Welcome To The Agents Course! Introduction to the Course and Q&A

and we are live welcome

everyone welcome to the first agent

course

live um let me see all of you uh we're

going to start the live uh in two

minutes so in the meantime please tell

us where you come

from uh in my case I'm based in Paris

what about you

jofrey I'm also based in Paris and

Ben uh I'm based in ANP in Belgium nice

so various people from Chicago Pakistan

India like Paris ah yes

nice um let's

see Ecuador Iran Florida USA turkey a

lot of people we saw that there is a lot

of people from uh from USA and from

India Paris nice

Liverpool Canada

India Saudi Arabia yeah it's really all

around the world we try to find an hour

that fit for everyone but you know it's

very hard because you know everyone is

distributed around the world so um Chile

but this this will be recorded uh France

bellarus Chile

Nepal nice yeah very is really around

the world so yeah a lot of people really

excited to to create agents so it's very

nice um yeah so as I said to to people

who who just uh arrive um we starting in

two minutes just to wait but everyone

who wants to join can

join Israel Florida India

Tunisia yeah very nice oh we're already

2,200

people that's amazing that's really

amazing um uh to be honest we I I I

really thought that will be and I think

we really thought that we be like I

don't know 500 people so it's it's very

nice to see all of you so thank you for

being here um yeah I think we can we can

start maybe seven yeah let's

start

so uh first of all hello everyone and

welcome and thank you for being here uh

for this first live of the agent course

uh the first thing we want to do is to

introduce ourselves so my name is Tomas

simonini I'm a developer Advocate at

hugging face and I'm um I'm some people

know me in this live because I wrote The

Deep reinforcement learning course two

years ago um and also work on the AI in

Games part at Hing face and currently

I'm one of the co-creator with jeofrey

and Ben of the agent course what about

you

Joffrey hi um nice to meet you everyone

I'm Joffrey so I'm a machine learning

engineer from huging face and um I was

working in the monetization team and now

um my focus is on

agents what about you

Ben thanks I'm Ben I'm a machine

learning engineer and I've given courses

in in data sets and in posttraining

llms and uh I worked on course as well

and I focused on most of the llm

material and I I built the quizzes and

the

certifications and yeah those are my

favorite parts of the course as well in

case you have any questions

about okay so let's get started um so uh

for people who don't know uh this live

is about the agent course it's a new

course uh at hugging face where you will

learn it's a free course from beginner

to expert uh the goal of the course uh

is to teach you how to create agent um

so it's a free course and I I think you

already know that you know agent is

really the topic in AI in

2025 uh we see a lot of companies being

created around that you know we see a

lot of project about that and so I think

it's really the skills to acquire this

year and fortunately with this course

you're in the right place um on this

course we have already

72k uh signups which is amazing

and we want to thank you for that and we

really hope you're going to like it we

work really hard for that and so let me

just uh give you some information about

how the this live is going to to to

happen uh there will be a first part

which is a presentation of the course

when I will give you a little bit of

information about the course the scope

the units and the challenges and then

there will be a Q&A uh during around 30

minutes and after that we will have a

small surprise to to announce um and one

thing is that don't hesitate to start to

ask your question now in the chat so

that you know my my colleague can you

know check the the

question um so as I said uh we launch

the course um like uh on Monday and so

you can have access here with this link

uh at

h.n agent course

um okay sorry so there is two ways to

find the course uh the first is uh with

this link which is

hf.org anization of the hugging face Hub

uh this this is where you're going to

find the quizzes uh the space uh

everything you need to do the enzone and

this is also where there is all the

links of the course and the second link

it's

hf.com Z and one are

published one thing I want to mention is

that we have a Discord server where we I

really advise you to join because it's

it's a place where you will be able to

exchange with us exchange with your

classmate and ask questions to do

together I think it's always easier when

you study something new to to to be able

to participate with multiple people um

and this is also where we give the

latest update

information so the link you have the

link here which is

h.c sorry SL jooin

Discord um but you also have the link uh

in the course when you go to unit zero

and you do the

onboarding um also um the best way to

keep being updated about the new unit

the new update Etc is uh to sign up to

the course uh so if you already signed

up to the course you should have

received some mails from us if it's not

the casee please check your spam uh

folder and if not uh the link is here uh

also you have the link directly when you

go to to the the course directly you

have a link uh to to sign up to the

course so yeah someone asking yeah it's

free the course is completely free there

is no uh it's totally free it's totally

open source so it's it's uh you it's

totally free uh yes and also the video

is recorded

um so again so what I said I put I put

the link um and what I said is that it's

uh really the best moment to start to

learn about agents and so we prepared

something very nice for people to start

with the theory than the practice so

that you know your at the end of this

course you have a very strong Foundation

to be able to build your own

agent so to give a little bit more of

information about what we do during this

