►
Description
Presented by: Jeff Boudier
This session is designed for software developers to learn how they can leverage GitHub and Hugging Face to build machine learning-powered features without a Ph.D. in machine learning. Jeff Boudier will walk the audience through optimal workflows building cool AI features leveraging code, models, and datasets in GitHub and Hugging Face, with examples in Computer Vision, Speech Processing, and Natural Language Processing.
As always, feel free to leave us a comment below and don't forget to subscribe: http://bit.ly/subgithub
Thanks!
Connect with us.
Facebook: http://fb.com/github
Twitter: http://twitter.com/github
LinkedIn: http://linkedin.com/company/github
About GitHub
GitHub is the best place to share code with friends, co-workers, classmates, and complete strangers. Millions of people use GitHub to build amazing things together. For more info, go to http://github.com
A
A
On
the
product
team
at
hiking,
face
and
today,
I'm
really
excited
to
tell
you
about
how
you,
as
a
developer,
can
take
advantage
of
all
the
latest
and
greatest
in
machine
learning
with
hugging
face
and
GitHub,
and
thank
you
very
much
to
the
GitHub
team
for
inviting
me
to
the
stage
today.
If
there
is
anything
that
I'm
going
to
be
talking
about
that,
you
have
questions
about,
don't
hesitate
to
send
me
an
email
I
am
at
Jeff
huggingface.com,
so
looking
forward
to
it
in
this
session,
I
want
to
have
two
main
parts
in
the
first.
A
Well,
let's
get
started
and
before
we
get
in
the
thick
of
it,
I
want
to
start
by
asking
you
sort
of
a
trivia
question,
which
is
what
do
the
Tesla
autopilot
in
Gmail
and
Facebook
and
Bing
all
have
in
common
and
well.
They
have
many
things
in
common,
but
the
one
thing
I'm
interested
in
is
the
Transformer
machine
learning
architecture.
It's
a
Transformer
model.
That's
the
Tesla
autopilot
uses
to
make
decisions
and
drive
the
car.
It's
a
Transformer
model
that
tries
to
complete
your
sentences
when
you're
typing
an
email
in
Gmail.
A
It's
a
Transformer
model
that
translates
automatically
posts
from
your
friends
from
every
corner
of
the
world
into
your
own
language
or
any
other
language,
and
it's
a
Transformer
that
gives
you
answers.
As
you
ask
questions
in
natural
language
in
search
engines
like
Bing
or
Google
and
there's
another
thing
that
all
these
experiences
have
in
common
is
that
each
of
them
is
being
powered
by
billions
of
Transformer
model
predictions
inferences
every
single
day
and
that's
a
huge
technical
feat.
A
We
have
a
deep
belief
in
hiking
phase,
which
is
that
machine
learning
is
becoming
the
default
way
to
build
new
technology
and
whether
mean
in
practice
well
say
your
product
manager
wants
to
create
a
new
feature
in
their
app.
The
old
way
of
doing
things
would
have
been
to
write
a
thousand
lines
of
code
and
see
how
that
works
with
machine
learning.
The
approach
is
to
find
a
machine
learning
model
that
does
a
decent
job
at
solving
the
problem
and
then
going
from
there.
A
A
Ai
is
going
to
be
eating
software
and
finally,
those
convolutional
networks.
Those
recurring
neural
networks
were
used
by
companies
applying
them
to
real
use
cases
in
production
to
do
powerful
things.
So
how
did
that
come
to
be
well?
The
way
that
it
started
is
that
neural
networks
were
basically
resurrected
from
the
research
lab
through
a
cheap
compute,
so
that
companies
could
apply
massive
amount
of
data
to
Crunch
into
accurate
models,
and
in
order
to
do
that,
work.
A
Well,
you
needed
to
use
what
I
would
call
Expert
tools,
tools
that
require
a
deep
understanding
of
the
deep
learning
of
machine
learning
tools
like
a
tensorflow
or
torch,
and
these
tools
were
not
accessible
to
to
developer,
not
quite
developer
friendly
enough
and
so
what
a
typical
project.
What
the
typical
machine
learning
project
looked
like
is
something
a
bit
like
this,
where
a
lot
of
very
tedious
work
is
spent
to
aggregate
collect
clean
data,
prepare
data
in
order
to
feed
into
a
model
that
needs
to
be
trained.
That
needs
to
be
evaluated.
