►
From YouTube: GitLab ModelOps Stage - Direction Walkthrough
Description
Please note, since this video was recorded "GitLab Learn stage" was renamed "Gitlab ModelOps stage"
Watch Sr. Director of Product David DeSanto and Principal Product Manager Taylor McCaslin discuss an overview of the Gitlab Learn stage, dig into the focus for each group, and answer common questions about this new area of investment for GitLab.
Links:
GitLab Learn Direction - https://about.gitlab.com/direction/modelops/
DataOps Direction - https://about.gitlab.com/direction/modelops/dataops/
MLOps Direction - https://about.gitlab.com/direction/modelops/mlops/
Applied ML Direction - https://about.gitlab.com/direction/modelops/applied_ml/
Join us in GitLab slack:
g_machine-learning: https://gitlab.slack.com/archives/C01ESHPNHS9
s_learn: https://gitlab.slack.com/archives/C01F1R4J7FW
A
A
Another
area
of
focus
is
ml
ops.
Sometimes
it's
referred
to
as
model
ops
and
by
the
way,
this
is
different
than
ai
ops,
which
I'll
touch
about
on
here
in
a
minute.
But
this
is
very
much
focused
on
enabling
customers
to
be
able
to
build
their
own
models,
as
well
as
maintain
manage
deploy,
partial
recall
and
all
the
other
components.
You
need
to
do
to
be
able
to
take
your
ml
models
to
production.
A
A
A
Correlating
events
back
to
determine
if
they're
legitimate
moves
by
a
valid
user,
as
well
as
getting
things
like
spam
detection,
abuse,
detection
and
so
forth,
and
you'll
see
a
lot
of
that
getting
highlighted
both
in
recent
keynotes
and
group
conversations
with
the
security
team,
intelligent
code
security.
It's
also
part
of
that.
This
is
really
about
us
trying
to
reduce
the
noise.
People
see
whether
that
is
from
security
scans
or
build
logs,
as
well
as
building
other
automated
recommendation
engines.
This
could
be
true
for
security.
A
A
B
Awesome
thanks
david.
Basically,
what
we're
going
to
talk
about
quickly
is
an
overview
of
all
of
the
different
areas
within
gitlab
learn,
so
the
focus
of
gitlab
learn
really
is
to
empower
and
enable
data
science
workloads
on
top
of
gitlab,
and
what
that
basically
means
is
today.
You've
got
lots
of
data
in
organizations.
B
That
is
really
what
we're
trying
to
help
enable
is
to
connect
the
data
that
organizations
have
with
the
actual
intelligence
and
machine
learning
that
we're
now
seeing
emerging
in
the
software
development
space
and
both
of
those
have
separate
problems.
We
already
know
this
with
data
ops
with
meltano.
It's
a
large
focus
space.
There's
tons
of
competitors,
notably
like
snowflake,
is
a
great
example
of
a
company
that
does
etl
well
data.
Ops
really
is
about
those
three
components
of
etl:
extract
load
and
transform
gathering
all
of
the
data
from
the
various
disparate
sources.
B
Next
we've
got
machine
learning
mlops,
so
this
is
really
about
what
you
actually
go
and
do
with
that
data.
How
do
you
extract
value
out
of
it
and
how
do
you
deploy
those
machine,
learning
and
artificial
intelligence
models
to
production
today?
This
is
a
really
messy
space.
It's
relatively
new.
There
are
tons
of
tools
and
frameworks
and
open
source
projects
that
machine
learning
experts
use
to
try
to
extract
value
out
of
data.
B
It
is
called
data
science.
For
a
reason,
it
is
very
much
a
scientific
effort
where
data
scientists
are
running
experiments,
trying
to
extract
intelligence
out
of
the
various
pieces
of
data.
They've
got
and
building
models,
training
models
and
deploying
them
into
production,
environments
and
part
of
the
struggle
today
with
ml.
Ops
is
you've,
got
data
scientists
who
are
very
highly
skilled
and
unique
individuals,
but
they
generally
don't
have
a
traditional
software
experience.