course you're going to study AI in

Theory design and practice the idea is

that we don't we we also want in

addition to the enzone that you

understand the theory behind the agent

so that when you will use a new library

you will be able to rapidly you know uh

keep up the pace with this library

because you will have the theory and so

it will be much easier to understand how

to use new libraries you're also going

to use establish AI agent libraries such

as small agent long long chain and Lama

index you're to share your agent on the

hugging face Hub and explore uh agent

created by the

community and you'll participate in

challenges where you will evaluate your

agents against over students this will

be during the unit four and finally you

will hun two certificate of completion

by completing assignment um we think

that it's important the certificate are

important because this is a good way to

show to your employer or future employer

that you have the skills on about agent

so yeah again uh I put the link uh of of

the course so uh but in any case what if

you didn't get the link what you just

need to do is that you type on Google

you know h.c learn or agent course bying

phas and you will find

it I just want to emphasize rapidly on

the free pass that you can do during

this course uh the first is to

participate as an auditor uh where you

it just you know you want to read one of

the unit but you know don't do the wall

certification

process um what I something important

because I see in some question this um

you don't need to tell us which PA path

you you you you choose it's very because

it's already it's automatic what I mean

is that if you you pass the quiz of the

unit one you can get a certificate if

you pass the unit one a use case

assignment and the final challenge you

can get the completion certificate

so what I mean is that you don't need to

tell us you just you know do the quizzes

and the assignment and then you go to

the certificate section and you can

generate your certificate my advice is

at least try to do the certificate of

fundamentals because I think it will

since you will get a good foundational

agent it will be important and useful

for your career to be able to show you

know this certificate to say I pass the

agent course bying phase so it shows

that you have some skills in agent so if

you don't want to do the world course at

least I think it's important to earn a

certificate of fundamentals all this

information that I'm telling you by the

way will also be are also in the unit

zero uh during the

onboarding one important thing is that

we sorry could I just add a bit there on

the fundamentals sure so one aspect to

the fundamentals that you should know is

that the quiz doesn't require any code

so you can get that certification

without following the code components

which we think is really good for like

decision makers and people that don't

use code but they're really interested

in in agents and they want to work on

agentic projects and get a certificate

for that so just know that that's yours

kind of certificate if you don't use

code yeah yeah good

yeah uh so there is one deadline um it's

you need if you want to get the

certificate uh you need to do the

assignments and finish before May the

1st

2025 so to help you to be on time with

the with the deadline what we done is

that we created what we call a

recommended pace obviously this is just

a recommendation you can you can work at

your own pace but we we provide you this

recommended pace where you see the

syllabus and you see you know what week

we advise to to work on um so yeah just

to rapidly quick uh to to go to the to

syllabus um you have unit one where it's

already published it's an introduction

to agent you have a theory part and a

practice part you're going to create

your first agent using small agent uh

next week we have a bonus unit where you

will learn to find your agent after that

we will have unit about Frameworks such

as longchain uh Lama index and uh small

agent and on the third unit will have

the use case like how you can use agent

in real world you know what type of use

case you can use with you know an

assignment and at the end you will have

on unit four what we call the final

assessment where you will create an

agent and you're going to Benchmark

against over um over classmates the idea

is that we're going to build a

leaderboard where you will be able to um

Benchmark your agent but we will give

you at the weeks go by we're going to

give you more information about the

syllabus because it's a work in progress

uh based on your feedbacks that's why at

the end of the section we have a

feedback form that you can feel and it

help us to know which element you want

us to add etc etc so don't hesitate to

fill

them um I just want to scheme rapidly on

the best practices to succeed in this

course they are always the same go on

Discord don't hesitate don't be shy we

have a very nice Community very friendly

and I want to thank personally the

personally sory the people who really

help on the Discord server but answer

question I see that the community is

very active so it's very it's very nice

thank you for that so yeah Discord is

really the place where you can you know

exchange find classmate Etc it's very

nice also in each unit we have what we

call ungraded quizzes that will help you

to verify that you really grasp the

material so don't hesitate to take them

it's really important I think to be sure

that you really you know that you not

fall into what we call illusion of

competence and finally Define a schedule

to stay sync because you know as I said

there is a deadline so don't don't so I

think it's important that you have at

least every week you know you work on

one part of the

units and yeah you need you don't don't

need a lot of tools because a lot of

people ask me you know can I use my

phone and I will not advise to use your

phone because you will need to code so

you will need a computer and you will

need aing face account and a free

