Build Anything with Llama 3 Agents, Here’s How

Summary

In this tutorial, David Andre demonstrates how to build AI agents using the Llama 3 model without extensive programming knowledge or powerful hardware. He utilizes AMA for local model running, VS Code for coding, and Gro for enhanced performance. The video showcases the setup process, from downloading the model to creating agents for tasks like email classification and response. Andre also highlights the potential issues with running the model through crew AI and offers a solution by connecting to the Gro API for improved speed and efficiency. The video concludes with a call to action to join his community for a comprehensive workshop on AI agent development.

Takeaways

• 😀 David Andre introduces a tutorial on building AI agents using the Llama model, which is accessible even for those with limited computer resources and no programming knowledge.
• 💻 The tutorial utilizes AMA for local model running, VS Code for coding, and Gradio for achieving high performance.
• 🚀 David demonstrates the impressive speed of the Llama model, achieving 216 tokens per second, highlighting the potential of the model even on less powerful hardware.
• 📈 The script showcases a comparison between the Llama 370B open-source model and the GPT-4 model, positioning Llama as a competitive choice.
• 🔧 David provides a step-by-step guide to downloading and setting up the Llama model, including instructions for using AMA and VS Code.
• 📝 The tutorial includes a practical example of building AI agents from scratch, focusing on an email classifier and responder scenario.
• 🛠️ David encounters and troubleshoots issues with the Llama model integration in crew AI, offering insights into potential solutions and workarounds.
• 🔗 The script emphasizes the importance of connecting Gradio to a team of agents to leverage the benefits of high-speed AI processing.
• 🔑 A guide is provided on how to securely use API keys with Gradio, including the creation and application of an API key for the Llama model.
• 🌟 The video concludes with a call to action for viewers to join David's community to stay ahead in the AI revolution and learn more about building AI agents.

Q & A

What is the main topic of the video?

The main topic of the video is teaching viewers how to build AI agents using the new llama free model, even without a powerful computer or programming knowledge.

What tools does David Andre recommend for building AI agents?

David Andre recommends using AMA to run the models locally, Visual Studio Code (VS Code) for writing the code, and Gro for achieving super-fast performance.

What is the significance of the 'llama fre model' mentioned in the video?

The 'llama fre model' is an open-source AI model that David Andre uses to demonstrate the creation of AI agents. It's significant because it allows for local running of AI models without needing powerful hardware or cloud-based services.

How does David Andre demonstrate the performance of the AI model?

David Andre demonstrates the performance of the AI model by showing the number of tokens processed per second, comparing the speed of the larger and smaller versions of the llama model.

What is the 'llm arena' mentioned in the video?

The 'llm arena' is a place where language models are ranked. In the context of the video, it's used to show that the open-source llama 370b model is better than the proprietary model GBD4.

What is the purpose of the community David Andre mentions?

The purpose of the community is to provide a step-by-step workshop for building AI agents, even for those who are not programmers, and to connect with others who are interested in staying ahead in AI development.

What is the first step David Andre suggests for setting up the AI model?

The first step is to download AMA from ama.com and then download Visual Studio Code from code.visualstudio.com.

How does David Andre handle the installation of the AI model in the video?

David Andre handles the installation of the AI model by showing viewers how to copy a command from the llama models page, open a terminal in VS Code, and run the command to download the model.

What is the issue David Andre encounters when trying to use the llama model through crew AI?

David Andre encounters an issue where the llama model works perfectly in the terminal but does not seem to work well when run as an agent through crew AI, causing unexpected results and slow performance.

How does David Andre resolve the issue with crew AI and the llama model?

David Andre resolves the issue by adding the Gro API, which allows him to use the llama model effectively and achieve the desired performance.

What is the final outcome of David Andre's demonstration with the AI agents?

The final outcome is the successful creation of AI agents that can classify and respond to emails, demonstrating the potential of the llama model and the crew AI framework for building AI applications.

Outlines

00:00

💻 Building AI Agents with Llama Model

David Andre introduces a tutorial on constructing AI agents using the Llama Free model, which is accessible even to those with limited computer capabilities and no programming background. The video demonstrates the use of AMA for local model execution, VS Code for coding, and Gro for enhanced performance. David showcases the Llama model's speed, processing 216 tokens per second, and compares it with other models, highlighting its superiority over models like GPT-4. He emphasizes the importance of keeping up with AI advancements and encourages viewers to build AI agents, offering a step-by-step workshop through his community.