B
In
many
cases,
organizations
don't
actually
have
the
infrastructure,
they
don't
have
the
experience
they
don't
have
all
of
the
data
science
tool
chains
running
in
a
production
environment.
So
just
deploying
those
models
is
very
difficult.
That
is
really
what
ml
ops
is
focused
on
is
connecting
machine
learning
experts
with
software
engineering
to
make
that
experience,
something
that
you
can
actually
deploy
models
very
quickly.
This
disjointedness
of
experience,
tool,
chains
and
software
makes
it
very
slow
and
expensive
to
deploy
production
models
if
you
can
even
get
it
to
production
at
all.
B
So
this
is
a
sort
of
a
fundamental
piece
of
making
this
happen.
One
thing
I'll
mention
with
ml
ops
as
well
is
the
partnership
angle
here,
because
there
are
so
many
different
tool,
chains
and
players
in
the
machine,
learning
and
artificial
intelligence
space.
We
are
never
going
to
do
it
all.
There
are
many
data
science
platforms
that
are
emerging
already,
where
all
of
these
technologies
are
tied.
Together,
we
want
to
partner
with
those
various
organizations
and
companies
to
help
make
the
data
science
tool
chain
just
work
with
gitlab.
B
We
don't
care
what
tools
you
decide
to
go
and
use.
We
want
you
to
use
whatever
the
best
tool
is
for
whatever
experiment
or
machine
learning
problem
you're
trying
to
solve,
and
we
want
that
to
work
seamlessly
with
gitlab
and
then
finally,
we've
got
our
applied
machine
learning
group.
So
the
idea
behind
this
applied
machine
learning
group
is
that
we
want
to
take
the
learnings
that
we
had
from
data
ops
and
from
ml
ops
and
apply
them
to
gitlab
itself,
and
I
think
what's
particularly
interesting
about
this-
is
this
is
strictly
dog
fooding
for
us.
B
When
you
think
about
anomalies
that
happen,
it's
very
hard
for
a
person
to
sit
and
watch
logs
all
day
long
and
make
decisions
based
on
them.
This
is
a
classical
machine,
learning
problem
that
you
can
put
heuristics
in
front
of
data
and
let
it
make
automated
decisions.
So
that's
kind
of
the
idea
behind
insider
threat
to
make
gitlab
smarter
and
more
secure
with
machine
learning
and
then
finally,
we've
got
intelligent
code
security,
which
is
an
area
I'm
particularly
passionate
about
with
sas.
B
So
when
you
think
about
code
security,
there's
lots
of
different
ways
to
express
vulnerabilities,
there's
lots
of
different
ways
to
detect
them.
It
is
really
a
a
problem
that
is
well
suited
for
machine
learning,
so
there
are
ways
that
we
could
reduce
false
positives,
that
we
could
detect
variants
that
we
could
detect
vulnerabilities
across
all
of
our
customers
in
a
privacy
forward
way
all
with
machine
learning
techniques.
B
So
you
can
see,
there's
lots
of
opportunities
for
us
to
enrich
git
labs
features
and
functionality
empowered
with
data
science,
and
to
do
that
we'll
need
the
learnings
from
data
ops
and
machine
in
ml
ops.
So,
overall,
that's
kind
of
how
we've
tied
this
whole
piece
together
these
different
groups
and
then
applying
them
within
git
lab.
So
we've
got
a
nice
sort
of
experience
here
where
by
us
trying
to
build
machine
learning,
features
into
git
lab.
We
can
develop
features
and
functionality
for
customers
in
the
data,
ops
and
mlaps
space,
and
vice
versa.
B
Today,
we're
already
improving
the
experience
of
running
data
science
workloads
on
top
of
git
lab.
We've
recently
exposed
gpu
runner
support,
which
was
called
out
in
our
last
release,
also
as
a
community
contribution
which
exposes
gpus
for
gitlab
runners,
so
that
you
can
run
data
science
workloads
which
depend
on
gpus
in
many
cases
we're
making
improvements
to
the
way
that
python
notebooks
are
represented
in
gitlab
and
there's
a
whole
host
of
other.