account is good enough you don't need a

paid account for this course

uh one thing I I didn't forgot to

mention the requirement for this course

is you need to have at least some skills

in Python and you need to have a little

bit of you know a little bit about llm

large language model if it's not the

case we put in unit zero and unit one

links that you can check you know we

have an NL natural language processing

course that you can follow but uh my

advice is that you need to already have

a little bit of skills in Python in code

and a little bit in llm but in any case

if you are very motivated we give some

links to to this course so you will be

able to keep up the

pace and one last thing be before I we

move to Q&A uh there is fre way that you

can help us to promote the course the

first is to like this video because it

helps to reach more people the second is

that we have a GitHub repository where

we put the course which is github.com

huggingface Agent course and liking ing

or GitHub report helps to make it more

visible to the GitHub community and

finally don't forget to follow the agent

course

organization uh uh because it's also

where you will have all the information

and the main entry point for all the

over

links

so again this is the link of the course

and I think now it's time for Q&A uh one

thing uh if you have technical question

about you know I can join this c um I

have PR you know generating my

certificate or you know on the quiz um I

think the best is on the certificate

space on the quiz space Etc to go to the

discussion tab uh because this where we

can track and we can you know rapidly

check what's going on based on your

username so that you know we can check

if you pass this quiz if the certificate

is generated ET it's much easier for us

and so now I let my two colleague to uh

to answer your question and if we don't

have time to answer your question don't

hesitate to ask in uh the agent course

question channel on

Discord hi so the first question that uh

I noticed was this one it was about

Python and it said what would you

suggest for those who are not familiar

with python and to get the most out of

the course so yeah this is a really good

question one that I'm I'm really

interested in as we said the first unit

of the course doesn't requ ire python to

complete but it does show some Python

and it does guide you through some steps

and obviously if you don't understand

that programming language you're going

to find those parts of it quite

difficult um and so we suggest that you

go to the reading material and you focus

on the fundamentals and you get the most

out of those fundamental parts of the

course from unit two to three it's for

um python programmers and so those units

won't be won't be ideal for you and it's

that first unit that you should focus on

we're also going to release bonus units

in the future and those bonus units will

have kind of different Target audiences

and they could be like no code bonus

units if that's something that you want

and so if you really enjoy that first

unit and because it's no code then say

that in the forms um and push us that

way and we'll release content like

that yeah sorry for the nonf French the

this question is in French but I will

translate so basically this question is

asking if the the course will be

available in French and let's uh

extrapolate in other languages too so at

the moment it's not this is an English

only um course but if you wish to to

support your local language you can do

PRS to translate in other

languages so uh a short question about

another course that we've done in the

over the last few months about small

course this was an experimental course

on loads of different subjects and it

was really a way of finding out what the

community wants and this course is

different to that one uh and it um is

fully fledged it's interactive U and it

has material that you can follow along

and live sessions like this small course

was kind of static and and asynchronous

and and small and so they're very

different courses and if you're here

then you probably know that and that's

great

I saw this question uh talking about the

NLP course so do you recommend doing

this after the AI agent course this

would be the other way around like if

you feel that at some point you don't

have the NLP background it would be

better to do the NLP course first but uh

that being said in the agent course we

start from scratch so all knowledge that

is necessary to understand agent are u

in the uh um unit

one okay I'm going to answer two

questions in a row that are both about

timeline the first one here is that why

is there a hard deadline on the 1 of May

um in real terms the deadline is is not

that hard but it is is about creating a

kind of cohort and a group of students

that are studying and discussing

together so that we can all collectively

get the most out of it and kind of

motivating ourselves the certification

process runs in parallel so we can

collect certifications as you go you can

adapt the pace to your own life and and

you can move slower and faster in

various Bits And if you're um getting

your certificate after that date uh

we're going to support you in that if

you need and so it isn't a really really

hard deadline it's about creating a

small community that's working on

something at the same time the other

question about timing was how much time

commitment is required this is um not an

easy question to answer really because

it depends on on your own life and and

how deep you go we're saying around

three to four hours per week uh per unit

sorry as we release them and we're going

to release a unit every two weeks we

have some bonus units in there there's a

lot of written material there are

exercises that you can go deeper on than

than we encourage so I'd say there's um

three to four hours is sort of minimum

and you can take it up from

there I saw question asking about the

longchain part so is it longchain agent

or long graph in fact it's going to be