A
It
limited
the
amount
of
value
it
had
with
most
projects
being
worked
on
by
data
scientists
in
companies
all
over
the
world,
never
making
it
into
production,
which
is
the
real
measure
of
value
the
real
measure
of
impact
of
the
technology
in
the
world.
But
then
came
about
Transformers
and
with
Transformers
came
the
advance
of
deep
learning
2.0.
So
what
happened?
A
So
starting
with
Transformers,
you
can
read
it
in
the
state
of
AI
report
that
just
came
out
a
few
weeks
ago,
Transformers
have
emerged
as
a
general
purpose
architecture
for
ML,
meaning
you
can
apply
for
any
kind
of
machine
learning
problem
you're
facing
and
the
resulting
Trends
in
usage
with
Transformers
usage
sharply
going
up
and
taking
over
what
had
been
traditionally
the
job
of
recurring
or
convolutional
neural
networks.
So
a
Transformers
architecture,
but
one
of
the
main
and
I
think
the
most
impactful
idea
that
came
out
of
the
attention
is
all
you
need.
A
So
that's
transfer,
learning
and
that's
very
powerful
is
the
ability
to
build
upon
each
other's
work
within
the
community,
and
so
you
remember
that
a
long
cycle
of
development
for
a
typical
project
of
machine
learning.
Well,
today,
the
process
of
machine
learning
project
is
really
two
or
maybe
three
steps.
A
The
first
step
is
to
identify
what
machine
learning
task
corresponds
to
the
problem
you're
trying
to
solve
the
feature
you're
trying
to
build,
and
there
are
tasks,
machine
learning,
tasks
that
are
pre-built
and
available,
whether
you're
working
with
text
or
you're,
working
with
images
or
videos
or
you're
working
with
speech
or
audio
for
text.
It
can
be
sentiment,
analysis,
sequence,
classification,
summarization,
translation
for
images.
It
could
be
classifying
images
or
recognizing
objects
or
removing
a
background
for
speech.
It
can
be
transcribing
speech
as
I'm
speaking
into
text,
or
you
can
be
turning
text
into
speech.
A
It
can
be
recognizing
who's
speaking
and
so
for
any
of
those
problems.
Any
of
those
type
of
features
there
are
pre-trained
models
that
you
can
find
that
are
working
great
at
this
particular
task,
and
so
once
you
have
the
task,
then
it's
easier
to
pick
the
pre-trained
model
that
will
give
you
state
of
the
art
performance,
meaning
that's
the
best
possible
accuracy.
You
can
get
in
the
model
predictions
and
you
don't
need
to
do
the
whole
cycle
of
gathering
and
cleaning
and
prepping
huge
data
sets
to
create
your
own
model
from
scratch.
A
In
fact,
it's
only
two
lines
of
code
to
take
an
out
of
the
box
pre-trained
model
from
a
hub
like
the
hiding
face
Hub,
and
then,
if
you
think
that
for
your
particular
data
you
can
have
even
better
accuracy.
Then
that's
the
fine
tuning
process
and
you
only
need
a
little
bit
of
labeled
data
in
order
to
take
the
pre-trained
model.
That
does
already
a
pretty
good
job
into
a
fine-tuned
models
that
that
does
a
great
job
on
your
specific
problem
and
data.
A
So
that's
the
new
process
of
transfer
learning
and
what
we've
seen
what
we've
seen
in
the
State
of
the
State
of
AI
report.
What
we've
seen
in
the
kaggle
survey
is
that
Transformers
are
now
eating
deep
learning
and
we're
seeing
that
the
general
purpose
architecture
of
Transformers
can
be
applied
to
any
kind
of
machine
learning,
problem
from
NLP
to
computer
vision,
to
speech,
to
reinforcement,
learning,
Etc
and
then
every
single
sector
of
the
economy
is
starting
to
adopt
those
models
in
order
to
build
AI
powered
features.
A
So,
in
this
whole
context
in
this
whole
Revolution,
what
is
the
role
of
hiding
phase?
Well,
our
mission
is
to
democratize
good
machine
learning
to
make
the
latest
and
greatest
in
machine
learning
built
from
ethics.
First
principles
accessible
to
as
many
people
and
as
many
companies
in
the
world
and
a
characteristic
way
in
everything
that
we
do
at
hagging
face.
Is
we
want
to
do
everything
in
a
Community
Driven
way?