B
A
A
B
So
when
you
think
about
the
the
machine,
learning
and
artificial
intelligence
space,
it's
still
relatively
new,
but
there's
been
an
explosion
of
players
of
tool,
chains
of
frameworks
being
used
to
power,
machine
learning
workloads.
This
is
an
area
where
we're
also
seeing
this
now
being
deployed
by
customers.
It's
something
that
customers
come
to
us
and
say:
hey.
We've
got
a
machine
learning
team.
We've
got
a
group
of
data
scientists
they're
really
well
paid,
and
we
don't
really
deploy
any
of
their
things
to
production.
B
It's
this
classic
problem
that
we
were
talking
about
earlier,
where
it's
just
a
different
kind
of
technology
that
companies
are
not
used
to
deploying
to
production,
and
so
there's
a
big
opportunity
here
for
gitlab
to
help
solve
this
problem.
When
you
think
about
gitlab
as
a
platform,
we
already
did
this
with
devops
devops
was
an
area
of
tons
of
tools
that
were
all
disjointed.
We've
brought
them
together
into
a
singular
platform,
structured
around
source
code
management
and
ci
cd.
It
just
makes
sense
to
apply
a
thin
layer
of
machine
learning.
A
This
is
me,
quoting
sid.
I
think
he
was
quoting
elon
musk,
if
I
remember
correctly,
but
either
way
the
the
statement
was.
The
future
of
software
is
a
hybrid
of
code
and
nl
models,
and
you
can
see
that
today,
with
the
amount
of
effort
going
in
by
startups
investment
from
vcs
on
how
to
essentially
make
smart
code
or
make
smarter
decisions
automatically
and
to
build
up.
A
B
Sure
yeah,
this
is
an
area
where
those
specific
areas
are
classical
machine
learning
problems.
When
you
look
at
more
traditional
security
companies
that
are
only
focused
on
security,
they
have
machine
learning
groups
that
are
already
trying
to
do
to
apply
machine
learning
techniques
to
security
use
cases.
So
this
is
something
that,
like
with
user
entity,
behavior
analytics
has
been
around
for
a
number
of
years.
Now
it
makes
very
logical
sense.
We
also
have
very
structured
data
within
git
lab
in
that
area
to
be
able
to
go
and
do
this
kind
of
work.
B
B
They've
been
working
to
apply
machine
learning
to
security,
application
security
testing
for
a
number
of
years.
So
this
is
just
something
that's
very
logical
for
us
to
go
and
do
and
then,
when
you
think
about
our
automated
portfolio
management,
suite
name
to
be
determined.
This
is
an
area
where
we
again
just
have
very
structured
data
within
gitlab
that
it's
very
easy
and
they're
very
simplistic
machine
learning
models.
For
example,
a
tanooki
bot
that
stan
has
put
together.
B
That
is
going
and
labeling
issues
is
not
doing
anything
super
sophisticated
with
machine
learning,
it's
using
very,
very
basic
techniques
and
getting
a
lot
of
value
out
of
it
very
quickly.
So
that's
when
we're
looking
at
opportunities
for
us
to
apply
machine
learning
within
get
lab.
We
want
to
take
those
very
high
value
and
low
effort
and
low
complexity,
solutions
and
start
with
those,
and
that
gives
us
that
dog
fooding
opportunity
to
figure
out
what
are
the
pieces
that
we
need
for
data
ops
to
work
better?
What
are
the
pieces?
A
A
B
Yeah,
so
I
the
first
piece
would
be
to
point
you
to
our
slack
channels.
We've
got
one
set
up
for
the
stage,
so
s
underscore
learn.
We
also
have
a
group
underscore
or
g
underscore
machine
learning
where
we're
hanging
out
and
talking
about
all
the
the
different
things
that
we're
working
on
in
the
machine
learning
space
so
definitely
feel
free
to
to
jump
into
those
channels.
