►
From YouTube: Sponsored Keynote: Evolving with Kubernetes: Embracing Model Ops - Steven Huels, Sr Director
Description
Don’t miss out! Join us at our next event: KubeCon + CloudNativeCon Europe 2022 in Valencia, Spain from May 17-20. Learn more at https://kubecon.io The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects.
Sponsored Keynote: Evolving with Kubernetes: Embracing Model Ops - Steven Huels, Sr Director, AI Cloud Services, Red Hat
Kubernetes changed the way in which applications are developed. AI and Machine Learning has changed what is required of an application which has driven changes in the way we put intelligent applications into production. Applying the DevOps principles to AI and Machine Learning outputs or Model Ops, can help you get your models from pilot to production safely, securely, and with repeatability.
A
A
I
don't
think
it
will
surprise
anyone
in
the
audience
today
if
I
was
to
tell
you
that
data
and
ai
workloads
are
one
of
the
fastest
growing
workloads
on
kubernetes.
Today,
of
course,
you
see
the
traditional
workloads
we're
all
used
to
deploying
so
databases,
logging,
web
servers,
but
increasingly
businesses
are
trusting
their
kubernetes
platforms
to
put
these
very
critical
workloads
into
production,
and
that
includes
everything
in
the
life
cycle
to
build
a
model.
A
First
and
foremost,
finding
the
right
set
of
resources
to
work
on
these
is
very
challenging
right
data
scientists
who
are
familiar
with
containerized
model
development
and
pipelines
that
are
then
able
to
hand
off
those
models
to
application
developers
for
integration
into
their
intelligent
applications,
can
be
a
challenging
workflow
and
then
having
I.t
administrators
and
kubernetes
administrators
who
are
able
to
support
those
users
and
be
able
to
impart
best
practices
and
give
them
a
dynamic
and
flexible
platform
is
a
new
paradigm.
That's
still
evolving.
A
In
that
model,
data
scientists
are
always
looking
for
access
to
the
latest
and
greatest
tools
right.
They
want
access
to
the
end,
all
that
innovation
that's
happening
in
open
source,
so
being
able
to
provide
an
environment
that
gives
your
users
the
ability
to
have
access
to
those
tools,
while
maintaining
the
integrity
needed
for
auditability
and
trust
in
the
models
that
go
into
production
is
critical
and
then
the
overall
complexity
to
operationalize
ai
models
lends
itself
to
be
a
slow
and
siloed
process,
let
alone
the
life
cycling
of
those
models.
A
A
And
this
is
where
kubernetes
really
provides
a
leg
up.
Among
some
of
the
traditional
approaches
that
are
out
there
right
with
the
kubernetes
platform,
you
get
the
agility
to
automate
resource
management,
to
be
able
to
respond
to
triggers
and
stimuli
in
the
environment
that
then
necessitate
those
models
to
be
rebuilt
or
the
data
to
be
re-aggregated
and
then
redeployed.
You
also
get
the
portability.
That's
inherent
in
the
platform.
Models
need
to
be
deployed
within
data
centers.
They
can
be
targeted,
edge
components
and
really
building
on
top
of
raccoon.
A
Eddie's
platform
gives
you
that
flexibility
and
what
your
target
platform
can
be
while
taking
advantage
of
those
centralized
resources.
So
you
get
your
users
out
from
underneath
their
desktops
and
onto
a
shared
platform
where
they
have
more
scalability
and
flexibility
in
the
environment.
They're
using
you
also
get
the
flexibility
of
how
you
provision
that
aiml
environment,
not
all
model
training
exercises,
are
built.
The
same
some
require
more
resources,
some
require
specialized
hardware.
A
The
kubernetes
platforms
have
the
ability
to
gain
access
to
those
specialized
hardware,
accelerators
or
specialized
storage
devices,
and
give
you
the
ability,
then,
to
train
multiple
types
of
models
and
really
conduct
experiments
in
a
more
rapid
pace.
So
you
get
to
try
out
and
get
to
a
better
answer
quicker
than
ultimately
what
you
put
into
production
and
then,
finally,
you
get
the
inherent
scalability
right.
A
A
Right
just
like
there's
application
development
test
deploy,
we
have
the
same
thing
in
the
model
life
cycle
right,
so
you're,
building
your
ml
model,
you're
running
multiple
experiments.
When
you
found
a
model
that
you
think
is
the
one
you
want
to
put
into
production,
you
can
validate
it,
those
that
pass
the
validation
go
into
an
image
registry.
A
Once
those
things
have
been
published,
there's
triggers
then
that
ultimately
manage
the
deployment
of
that
model
as
a
service
for
integration
into
the
intelligent
application
and
they're
ultimately
deployed
out
to
their
their
final
platform.
Once
they've
been
deployed,
then
they
can
be
monitored
for
drift
and
any
other
types
of
of
metrics.
That
would
ultimately
require
you
to
retrain
and
redeploy
and
the
secret
in
all
of
this,
and
the
trick
is
really
to
have
this
be
done
in
a
continuous
pipeline.
A
So
again,
once
you've
been
able
to
build
these
models,
you're
able
to
react
and
put
these
into
production
continuously.
You
know
it's
challenging
enough
to
put
a
model
into
production
once,
but
to
be
able
to
do
it
continuously,
with
confidence
and
on
demand
is
really
a
critical
advantage,
and
you
don't
have
to
take
my
word
for
it.
We've
been
doing
this
with
a
number
of
customers
on
top
of
red
hat's
openshift
platform
in
the
healthcare
world,
where
we're
helping
with
diagnosis.
A
Obviously,
in
the
financial
services
world,
we're
helping,
monitor
and
analyze
large-scale
data,
there's
a
number
of
use
cases
across
industry,
and
so,
if
you'd
like
to
learn
more,
you
can
visit
redhat.com
and
read
more
about
these
customers.
I
thank
you
for
listening
to
me
and,
if
you're
not
already
using
your
kubernetes
as
a
model,
ops
platform,
I'd
encourage
you
to
take
a
look
at
it.
It's
a
great
platform
and
I
think
it
will
meet
a
lot
of
the
needs
of
your
user
base.