A
You
can
see
that
the
rate
of
growth
in
the
last
three
years
from
0
to
70,
000
Stars,
has
been
staggering,
and
today
there
are
over
eight
Millions
downloads
of
the
Transformers
Library
happening
every
month,
and
so
Transformers
and
hugging
face
are
becoming
the
de
facto
way
through
which
companies
all
over
the
world
are
applying
machine
learning
to
their
business.
And,
of
course,
I
talked
about
natural
language
processing.
Nlp
dealing
with
text
because
that's
where
Transformers
come
from,
that's
where
they
were
first
applied,
but
today
on
the
Hub,
you
can
find
models.
A
You
can
solve
problems
in
computer
vision
in
speech
with
biomedical
data,
drug
Discovery
protein
with
reinforcement,
learning
with
time-based
series
data.
Any
kind
of
problem
can
be
solved
through
machine
learning
and
Transformers,
and
so
the
end
result
is
that
today,
on
the
hugging
phase
Hub
on
huggingface.com,
where
you
can
find
all
these
models
where
you
can
find
all
these
data
sets.
We
see
over
a
hundred
thousand
people
coming
every
single
day
and
downloading
over
a
million
models
every
single
day
and
I
talked
a
lot
about.
A
Transformers
and
Transformers
are
cool,
but
the
coolest
new
kit
on
the
Block
is
diffusers
I'm
sure
you've
seen
on
your
Twitter
on
Reddit,
these
cool
images
that
are
created
by
just
adding
a
prompt
with
a
stable
diffusion
demo.
Diffusers
can
be
applied
to
generate
images,
to
generate
videos
to
generate
music
sound
and
it's
only
the
beginning
and
just
like
with
Transformers.
If
you
want
to
access
any
of
those
state-of-the-art
models,
just
like
with
Transformers
with
the
diffusers,
a
library
from
a
hugging
face,
you
can
do
so
as
well.
A
Whichever
framework
you're
working
on
whichever
model
architecture
you
want
to
access,
whichever
model
checkpoint,
you
want
to
use
all
right.
So
I
gave
you
a
bunch
of
context
on
like
what
happened
over
the
last
two
and
three
years
with
machine
learning
and
why
it's
able
to
do
all
the
things
that
it's
able
to
do
now,
and
now
is
the
time
for
me
to
tell
you
how
you
can
use
all
this
stuff
using
tools
that
are
available
today,
taking
advantage
of
GitHub
of
the
hugging
face,
Hub
to
build
machine
learning
into
your
own
applications.
A
So
the
good
news
is
this:
the
good
news
is
that
with
GitHub
and
with
hikingface,
you
have
everything
you
need
to
build.
Machine
learning
part
features,
and
the
other
good
news
is
that,
as
you
know
how
to
work,
how
to
build
with
GitHub,
you
already
know
how
to
work
and
how
to
build
with
hiding
face
so
that
you
can
build
and
maintain
and
collaborate
around
your
code
with
GitHub,
and
so
you
can
build
and
you
can
collaborate
around
your
models.
A
Your
data
sets
around
your
demo
apps
with
the
hugging
face
hub
and
at
the
end
of
the
day,
for
you
all
of
this
are
git
repos,
which
you
can
use
very
easily
already,
as
you
know
how
to
work
with
code
all
right.
So
what
can
you
do
with
the
hugging
face?
Hub?
How
do
people
go
about
building
machine
learning
features
with
hiding
face?
The
first
thing
that
you
can
do
is
explore
all
these
models
and
data
sets
that
are
free
and
accessible
to
you.
A
I
told
you
that
are
about
over
a
hundred
thousand
of
them
that
are
free
and
accessible.
If
you
go
to
huggingface.co
today,
and
you
go
to
look
at
models,
you
will
see
about
80
000
different
models
that
you
can
apply
to
your
business
problem.
How
do
you
go
to
sort
through
that
Jungle
of
models?
Well,
you
can
very
quickly
and
easily
filter
them
down
to
the
problem
that
you
want
to
solve.
A
As
I
said
earlier,
the
new
way
of
doing
machine
learning
first
thing
to
identify
the
machine
learning
task
that
you
want
to
apply,
and
you
can
do
that
very
easily
on
the
hiking
face
Hub
by
just
selecting
the
task,
and
then
you
can
select
the
language
that
you
want
to
work
in
the
framework
that
you
want
to
work
in
and
quickly
narrow
down
to
the
set
of
pre-trained
models
that
you
can
use
in
your
application.