B
Certainly,
if
you've
got
customers
interested
in
how
gitlab
is
using
ml
or
ai
or
how
we're
enabling
customers
to
run
data
science
workloads
on
top
of
git
lab,
I'm
always
happy
to
to
chat
with
customers
we're
actively
looking
for
customers
to
explore
this
area
as
we
build
out
the
personas
the
use
cases
and
work
out
what
functionality
we
need
to
support.
So,
if
you've
got
those
customers
reach
out
to
me
and
we'll
we'll
get
you
sorted.
A
B
Yeah
first
definitely
join
those
slack
channels,
we're
there
exchanging
articles
and
talking
about
customers
we're
talking
with
if
you've
got
customers
or
if
you've
got
experience
with
machine
learning,
we
would
just
love
to
know
what
doesn't
work.
Some
of
the
the
things
that
we've
just
started.
Working
on,
like
enabling
gpu
support
in
gitlab
runner,
improving
the
python
notebook
support
are
just
very,
very
basic
things
that
anyone
trying
to
run
machine
learning
workloads
on
gitlab
just
knew
didn't
work,
the
way
that
they
needed
to
or
didn't
function
at
all.
B
So
those
little
tiny
things
that,
like
I
just
wish
it
were
a
little
bit
easier
to
do
x,
y
or
z
related
to
a
machine
learning
workload.
Those
are
things
that
we
can
easily
prioritize
early
on
to
just
help
make
all
of
this
go
easier,
and
certainly,
if
you've
got
connections
to
the
machine,
learning
and
artificial
intelligence
community.
Definitely
let
us
know
we're
chatting
with
partners
in
the
space
other
technology
companies,
so
we're
happy
to
engage
any
way
that
we
can.
A
B
A
To
kind
of
think
of
something
taylor
said
a
minute
ago,
is
that
today
this
is
being
supported
by
no
one
who
whose
full-time
job
it
is
it's
people
who
are
taking
time
so,
if
you're
interested
in
your
developer
and
even
if
you're,
not
a
developer,
but
you
want
to
commit
some
code
to
help
what
we're
trying
to
do
here
to
get
everything
kicked
off.
Please
do
reach
out
in
those
channels
and
taylor
or
stan
or
mom
completely
to
the
issues
that
we're
currently
tracking
what
we
think
we
need
to
tackle.
Next,
everyone
can
contribute.
A
We
would
love
to
get
additional
help
from
people
as
they
have
time.
So
those
are
typical
questions.
If
you
had
more
questions,
please
feel
free
to
put
them
in
the
slack
channel
for
kayla
mentioned
before.
I
asked
him
to
make
his
final
comments.
I
do
want
to
take
a
moment
to
thank
the
people.
Who've
been
working
on
this
already
so
taylor
who's,
obviously
called
here
monray,
who
is
based
out
of
the
apac
region
as
part
of
the
sa
orb
as
well
as
stan
hughes,
one
of
our
engineering
fellows.
A
It's
always
been
possible
without
their
time
and
dedication
to
making
it
possible
and,
as
taylor
said
third
tonight,
we
had
our
first
demo
related
feature
ship,
which
is
exciting,
and
that
was
in
collaboration
with
a
customer
who
wanted
to
see
us
get
nvidia
gpu
support,
so
big
thanks
to
them.
They're
the
ones
who
got
the
direction
pages
together
and
are
are
giving
their
time
on
top
of
their
current
responsibilities
where
they
can
to
help
move
this
forward.
B
Oh,
I
would
just
say
this
is
an
exciting
opportunity
for
get
lab.
It's
an
area
that
is
relatively
cutting
edge.
There's
lots
of
new
innovation
happening
in
this
space.
Get
lab
today
doesn't
really
have
anything
to
offer
here.
So
this
is
truly
a
green
field
opportunity
for
us
and
it
strikes
the
same
chords
as
the
success
that
we
found
in
the
devops
space
in
the
security
space,
and
really
it
just
puts
us
forward
as
a
thought
leader.