05:01

🔧 Setting Up the Development Environment

The tutorial continues with setting up the development environment. David guides viewers to download AMA from their official website and VS Code from visualstudio.com. He then instructs on acquiring the Llama model, suggesting the 8 billion parameter version for its balance of size and performance. He demonstrates how to download the model using a command in the terminal within VS Code. Afterward, he shows how to import the Llama model into a Python script using the 'crei' package and sets up a basic Python file to begin coding the AI agents.

10:04

📧 Creating Email Classifier and Responder Agents

David proceeds to define the roles and goals for two AI agents: an email classifier and an email responder. The classifier's task is to accurately categorize emails as important, casual, or spam, while the responder's role is to draft responses based on the email's importance. He sets up the agents using the 'crew' library, detailing the process of assigning each agent a model, verbosity level, and delegation settings. David then outlines the creation of tasks for the agents, which involve classifying an email and responding to it accordingly. Despite initial issues with the Llama model's performance within the 'crew' framework, he successfully demonstrates the model's capabilities in a terminal test.

🚀 Integrating with Gro for Enhanced Performance

Facing performance issues with the Llama model in 'crew', David explores an alternative approach by integrating with Gro, a platform known for its API capabilities. He guides viewers through generating an API key from Gro Cloud and setting up the necessary environment variables for API integration. After modifying the script to use Gro's API, he successfully runs the AI agents, achieving a significant improvement in speed and performance. David concludes by reflecting on the rapid advancements in AI and the importance of staying updated with the latest technologies, inviting viewers to join his community for further exploration and learning.

Keywords

💡AI agents

AI agents refer to autonomous software entities that perform tasks or services on behalf of users. In the context of the video, the host is demonstrating how to build AI agents using the llama fre model, which is a type of AI model. The video aims to show that even without extensive programming knowledge, one can create AI agents to perform specific tasks like classifying emails or generating responses.

💡Llama fre model

The 'llama fre model' is an AI model mentioned in the video, which is used to build AI agents. It is part of the open-source movement in AI, offering an alternative to proprietary models like those from OpenAI. The video emphasizes the accessibility of this model, suggesting that it can be used even on less powerful computers.

💡AMA (Agent Management API)

AMA, or Agent Management API, is a tool used in the video to run AI models locally. It is showcased as a means to manage and interact with AI agents without the need for cloud-based services. The script mentions using AMA to download and run the llama fre model, highlighting its role in the local execution of AI tasks.

💡VS Code

VS Code, or Visual Studio Code, is a popular source-code editor used by programmers for tasks like writing and debugging code. In the video, VS Code is recommended for writing the code necessary to create AI agents, suggesting that it is a user-friendly and powerful tool for coding, even for beginners.

💡Gro

Gro is mentioned as a tool to achieve 'super fast performance' in the context of running AI models. It is used to enhance the speed at which AI agents operate, which is crucial for real-time applications. The video shows how to integrate Gro with the AI agents to leverage its performance capabilities.

💡LLM (Large Language Model)

LLM stands for Large Language Model, which is a type of AI model designed to understand and generate human-like text. In the video, the host uses the 'llama fre 8 billion' model, a smaller version of the 'llama 370b' model, to demonstrate the classification and response capabilities of AI agents.

💡API (Application Programming Interface)

An API, or Application Programming Interface, is a set of rules and protocols for building and interacting with software applications. The video discusses using an API key from Gro to connect and utilize its services for enhanced performance of AI agents. This demonstrates the practical use of APIs in integrating different software services.

💡Email Classifier

An 'email classifier' is a type of AI agent that categorizes emails based on their content. In the video, the host creates an AI agent with the role of an 'email classifier' to automatically sort emails into categories like 'important', 'casual', or 'spam'. This showcases the practical application of AI in automating routine tasks.

💡Crew AI

Crew AI is a framework or library used in the video for creating and managing AI agents. It provides a structured way to define agents, their roles, and tasks. The script mentions using Crew AI to import necessary components and set up the agents for tasks like email classification and response.

💡Community

The term 'community' in the video refers to a group of people with shared interests, in this case, AI and its applications. The host encourages viewers to join his community to learn more about building AI agents and stay updated with the latest developments in AI technology.

Highlights

Introduction to building AI agents using the llama free model without programming knowledge.

Utilization of AMA for local model running and VS Code for coding.

Achieving high performance with Gro and comparing it with the big model.

Demonstration of instant response with the smaller llama free 8 billion model.

Instruction on connecting Gro to a team of agents for enhanced speed.

Llama 370b, an open-source model, outperforming gbd4 in the llm arena.

Urging the audience to build AI agents to stay ahead in the AI revolution.

Step-by-step workshop to build AI agents from scratch.

Downloading AMA and VS Code as prerequisites for the project.