both because we are going to start

explaining the the principles of

longchain and then moving on to L

graph so I saw this question do we need

a GPU for this course but I'd kind of

extend this out as well say what sort of

Hardware do we need for this course no

you don't need a a

GPU we use uh inference apis and the hub

for the course and so and we also use a

number of smaller models at various part

of the course so you don't need a GPU if

that was something that you had you

could kind of get more out of the course

you could try different models from The

Hub or you could download those and kind

of take it further and uh that would be

pretty cool and quite exciting so it's

definitely something you can take

advantage of but you should be able to

run this on a on a general uh computer

nothing fancy just that you can access

Python and that will take you all the

way to the end of the

course I'm seeing question about which

which kind of llm are we going to be

using so at the moment in unit one we

used two uh but I'll try to be um

as generic as possible meaning like in

future unit we will use other llms and

small uh LM that we haven't used in

previous

units

okay sorry um okay so this question

asked about the C certification process

and the person said that their hugging

face account doesn't have their their

full name that's okay uh within the

certification process you can Define

whatever name that you like but the the

certificate will be attached to your

hugging face profile and the reason that

we do that is is just so that the

certificate has validity and it links

back to her profile because this is just

something that um people want with

certificates so that they're not just an

image that they actually have some kind

of credential and and people know that

they're a valid certificate so it just

kind of adds some some value to you to

have those certificates because they're

they're maintained and they're linked

back to a

reference so can we make the final

assignment if we do not have prior

knowledge of LM that's the purpose of

the course uh is to guide you through

from no knowledge of n llm to being able

to actually code um an agent and compete

in a in a benchmark in a leaderboard

against other

students this question asked why have

you chosen only three Frameworks okay I

mean you could ask the question both

ways why have we chosen three Frameworks

and and not one most courses will just

choose a framework and teach you through

it and um I I suppose we could choose 10

10 Frameworks the reason we haven't

chosen more than three is because we

don't think it would be a a feasible

course for us to build to a high

standard and and for students to follow

that's that's a lot of Frameworks to

learn in a few months and in reality

most people don't really need more than

a few why have we chosen three and not

one well that's because we wanted to

build an agnostic course that wasn't

really about Frameworks Frameworks are

just a way of doing something this

course is about agents and about the

fundamentals and understanding them and

we've chosen three of what we think are

the best Frameworks out there and if you

look at the statistics on them they're

the most used as well and they're kind

of three diverse Frameworks so three

diverse tasks so that's why we've chosen

those so is this course relevant for

project managers uh we are trying to

make this course as general as possible

so that's what we divided it into

multiple units unit one is here for the

fundamentals so if you feel that your

python uh skills are too light at some

point it's it's okay to to just read the

materials and not do the coding steps um

and not necessarily like do the final

assignment which will be a little bit

more code uh

demanding so this is a kind this is a

cool question I really like it um what

was your motivation to come up with a

course at this moment in time can you

share more of the perspective about AI

agents in the industry to 2025 it's a

really kind of long-winded question

maybe Joffrey wants or Thomas want to

answer this as well after me because

it's kind of like just a a position

question um so for me my personal place

in it is that I've been building small

courses around llms and teaching people

how to use LM at different kind of

levels and um it's really exciting but

it's as we move into agents that they

actually become both exciting and kind

of fully usable so it's really where you

let these models loose and you can

actually do things with

them also now in 2025 like the problem

of inference and compute seems to be way

less significant than it was the year

before and people can actually access a

diverse range of models on the Hub and

so we can kind kind of get closer to

building applications and switch between

different open source models so there's

so much more you can kind of do there

without needing to worry about hardware

and go into those kind of complexities

which means that more people can build

this stuff so so for me that was why

this is the right time to build AI

agents and there's also a wider range of

of Frameworks and libraries out there

now so from small agents up to to langra

there's kind of more to learn there than

there was here

ago so maybe are there only Theory or

video tutorials um it's exactly the

right time to ask this question there's

going to be uh an optional unit next

week that is going to cover fine tuning

um that I'm going that I recorded and

it's a video in death

case cool uh this question asked was

kind of extra study material I finished

the first unit yesterday what should I

do to practice in preparation for the

future units so we've added links

throughout the material that you can

collect and kind of go off and do some

extra

reading uh and obviously it's not a it's

not two weeks of material that you know

80 hours of work or something it's

definitely not not that much and and it

kind of will fit in most people's lives

so there's a lot of space there

