, that may be utilized by the AI to carry out particular duties.

import json
import requests
from langchain.instruments import software
from loguru import logger

@software
def get_current_weather(metropolis: str) -> str:
   """Get the present climate for a given metropolis.


   Args:
     metropolis (str): The town to fetch climate for.


   Returns:
     str: present climate situation, or None if an error happens.
   """
   attempt:
       knowledge = json.dumps(
           requests.get(f"https://wttr.in/{metropolis}?format=j1")
           .json()
           .get("current_condition")[0]
       )
       return knowledge
   besides Exception as e:
       logger.exception(e)
       error_message = f"Error fetching present climate for {metropolis}: {e}"
       return error_message

This code defines a number of instruments that can be utilized in an AI system, seemingly along with the OpenAIChatCompletion class we mentioned earlier. Let’s break down every software:

get_current_weather:

• Fetches real-time climate knowledge for a given metropolis utilizing the wttr.in API.
• Returns the climate knowledge as a JSON string.
• Contains error dealing with and logging.

Tool_executor.py

It handles the execution and administration of instruments, making certain they’re referred to as and built-in appropriately inside the AI’s response workflow.

import json
from typing import Any, Record, Union, Dict

from langchain_community.instruments import StructuredTool

from langchain_core.utils.function_calling import convert_to_openai_function
from loguru import logger
from brokers.specs import ChatCompletion, ToolCall

class ToolRegistry:
   def __init__(self, tool_format="openai"):
       self.tool_format = tool_format
       self._tools: Dict[str, StructuredTool] = {}
       self._formatted_tools: Dict[str, Any] = {}

   def register_tool(self, software: StructuredTool):
       self._tools[tool.name] = software
       self._formatted_tools[tool.name] = convert_to_openai_function(software)

   def get(self, title: str) -> StructuredTool:
       return self._tools.get(title)

   def __getitem__(self, title: str)
       return self._tools[name]

   def pop(self, title: str) -> StructuredTool:
       return self._tools.pop(title)

   @property
   def openai_tools(self) -> Record[Dict[str, Any]]:
       # [{"type": "function", "function": registry.openai_tools[0]}],
       consequence = []
       for oai_tool in self._formatted_tools.values():
           consequence.append({"sort": "perform", "perform": oai_tool})

       return consequence if consequence else None

   def call_tool(self, software: ToolCall) -> Any:
       """Name a single software and return the consequence."""
       function_name = software.perform.title
       function_to_call = self.get(function_name)


       if not function_to_call:
           increase ValueError(f"No perform was discovered for {function_name}")


       function_args = json.hundreds(software.perform.arguments)
       logger.debug(f"Perform {function_name} invoked with {function_args}")
       function_response = function_to_call.invoke(function_args)
       logger.debug(f"Perform {function_name}, responded with {function_response}")
       return function_response

   def call_tools(self, output: Union[ChatCompletion, Dict]) -> Record[Dict[str, str]]:
       """Name all instruments from the ChatCompletion output and return the
       consequence."""
       if isinstance(output, dict):
           output = ChatCompletion(**output)


       if not need_tool_use(output):
           increase ValueError(f"No software name was present in ChatCompletionn{output}")

       messages = []
       # https://platform.openai.com/docs/guides/function-calling
       tool_calls = output.decisions[0].message.tool_calls
       for software in tool_calls:
           function_name = software.perform.title
           function_response = self.call_tool(software)
           messages.append({
               "tool_call_id": software.id,
               "function": "software",
               "title": function_name,
               "content material": function_response,
           })
       return messages

def need_tool_use(output: ChatCompletion) -> bool:
   tool_calls = output.decisions[0].message.tool_calls
   if tool_calls:
       return True
   return False

def check_function_signature(
   output: ChatCompletion, tool_registry: ToolRegistry = None
):
   instruments = output.decisions[0].message.tool_calls
   invalid = False
   for software in instruments:
       software: ToolCall
       if software.sort == "perform":
           function_info = software.perform
           if tool_registry:
               if tool_registry.get(function_info.title) is None:
                   logger.error(f"Perform {function_info.title} shouldn't be out there")
                   invalid = True


           arguments = function_info.arguments
           attempt:
               json.hundreds(arguments)
           besides json.JSONDecodeError as e:
               logger.exception(e)
               invalid = True
       if invalid:
           return False

   return True

Code Clarification

This code defines a ToolRegistry class and related helper features for managing and executing instruments in an AI system. Let’s break it down:

• ToolRegistry class:
  • Manages a group of instruments, storing them in each their authentic type and an OpenAI-compatible format.
  • Offers strategies to register, retrieve, and execute instruments.
• Key strategies:
  • register_tool: Provides a brand new software to the registry.
  • openai_tools: Property that returns instruments in OpenAI’s perform format.
  • call_tool: Executes a single software.
  • call_tools: Executes a number of instruments from a ChatCompletion output.
• Helper features:
  • need_tool_use: Checks if a ChatCompletion output requires software utilization.
  • check_function_signature: Validates perform calls towards the out there instruments.

This ToolRegistry class is a central part for managing and executing instruments in an AI system. It permits for:

• Straightforward registration of recent instruments
• Conversion of instruments to OpenAI’s perform calling format
• Execution of instruments primarily based on AI mannequin outputs
• Validation of software calls and signatures

The design permits seamless integration with AI fashions supporting perform calling, like these from OpenAI. It offers a structured strategy to lengthen an AI system’s capabilities by permitting it to work together with exterior instruments and knowledge sources.

The helper features need_tool_use and check_function_signature present extra utility for working with ChatCompletion outputs and validating software utilization.

This code types an important half of a bigger system for constructing AI brokers able to utilizing exterior instruments and APIs to boost their capabilities past easy textual content technology.

These have been the exterior scripts and different helper features required to incorporate exterior instruments/performance and leverage all AI capabilities.

Additionally learn: How Autonomous AI Brokers Are Shaping Our Future?

Now, an occasion of OpenAIChatCompletion is created.

The get_current_weather software is certain to this occasion.

A message checklist is created with a person question about London’s climate.

A chat completion is requested utilizing this setup.

llm = OpenAIChatCompletion()
llm.bind_tools([get_current_weather])

messages = [
   {"role": "user", "content": "how is the weather in London today?"}
]

output = llm.chat_completion(messages)
print(output)

• The AI understood that to reply the query about London’s climate, it wanted to make use of the get_current_weather perform.
• As an alternative of offering a direct reply, it requests that this perform be referred to as with “London” because the argument.
• In an entire system, the following step could be to execute the get_current_weather perform with this argument, get the consequence, after which doubtlessly work together with the AI once more to formulate a last response primarily based on the climate knowledge.

This demonstrates how the AI can intelligently resolve to make use of out there instruments to assemble data earlier than offering a solution, making its responses extra correct and up-to-date.

if need_tool_use(output):
   print("Utilizing climate software")
   tool_results = llm.run_tools(output)
   print(tool_results)
   tool_results[0]["role"] = "assistant"


   updated_messages = messages + tool_results
   updated_messages = updated_messages + [
       {"role": "user", "content": "Think step by step and answer my question based on the above context."}
   ]
   output = llm.chat_completion(updated_messages)


print(output.decisions[0].message.content material)

This code:

• Test if instruments should be used primarily based on the AI’s output.
• Runs the software (get_current_weather) and prints the consequence.
• Modifications the function of the software consequence to “assistant.”
• Creates an up to date message checklist with the unique message, software outcomes, and a brand new person immediate.
• Sends this up to date message checklist for an additional chat completion.

• The AI initially acknowledged it wanted climate knowledge to reply the query.
• The code executed the climate software to get this knowledge.
• The climate knowledge was added to the context of the dialog.
• The AI was then prompted to reply the unique query utilizing this new data.
• The ultimate response is a complete breakdown of London’s climate, immediately answering the unique query with particular, up-to-date data.

Conclusion

This implementation represents a major step towards creating extra succesful, context-aware AI programs. By bridging the hole between massive language fashions and exterior instruments and knowledge sources, we will create AI assistants that perceive and generate human-like textual content that meaningfully interacts with the actual world.

Often Requested Questions