Selection of the llama 3 model for its balance between size and performance.

Downloading the model using a command in the terminal.

Quick tip on ending chats without killing the terminal in VS Code.

Importing the llama model in a Python file using the LangChain community package.

Setting up the email classifier agent with a specific goal and backstory.

Creating a responder agent to write email responses based on classification.

Defining tasks for the agents within the crew AI framework.

Troubleshooting the integration of llama 3 with crew AI and finding a solution.

Adding the Gro API for improved performance on lower-end computers.

Demonstration of the significant speed improvement with the Gro API.

Invitation to join the community for those interested in cutting-edge AI.

Transcripts

my name is David Andre and I'll show you

how to build AI agents using the new

llama fre model even if you have a bad

computer and know nothing about

programming now we're going to be using

AMA to run the models locally vs code to

write the code and Gro to achieve super

fast performance and look at this and

now boom 216 tokens per second now this

is the big model let's try the smaller

one Lama free 8 billion and let's do

tell me a long joke

like that was instant so at the end of

this video I'll show you how to connect

Gro to your team of agents that way you

can get the benefits of this insane

speed now check this out this is the llm

arena where language models are ranked

llama 370b an open source model is

better than gbd4 the future is here if

you aren't doing anything about it if

you aren't building AI agents you're

falling behind so I urge you to take the

first step and build something to make

this as easy as possible for you I've

built an entire Workshop that will teach

you step by step how to build AI agents

even if you're not a programmer you can

get it inside of my community the link

will be in the description so let's get

to building agents I was planning on

building on top of a previous project

but then I realized I think you'll get

more value if I build everything from

scratch so the basics right go to

ama.com and hit download then you need

vs code go to code. visual studio.com

and download it obviously the next thing

you need is the model so so if we go to

all Lama click on models and select

llama 3 obviously I would recommend

going with 8 billion model which is

already really good so copy this command

just click on here copy it go to your

terminal and if you don't have u k

that's okay in vs code just click on

Terminal and do new terminal and you

should be able to run the AMA right here

right yeah exactly the first time you

run this obviously you won't be able to

run the LM you cannot just be like hey

llama because you haven't downloaded the

model yet so the first time do this it

will start downloading the model now 4.7

GB the smaller one so it will take like

20 minutes the bigger one though if you

have a really good PC this one is 40 GB

so this will take like 3 hours probably

either way you have to download it the

first time you use it right which is

fine because you only have to do this

once by the way quick tip whenever

you're in a chat and you want to end it

just go SL bu and you don't have to kill

the terminal each time so we run this

and this is just to chat with the llm

locally and that's already pretty good I

mean most people can't do that but

honestly we're not interested in that

this is kind of a basic we want to build

AI agents with Lama fre so let's go into

vs Studio by the way if you're not

subscribed Please Subscribe and I

created just an empty file you know

clicking right clicking new file naming

it main.py just an empty python file

literally as simple as that so what we

need to do is we need to import AMA so

go from Lang

chain community. llms import or Lama and

as you can see it's highlighted meaning

we have not installed it I have to be in

the correct um environment okay so yeah

if you have not installed this package

yet don't worry just go in here go into

your terminal by the way again you can

do this straight in vs code just go into

the terminal make sure you're in the

right environment I mean you can just

use the base environment if you want and

just type in PIP install crew AI

literally just type that and it will

install everything and as you can see I

already have it if you don't have it it

will take like 20 seconds so don't worry

inside of crei um you'll get the L chain

Community package which is what we'll

use to import o Lama next we need crew

AI so from crew AI we need to import

some things we need to import agent task

crew and process that's it those are the

two Imports we need next thing we need

to uh decide the model so I'm just going

to use a variable and O Lama we have to

do model

equals and put the name of the model as

it named in ama if you're using the

default version this the 8 billion just

type in Lama 3 if you want to use the

70b version um then do llama free 70b so

I'm just going to do llama free because

um faster it's faster so yeah and now I

need to decide what type of Agents I

want to build and I'll just keep this

simple that way all of you can follow

along and what I came up with is a

simple email classifier so we'll go the

first agent will receive receive a

random email and classify it and based

on the classification the second agent

will write a response so let's create a

new variable named email and this is

just going to be a simple string and I'm

going to do a random one uh let's do the

classic Nigerian PS sending some gold

let's see if the llms are smart enough

to recognize this is a spam by the way

all of the code and prompts from this

video will be linked below the module so

another reason to join the community

anyways let's create our first agent and

this is going to be

classifier and we're going to do agent

like this beautiful and first we need to

give it a role which is basically just a

second name right and this is going to

be email