personally I'd encourage you to go

through the reading

material first and kind of collect that

together and go into that I'd then

encourage you to kind of enter into the

discussions in Discord and if you really

feel like you're getting your head

around that material then maybe you're

one of the um you know one of the the

Vanguard students a bit further ahead

and maybe you can help your colleag and

kind of work through those questions and

maybe some of those questions are quite

difficult for you and you can go back

and work on those and help each other

that's really cool and then maybe you

can go back to the exercises and kind of

go deeper on them and build more complex

applications and start to find the

limits of your abilities so that you can

have ideas about the kinds of use cases

that you'd want to build or or what you

find interesting and challenging because

then as you come into unit 2 and you

start to see framework

then you'll start to think okay now I I

know there's this problem I want to

solve maybe this itch I want to scratch

how can I do that with this framework

and you'll start to unpack and grow like

that that that would be my my

recommendation so does this course

require knowledge about the hugging face

ecosystem well um the course is

happening on hugging face and we are

using a lot of hugging face tools

so this is something that we need to

work through to to get you that

knowledge about using the hugging pH

ecosystem but hopefully um the lesson

are clear enough so that you you get

that knowledge on the

goal cool so this question asks can we

use an agent as a slm a small language

model or just a large language model an

llm so um yeah this is a a difficult

question in the sense that they're not

really two separate categories you know

there's a bit of a spectrum and on that

Spectrum you have small models of 135

million parameters if you look at the

smallest smaller models and models right

the way up to 72 billion parameters if

we look at quen really powerful models

and even some models with greater

parameters than that but

those models are really relevant to the

task at hand so there might be some

tasks that are about structuring text

and restructuring text in certain

formats and smaller language models can

can do that and they can do it pretty

effectively and you might be able to

find that a 1.7 billion parameter small

LM can perform that task uh and maybe

even smaller than that then you might

work on a on a kind of reasoning or math

or code problem and that that model

probably might not be able to do that

and so you'd go up to a a 7 billion or

even 72 billion parameter model in order

to perform this kind of complex task and

I'd encourage you to to explore that and

to find out both what models of certain

sizes can do and what they can't do uh

and if you've got kind of real Hardware

constraints um then I would encourage

you to focus on small language models

and to work on TKS that they can solve

and to figure out what that

is um so are we going to see use case

that show how to move in production

environment or AI

agent um exactly so basically uh at the

moment we only s the fundamentals but

going on with next week we're going to

see for works and unit three is about

use cases and use cases is actually

building a real case like you would be

in production into um to to fulfill an

objective so I I think it relates to

your

question so this question asks about

submitting code as part of the

assignment for the unit completion so in

unit one as we mentioned there's no code

submission these are multiple choice

questions within a quiz format that are

that challenge your understanding of the

fundamentals as we go on we will um

grade we will not use quizzes in every

single unit for assessment and in the

third unit we may use a quiz kind of

based on feedback on where people are

going and that might be a kind of more

code related quiz because units two and

three challenge those things but what

we're certain of is that there will be

an assignment at the end of the project

which does contribute towards your sort

of certificate of

completion so does the course teach VM

agent and vision agent uh yes it will to

some extent um VM are really used oh

sorry are really used as tools um in

onance we browsing um so this is a

fundamental also um component of what we

need to to learn in order to to complete

the last

assignment so this question asks whether

we'll build deep research um deep

research agents that's how I understood

it like the ones recently released by

open AR and or even the open alternative

by the makers of of small agents I'm not

sure if that's going to be a specific as

the specific title of of one of the

exercises but we're working towards a

guia style evaluation and that's very

similar to that application and so in

effect if this was something that really

interested you you could definitely

Point your final exercise towards that

kind of use case and maybe add an

interface on top of that

and work on that I'm not sure if Joffrey

wants to add anything more to that

question

because that's perfect perfect answer so

basically open deep research just took

the second place of the Gaia Benchmark

and the goal of the final assignment is

a Gaia style Benchmark with different

question and so if you want to reproduce

that kind of um of of coding that they

did in open deep research this is

totally possible

here's another nice question

um are we building our own agents or

building on top of other models

so uh This what I I get from this

question is maybe that you want to take

a look at some of the chapters in the r

material what are agents what are llms

and kind of go through that section and

just to get an idea of of the difference

between an agent and a model so when we

talk about models we're talking about

the llm at the core of the agent this is

a a set this is a a binary process it's

it's not a piece of code it returns back

a kind of reasoning or or text that it's

generated about a decision the agent as