So
now
you've
explored
the
hiking
face
Hub
and
you
you
found.
A
A
It's
an
incredible
model
to
do
automatic
speech,
recognition,
which
is
how
you
transcribe
speech
into
text,
and
so
what
you're
looking
at
right
now
is
the
model
card
for
the
whisper
large
model
of
open
Ai,
and
it
gives
you
all
the
information
that
you
need
to
know
about
the
model
how
it
was
built
which
paper
it
comes
from.
What
are
the
possible
risks
associated
with
that
model?
The
performance
of
that
model?
You
can
even
try
it
right
from
the
page
and
because
everything
on
the
hiking
face
Hub
is
essentially
under
the
hood,
a
git
repo.
A
You
also
have
access
to
all
of
the
repo
contents,
and
within
that
whisper
large
repository,
you
can
find
every
single
file
that
is
part
of
it
and
for
the
whole
repo.
You
can
look
at
the
full
history
of
the
repo,
its
contributions
over
time
for
every
country
distribution.
There
is
a
unique
commit,
ID
that
you
can
use
and
refer
to
to
have
clear
version
control
in
how
you're
integrating
your
model
and,
of
course,
for
each
of
those
files.
A
You
have
the
full
history
and
the
content
of
the
file
that
you
can
explore
right
there
in
place
from
The
Hub,
but
of
course
you
want
to
interact
with
this
with
code.
So
how
do
you
do?
You
can
simply
git
clone
at
the
repository?
So
you
have
access
on
your
local
machine.
You
can
get
ad
and
git
commit
and
git
push
to
contribute
to
the
model.
A
So
I
showed
you
how
you
can
explore
models
on
a
hiding
face,
Hub
how
you
can
manage
your
own
models
with
a
hugging
face,
Hub,
taking
advantage
of
Version
Control,
taking
advantage
of
git
based
workflows.
The
next
thing
that
you
can
do
once
you
have
a
model,
is
to
collaborate
around
that
model
and
to
collaborate
around
that
model.
A
When
you
can
see
the
changes
as
they
happen,
you
have
the
ability
to
create
discussions
to
engage
and
ask
questions
to
the
model
authors.
What
is
the
license
of
this
model?
Have
you
thought
about
adding
this
feature?
Etc
everything
you're
used
to
doing
on
GitHub
you
can
do
to
on
the
hyping
phase
Hub
to
interact
around
the
models
the
data
sets
and
the
demo
apps.
So
we
talked
about
how
you
can
explore
models,
how
you
can
manage
your
own
models,
how
you
can
collaborate
around
new
models
now?
A
The
next
thing
is,
how
do
you
use
these
things
and
how
you
use
those
models
where
there
is
a
very
useful
abstraction
on
a
hugging
face
called
tasks
and,
as
I
told
you
at
the
beginning,
the
first
thing
you
want
to
do
is
to
identify
the
machine
learning
task
that
can
solve
your
problem
and
on
hiking
phase
there
are
a
ton,
a
different
tasks
for
very
different
kind
of
modalities.
So
here
you
can
see
tasks
for
natural
language
processing
for
text.
A
You
can
easily
summarize
translate
classify
analyze
the
sentiment
of
text,
but
there
are
also
tasks
for
audio.
There
are
also
tasks
for
computer
vision.
There
are
also
tasks
for
reinforcement,
learning
and
Etc,
but
what
do
those
tasks
mean
in
practice
for
you
in
code?
Well
in
the
Transformers
Library
there's
a
class
called
pipeline
in
a
Transformer
pipeline
is
the
implementation
of
a
machine
learning
task
so
that
you
don't
have
to
worry
about
the
tokenizer
and
the
token
level
things
and
how
to
pre-process
your
input
and
how
to
post-process
your
output.
A
Everything
is
encapsulated
and
abstracted
at
the
task
level
through
the
pipeline
and
so
using
a
machine
learning
model.
A
pre-trained
model
that
you
find
on
a
hiking
face,
Hub
that
will
solve
your
problem,
and
here
it's
a
Facebook
model.
It's
the
bark
mnli
modeled.
In
order
to
do
zero
shot
classification,
it's
only
really
a
couple
lines
of
code
right,
so
you've,
pipped
install
Transformer
you
pulled
pipeline.
Now
you
define
that
object
referring
to
the
model
checkpoint.