classifier next we need to give it a

goal so what is what will this agent do

what's his main goal goal equals

accurately classify emails based on

their importance give every email one of

these ratings

important casual SP or spam and then we

need to give it a backstory and I'm just

going to keep it super simple that way

this is not a 2hour video the next thing

I need to do is set verbos so ver both

equals true this will let us see the

flow of the messages inside of the

terminal next we do allow delegation

equals false because we don't want this

agent delegating this task to somebody

else else and lastly and this is super

important we need to assign the llm

which is model if we don't do this it

will automatically default to gp4 which

will not work because we haven't

provided the API for open AI but we

don't want gbd4 we want llama free and

that's our agent done crew AI makes this

super simple next agent is going to be

responder so responder equals agent and

then we basically do the same but for an

email responder and again kab copal just

Sav me some time based on the importance

of the email write a concise and simple

response all right we're both true

beautiful all of this is the same so now

we have our agents created super simple

now we need to define the task so first

task I'm just going to name classify

email equals task and creating tasks

inside of crei is even simpler than

creating agents first we do a

description description equals and then

I'm going to do an F string and the

reason for that is because I want to

enter the variable we've created right

for the email so F strings in Python

just allow you to insert the variable

inside of a string agent equals

classifier and then expected output one

of these three options import on casual

or spam yes beautiful so that's our

first task created then we're going to

create a second task respond to email

okay

exactly beautiful description respond to

the email now we have created both of

the tasks next step is to define the

crew crew equals crew and first we need

to assign the agents exactly a list of

Agents so just create a list and enter

all of the agents you have can be two

can be three can be 10 it's up to you

then we do tasks and be careful because

I used to put task here and this will

not cause an error but it will cause

your crew not to work verbos is

different here it's either one or two

I'm going to do two and then process

equals process. sequential and the last

thing we need to do is output equals

crew. kickoff okay and then obviously I

need to print it out to see it and now

this should work fingers crossed let's

see it do we have a team of llama free

agents working or

not okay this is promising let's look at

the activity

monitor so we have memory used 36 it

went up from like 30 which is means that

the llamas 3 8 billion uses like 3 to 6

GB of RAM which is very good news

because most of you can run it but this

is bad news

because it's probably going to include a

bunch of Chunk it should not be this

slow it should be much faster and I'm

suspecting that we're going to see a

bunch of hallucinations oh my God what

is this so I don't know why this is

happening but inside of crew AI

llama 3 does not seem to be working well

it works perfectly in the terminal but

some like somehow through crew AI it's

uh not really working well I mean I can

try verbose false it's not the issue of

the model clearly like look at this this

is perfect right look at this the model

said a classic rating spam the Nigerian

pen is a welln example of a fishing scam

yeah so perfect I mean it gave it the

rating exactly as I wanted but somehow

running it as a agent through crew AI is

causing troubles I have no clue why so

we tested this now I'm going to show you

how to add the gro API so if you have a

bad computer and you want to use the API

I'm going to show you how so okay let's

go Gro and let's go Gro cloud and let's

go API Keys create API key I'm just

going to do llama 3 submit and I'm going

to delete this before posting the video

you know never share API keys with

anybody

I'm going to do Gro API key equals let's

go into documentation see how we connect

it let me duplicate this rename cuz we

want to do Gro so now this is for the

gro right so we delete AMA and okay

let's follow along with this we can just

probably delete this all together

honestly following the GitHub official

GitHub from crei might be the move right

so we need to assign these three

variables and just make a default right

so open a API

base always that environment yes so we

need to import OS we copy this we need

these three lines and we actually need

them not commented out beautiful let's

go to Gro

documentation and open a compatibility

and the base URL copy this API base

equals model name let's copy this

again and the API key test Moment of

Truth oh my god it works it works and

look at the speed though let's try the

different email and let's do it look at

the

speed okay what am I doing okay so I'm

just going to explain what I did so I'm

I was completely trolling the answer was

right in front of my eyes right here

inside of the main GitHub from crew AI I

just needed to assign these three

variables right and the reason why I

deleted the llm is because crew AI is

automatically made to use the default

one so like this is like basically

hidden right something like this is

hidden and you don't have to do that if

you're using the open endo and guys

let's just appreciate the speed one more

time like this is an API look at this

insane this is the AI Revolution we are

watching the the AI Revolution happened

right in front of our eyes unfortunately

99% of people will miss out on the AI

Revolution and I don't want you to be

one of them the only way to stay ahead

is to surround yourself with people who

are at The Cutting Edge of AI which is

exactly why I created the community I'm

looking for people who want to take AI

seriously so if that sounds like you

make sure to join for this link in the

description