a whole is a is a system that um reacts

and and acts upon events and and and

performs those those functional steps

based on the decisions made by that llm

at the course the model of the course so

they're two different things and um will

you you'll be building agents using

existing models from the hubs and you

won't make any real modifications to

those models you'll just be building

agents so does this course connect llm

agent with agents might be a question

for Thomas but basically no LM agent are

very different from LL agents so if

you're interested in erl agent instead

you may be better by checking the the

Perl

course yeah there is a it does not

connect um except you know if you train

your LM using lfh but the Deep

reinforcement learning course is not on

this topic is more on classical deep

reinforcement learning from Q learning

to PO so now there is no connection

between both but eral course is also a

very good course I'm not very objective

because I wrote it but yeah if you're

interesting on that after the agent

course obviously you can

join so this question asks are we going

to see use cases that that show how to

move into production environment this

isn't something that we're really

discussing in the course um yeah we

don't have a kind of production unit but

definitely a unit that I could see and I

could definitely see it as a bonus unit

what I'd say is that the libraries that

we're using are are robust and and a use

across the industry so the transition

from these agents to Something in

production is not um is not unimaginable

right it's it's you'll be using the same

code and you'll need to consider oper

operationalizing that but um if that's

something that you really want I'd go to

the feedback form and I'd mention it and

we'll definitely introduce a bonus unit

on that there's a few people at hugging

face that I think would really enjoy

making a bonus unit like that so um yeah

please bring that

up so the crisp question uh are we using

DPS exactly this is part of the plan at

the end of uh unit one in the final

assignment the model that is currently

in the template of the final of the unit

one final assignment your first agent is

a deeps distilled model so in in that

sense you are already using deeps as

part of unit one the deeps will come

back uh in later units

too cool so this question asks is there

any limitation for the number of

attempts that that uh you can make so no

there's no limitation in fact you could

even try to follow all of the units

right at the end end of the the period

and collaborate differently it's

completely self-paced and you can run

through those as many times as you as

you need my recommendation would be if

you do take the

test that to and you and you don't pass

would be to go back to the written

material and kind of take take another

look through that and and then try again

I wouldn't kind of try to Brute Force

the quiz you you may pass it if you've

got enough time but I I think you'd

probably only be wronging yourself

because some of the stuff we're covering

in this unit one is actually fundamental

and um the components of unit 2 rely on

it and so getting the most out of unit 2

relies on those kind of fundamental

components so I'd really encourage you

to to try and then go back and and

iterate like

that so there's a another question on

the final assignment how is it evaluated

so if you look at the GAA Benchmark

actually uh they here some ground truths

answers so there's a a question you have

a series of task to performed it is

usually either having tools like web

search having tools that uh can evaluate

like look at a local file or this kind

of thing and then at the end of multiple

steps there is a ground through sensor

and this is a kind of exact match

pairing so this is going to be the same

it's going to be evaluated from um a a

computer program uh a code that is going

to check the ground truth of the

question to uh the precomputed ground

TRS but those question are usually

pretty hard so um there's no way you you

get the maximum score on those kind of

benchmarks

so this question asks about olama but I

would just extend it out to other like

inference tools inference libraries and

platforms we use um inference apis that

are integrated into the Hub most of the

way through the the um the course and

the the reason for that is that these

are kind of consistent and things that

we can give you the most support on and

make it easier for you to ask questions

and know whether it it's you or the

model that that's creating a problem we

kind of reduce down the the number of

constants so that we can just make it

easier to support you and not have a

kind of um too many too many libraries

and too many variables in the learning

process like a llama Or BLM these kind

of other Alternatives that said all of

the Frameworks that we're using have

really wide selection of inference

options and so you can kind of follow

along the course and learn the

fundamentals and learn about agents and

then you can go away to those Frameworks

and learn about their

Integrations and you'll also be able to

swap between models on different

inference platforms so you'll be able to

say Okay I I used small lm2 via hugging

face inference endpoints and now I'm

going to use that same model via ol and

run it locally and you'll be able to

swap those out or you'll be able to

switch over to proprietary apis like

open AI or ropic or Google and and try

your agents out in that way but uh the

course keeps it kind of simple so that

we can help each other the

best so should we have knowledge of

machine learning fundamentals like by

propagation stochastic calculus soft Max

and so on um this is a nice to have

obviously uh but the fundamental in unit

one easier to normally um even out the

knowledge of

everyone cool um this is a similar

question to the last one do we code on

our devices or are there a Sandbox so

the way that the course has been built

is that are many of the units use spaces