Referring
to
the
task
that
you
want
to
apply,
you
send
an
input.
A
You
get
the
output
back,
so
I
told
you
about
how
you
can
manage
models,
how
you
can
collaborate
around
models,
I
told
you
how
you
can
use
tasks
to
use
those
models
to
solve
your
problems
now.
The
next
step
is
what
what,
if
you
want
to
adapt
that
model
to
have
even
better
accuracy
to
your
specific
sort
of
business
problem,
your
specific
data
and
that's
the
training?
A
Well,
the
good
news
is
that
we
built
a
no
code
tool
so
that
with
zero
knowledge
of
machine
learning,
but
just
a
few
examples
of
labeled
data,
you
can
go
from
zero
to
a
state-of-the-art
model,
and
this
is
what
creating
a
new
Auto
Train
project
looks
like
you
can
select
the
number
of
model
candidates
that
you
want
to
start
with.
You
can
upload
your
label
training
data.
It
can
be
for
natural
language
processing
tasks,
it
can
be
for
image,
classification,
computer
vision,
tasks,
tabular
data
and
everything
just
happens
automatically.
A
A
In
your
particular
context,
now
I
want
to
talk
to
you
about
how,
now
that
you
have
a
model,
how
you
can
demo
it
to
the
community,
how
you
can
show
to
people
what
it
can
do
and
to
do
that,
you
want
to
create
a
great
demo,
app
experience
and
that's
what
the
Spacey
service
on
hiking
face.
Hub
allows
you
to
do
so
here
is
pictured
a
space,
a
machine
learning
app
that
is
called
Huggy.
That
is
built
to
let
you
interact
with
an
agent.
A
That's
been
built
using
the
hiding
face,
reinforcement,
learning
tools
so
with
Huggy
you
can
interact
with
the
agent.
Everything
is
running.
Everything
is
hosted
within
within
this
webpage
that
you
created
as
a
space
and
every
single
week.
We
feature
some
of
the
most
interesting
spaces.
Some
of
the
most
interesting
machine
learning
demo
apps
created
by
the
community-
you
can
see
this
week
in
addition
to
Huggy,
you
have
an
amazing,
stable
diffusion
powered
out,
painting
experience.
A
You
have
great
natural
language
processing
applications
to
summarize
text,
to
give
a
positive
pin
to
any
text
that
you
enter
so
really
cool
demo
apps
that
you
can
try
and
are
all
powered
by
spaces
and
to
talk
to
you
about
spaces.
I
want
to
ask
Nate
to
walk
you
through
how
it
works
and
how
you
can
use.
B
The
team
here
at
hugging
face
is
super
excited
to
announce
the
public
release
of
spaces,
the
quickest
and
easiest
way
to
build,
deploy
and
share
Machine
learning
demo
apps
machine
learning
projects
come
in
all
sorts
of
different
flavors.
We
build
spaces
because
we
wanted
to
provide
a
more
flexible
solution
for
folks
to
share
their
projects
with
spaces.
You
can
take
any
existing
code.
You
might
have
build
an
app
around
it
in
minutes
and
share
it
instantly
on
our
site.
All
for
free,
let's
see
just
how
easy
it
is
to
create
a
new
space.
B
First
make
sure
you're
logged
into
your
hugging
face
account
then
simply
click
your
user
icon
in
the
top
corner
and
click
new
space,
Here's
Where.
We
can
name
our
project
and
choose
one
of
the
supported
open
source
libraries
we'd
like
to
build
our
app
with
right.
Now
you
can
select
either
streamlit
or
gradio,
we'll
name
our
app
demo
app
and
choose
streamlit.
For
the
sake
of
this
demo,
once
we're
ready,
we
can
click
create
space
to
create
a
new
project.
B
Now
this
project
is
actually
a
git
repo,
so
we
could
clone
it
locally
to
edit
the
files
if
we
wanted
to,
we
also
have
the
option
to
just
add
and
edit
the
files
in
the
browser.
So
that's
what
we'll
do
for
now.
Let's
add
a
requirements.txt
file
to
our
repo,
which
is
how
we
can
Define
our
app's
python
dependencies
to
do
that,
we'll
navigate
to
the
files
and
versions
Tab
and
create
a
new
file
using
the
drop
down
on
the
right,
we'll
name,
the
file,
requirements.txt
and
add
a
couple
dependencies
in
this
case.