on the hubs and not and notebooks there

um we also have spaces and you can spin

up your own notebooks you will'll give

you template spaces that you can

duplicate and try out in different ways

so actually um you don't need to do that

kind of uh that much environment

Management on your own machine and if

you've got say you know a Windows

machine or or a Chromebook or these kind

of um various machines as long as you

can connect to a browser and you can get

it to spaces you'll be able to go quite

far with the course and go along with

the coding test

so maybe related question how exactly do

we use llm with agents so do you need

external API um no this do we have

something called the SS API where in

which you can already call some of the

most trending uh model directly from the

hugging face environment so this is what

we use um and in any case we can also

provide some inference end point with

the model deployed for the content of

this

course so you don't have nothing to pay

to follow this

course so um this question asks about

the field of llms and and if there are

any kind of prerequisites that that

would support this course I would say

that the agent course comes in right

after

the I think um part three or four of the

NLP course where you start to talk about

tokenizers and using LL using in that in

that course models for inference it

comes in right after that and it

actually carries on explaining

tokenization and chat templates and

inference it it leads on really well

from the NLP course there's some

material in that NP course that aren't

relevant here like for example fine

tuning models and these kind of things

we won't do any of that in the agents

course so you can take the that certain

part of the NLP course and follow on

into this one and they they work

together quite well I think and they'll

just give you a bit of um background

knowledge that said if you've been using

llms for the last year or so you have a

basic understanding of tokenization

basic understanding of generation we're

going to build on top of that basic

understanding with a bit more

information and then you can kind of go

from

there so does this class as agent

security related content um I would say

at the moment not directly we will grasp

a little bit of that on when we talk

about Frameworks for instance code agent

are in a sense very free uh because they

can code their own tool and they can

then call them uh while Json agent are

less free and we have some some

framework that are very free like small

agent on one end of the spectrum and

some framework that are very

deterministic like L graph on the other

end of the spectrum so this is where we

will tackle a little bit in agent

security uh which is related to going to

production you you want something secure

if you if you need uh your agent to

actually do something

um something meaningful so this is in

that context that we will talk about

it this is another question that I I I

just find really interesting even though

it's quite a simple question can I use

JavaScript instead of python the answer

is no for this for this course in its

current format but uh it personally is

something that I think we could

definitely work on and it would

definitely be cool if this course was

was rebuilt in typescript and JavaScript

so that more people could use it we're

dependent on the libraries that we're

using but there are libraries in in

other languages that are really

compatible so it's one of those things

again if this is something that's really

important to you and it makes total

sense to build agents in JavaScript

there's no kind of python specific

attributes that we really want for

agents then um I would start to talk

about that on Discord I'd start to bring

it up in the form and maybe kind of

Build Together a sort of JavaScript

agents Community which I would

definitely love to become the ambassador

of and uh so yeah if that's something

that you want let's let's talk about it

but right now it's not something that's

in the course and we've already got a

python implementation

unfortunately um I think there is two I

can answer very rapidly uh was one about

um um if we're going to do lives and in

uh and when yes we think to do more

lives during the course uh we keep you

updated with the email and Discord uh we

try to find an hour that fit for most of

people because you know as I said it's

very complex since you all distributed

around the world so we will keep you upd

on that and there was another about you

know how how this course you know is not

going to be obsolete um the idea of

course at huging face is that there is a

first version that is made by us and by

the community and then we try you know

people can contribute and we current we

always try to update based on what's

Happening you know agent is a very fast

topic you know in term of things are

moving fast so yeah for for at least the

next month next TR will be updated and

then you know it's an open source course

so IDE is that the community can always

partici ipate and one last for my from

my side is can we get Hing face agent

swag stuff like that uh we don't promise

anything for no but we are thinking

about you know stickers and stuff like

that but again it's not decided yet but

we would love to do that we also uh I we

made some illustration and stuff like

that that you can already share um and

so we also on Discord we want to create

like stickers and stuff like that that

you can put you know on your message uh

but yeah not everything is is already

you know decided on that but yeah there

is some things that we're working

on and I think we have time for at least

one or two question I think and for the

over if um we we do apologize if we

don't have time to answer your question

but you can ask us during the next weeks

like today tomorrow Etc on agent cour

course question and also in the upcoming

uh um lives

so maybe this one last question oh sorry

yeah go

ahead one last question I noticed was is