B
Pytorch
and
hugging
faces
Transformers
when
we're
done
we'll
hit,
commit
new
file
to
add
it
to
our
spaces,
repo
to
create
the
app
itself.
We'll.
Add
another
new
file
named
app.pi
in
this
file,
we'll
quickly
Define
a
simple
streamlit
app
that
uses
the
default
sentiment.
Analysis
pipeline
from
Transformers
to
predict
how
positive
or
negative
a
given
input
text
is
just
like
before,
when
we're
done.
B
We'll
hit
commit
new
file
to
add
that
application
file
to
our
repo
in
just
a
few
lines
of
code
were
finished,
and
we
can
view
our
running
app
from
the
app
tab.
If
we
choose
to
make
our
space
public,
we
can
easily
share
our
app
with
anyone
around
the
world
by
sharing
the
link
to
the
space
in
this
video,
we
saw
just
how
easy
it
is
to
build,
deploy
and
share
Machine
learning
demo
apps
with
hugging
face
spaces.
B
A
Watching
thanks
Nate,
so
that's
spaces
on
the
hugging
face
Hut
and
just
a
few
days
ago
we
released
an
exciting
new
feature
that
you
can
use
today,
which
is
the
ability
for
you,
as
you
created
the
space
to
upgrade
the
underlying
Hardware,
the
underlying
infrastructure
that
powers
your
space
and
so
very
easily.
You
can
select
a
higher
CPU
infrastructure
or
a
variety
of
GPU
infrastructures
to
provide
you
even
better
experience
to
people
who
are
going
to
try
your
space
app.
A
A
Here's
the
best
part
how
you
can
actually
deploy
it,
so
you
can
use
it
in
your
application
at
scale
very
easily,
and
so
to
do
that
we
created
a
new
service,
called
hugging,
face,
inference
endpoints
and
when
you
go
to
deploy
a
transfer
former
model
or
a
machine
learning
model,
you
very
quickly
face
a
whole
set
of
new
types
of
problems
with
gpus
and
dependencies
and
Dockers
and
kubernetes.
So
it's
it's!
It's
a
big
Endeavor
to
try
to
take
a
model
and
turn
that
into
a
service.
That's
going
to
scale!
That's
going
to
be
robust!
A
That's
what
hiding
face
inference!
Endpoints
do
for
you.
We
basically
made
machine
learning
models
completely
Plug
and
Play,
for
you
pick
the
model
get
the
service
integrate
into
your
app
here's,
how
it
works.
First,
you
select
the
model
that
you
want
to
deploy.
Then
you
select
the
cloud
provider
that
you
want
to
use
today.
It's
can
be
AWS
or
Azure.
A
You
select
the
region
that
you
want
to
use
the
closer
to
your
data,
the
better
and
then
you
can
even
select
the
type
of
instance,
the
type
of
compute
that
you
want
to
use
the
auto
scaling
logic
so
that
you
can
scale
horizontally
your
model
and
the
security
setting
that
you
want,
so
that
you
can
go
from
a
public
endpoint
all
the
way
to
a
secure
private
endpoints
that
uses
a
private
link
under
the
hood
to
build
a
direct
connection
to
your
VPC.
So
that's
anybody,
but
your
VPC
can
connect
to
the
endpoint.
A
Nobody
on
the
internet
can
connect
to
your
endpoints.
That's
where
the
secure
endpoints
does,
once
you
created
your
endpoint,
it's
available
to
you,
it's
dedicated
to
you.
If
it's
protected,
only
you
or
people
in
your
organization
can
have
access
to
it
and
you
have
easy
access
to
logs
to
monitoring
so
that
you
can
manage
your
application
and
seat
scale
up
and
down
and
update
your
endpoints.
As
you
update
your
models
and
so
very
easily,
you
can
create
a
fleet
of
models
that
are
working
for
you
without
having
to
deal
with
any
infrastructure.
A
So
today,
I've
told
you
what
happened
in
machine
learning,
what
it
means
for
companies
and
I
walked
you
through
what
you
can
do
using
the
tools
that
are
available
today
to
take
advantage
of
all
these
new
possibilities
that
machine
learning
models
offer
to
you.
So
you
can
manage
and
collaborate
with
the
Hub
around
your
own
models.
So
you
can
use
these
models
using
hiking
face
tasks,
so
you
can
even
improve
upon
them
with
Auto
Train.