it possible to finish before the

deadline um in real terms no because we

um we'll release that last unit and that

last um assignment and that will lead up

to the deadline so there's a certain

amount of leeway if you worked faster on

that very last assignment and you'd be

able to finish before then but uh you

wouldn't be able to kind of do the whole

course in the next week you'd need to

wait for the the next few units to be

released and once that last units

released you could go fast from there

and finish early and the reason for that

as I said before is that we're kind of

maintaining a community and a cohort

people that are working through it

together and last question on my side so

what happens is somebody fails the test

can he she try again uh and does he she

gets the certificate anyway so for this

question there's going to be two

certificate one for the fundamental

which is on quiz uh that quiz can be

taken again as many time as you wish

until you meet the minimum

criteria uh of

80% and so you you can take it again and

again uh if you fail it there's no

problem and on the final assignment

which is going to be code related where

you going to have to send your result

result to a leaderboard we're going to

have a passing grade let's say um

there's 50 question you need to at least

have 25 out of 50 correct um the this is

the passing grade and you can also

retest As Long As You Wish until you you

meet that certain criteria so there's no

no failing

basically and the one thing I would add

there is that for that final quiz it's

your agent that needs to answer those

questions and you so you can try as many

times as you like yeah

exactly um so let's move on for the um

surprise I let Ben

explain okay so this comes from our

colleagues at at gradio and and the

small agents Library which you'll be

familiar with you'll use both these

libraries in the course and this and

it's a it's a hackathon it's a live and

and partially hybrid event we're told

and it will use all of the skills that

you're going to build up on the course

it's that last week of March the date

shown on the slide there one of those

weekends and so this is a great moment

for you to apply these skills you will

already be familiar with both of these

tools and building agents for them and

so this is a chance to kind of um yeah

use it practice one thing I would really

encourage is you to come together as

students and and we'll kind of support

that and if there were a whole series of

small uh sorry agent course teams that

you work together on and went to this

hackathon that would be really cool um

and it would be cool to see you on the

leaderboard for that

hackathon yeah and we will give you more

information in the upcoming days and

weeks um don't so don't forget to up to

Discord and to the mail to get the

latest

information cool and then this is just a

recap really of everything that we've

said at the beginning and through all

these questions just before we close so

if you need to stay up today uh you

don't want to miss anything as the

course moves forward the first thing to

do would be to follow the organization

on the Hub the way that HF works the way

that the Hub works is that any kind of

changes whether they're bug fixes

comments anything you'll you'll get um

you'll get notifications you'll see that

on your on your feed and so you'll know

is of course evolved especially when the

new units come out uh and it's also a

nice way of kind of getting an inside

line on some of the material I noticed a

few students were working on the quiz

kind of before the material came out so

if you want the inside line uh

definitely follow the or on the Hub as

you said a few times join the Discord if

you want to uh discuss things and speak

to students also studying

another way of getting the inside line

on our material is to Star the GitHub

repo and then you'll get notifications

about our pool request so you kind of

see the material even before it's

published uh and that's the best way of

getting that um that inside

information if you're already working on

the

material and you have questions kind of

about the the subject about agents I

would first go to

Discord start a thread if you like if

you see an interesting comment there

already and um start to ask questions

I've noticed that a number of students

are are very engaged and are kind of

taking the conversation up and we're

doing our best to join in um as many

conversations as we can so I would go

there if you um have kind of technical

issues like you're not sure if you know

the the application is broken or is your

code wrong and you really just want help

with errors and bugs and things like

this I would take those to the hub and

to the spaces and so that way the

various Builders of those tools can help

you right there it just makes more sense

instead of stuff getting kind of lost in

Discord so I would go to the hub for

those kind of

things and then going forward what

should you do straight after this uh

live session well um if you've already

finished the material let's say

I would go and get your certification so

I go and take the quiz and then do your

certification and if you want you can

you can share that um it's already got a

link so you can add it to your LinkedIn

and stuff um and then I would look out

for the new units coming if you followed

us in on GitHub and and on the Hub then

then you'll get notifications about that

and then I would also look out for more

live sessions like this one and like the

kind that we've discussed during the

session just to yeah so we can see you

again and and that's everything from the

recap yeah thank you very much um thank

you all for being here uh we are super

happy uh I think both of us didn't

thought that we will get this amount of

signups we are about 74k signup which is

like crazy and amazing at the same time

um yeah so thank you very much for this

live um we'll see you next time and uh

yeah keep learning and stay

awesome bye thank you very much bye

everyone