►
Description
On today's stream we're joined by Robert from the RHODS team to talk about the data science platform and OpenShift!
Join OpenShift Commons:
https://commons.openshift.org/join/
Data and AI Slack Channel: https://openshiftcommons.slack.com/archives/C8L8S4XFF
OpenShift Users Slack Channel: https://kubernetes.slack.com/archives/C6AD6JM17
Get the slides here: https://speakerdeck.com/redhatlivestreaming/ask-an-openshift-admin-ep-107-red-hat-openshift-data-science
Want to learn more?
Try Red Hat OpenShift: https://red.ht/3uVxkya
Follow us:
Andrew Sullivan https://twitter.com/practicalandrew
Jonathan Rickard https://twitter.com/jrocktx1
OpenShift https://twitter.com/openshift
Subscribe to Red Hat's YouTube channel: https://www.youtube.com/redhat/?sub_c...
About Ask an OpenShift Admin:
Do you have questions? We want to answer them! Every Wednesday at 11am ET, hosts Andrew Sullivan and Jonathan Rickard sit down to answer your questions about Red Hat OpenShift.
About OpenShift:
Red Hat OpenShift is an open source container application platform based on the Kubernetes container orchestrator for enterprise application development and deployment.
OpenShift on redhat.com https://red.ht/3rYz5ZS
OpenShift on GitHub https://github.com/openshift/
#RedHat #OpenShift #Kubernetes #ContainerPlatform
A
B
A
Morning,
good
afternoon,
good
evening,
wherever
you
are
in
the
world,
this
is
the
ask:
an
openshift
admin
live
stream.
You
may
notice
that
my
lovely
co-host
Andrew
is
out,
and
that
is
because
he
is
saving
America
at
the
tower
doing
doing
what
he
does.
He
had
a
meeting
that
he
couldn't
miss,
and
so
we
will.
You
are
well
you're,
going
to
be
forced
to
suffer
through
this
with
me
and
and
Robert
here,
and
Andrew
will
be
back
hopefully
next
week.
A
Yeah
I
love
it
and
we
were
talking
in
the
pre-show
a
little
bit
like
it's.
It's
such
a
broad
conversation
right,
there's
so
many
different
things
that
we
could
fit
into
this
bubble,
like
the
AI
bubble
right,
so
I
I
think
it's
I'm
looking
forward
to
exploring
this
with
you
like.
It's
super
exciting
for
me.
So
let's
go
ahead
and
do
the
top
of
Mind
topics.
So
there's
not
a
whole
lot
going
on.
A
A
You
know
the
streams
that
we
missed
before
kind
of
go
over
all
the
things
that
are
in
4.13
and
hopefully
do
like
a
little
bit
of
a
demo
with
some
of
the
hypershift
stuff
that
I've
been
working
on
so
really
looking
forward
to
that,
and
then
We've
also
got
Jack
from
the
Scupper
team.
He's
gonna
come
in
and
he's
going
to
talk
about
red
hat
service
interconnect
after
that,
so
we've
got
a
lot
going
on.
A
A
Also
on
all
the
socials,
so
Twitter
I'm,
Jrock,
tx1,
Reddit
same
thing,
kubernetes
slack
and
then
Andrew
is
andrew.sullivan
redhat.com
and
he's
also
at
practical
Andrew
on
Twitter
and
the
other
thing
is,
you
know
before
you
get
moving
over
to
the
open
or
the
the
data
science
stuff.
We
are
also.
You
know.
This
is
an
office
hours
presentation
right,
so
we're
here
to
answer
your
questions.
So
if
you've
been
watching
the
stream
or
if
this
is
your
first
time
there
are
no
dumb
questions
here.
A
You
know
that
we
love
answering
questions.
We
love
the
interaction
with
the
crowd,
so
feel
free
to
ask
away.
Hopefully
it's
about
the
topic,
but
even
if
it's
about
openshift
or
it's
about
kubernetes
or
something
that
we
can't
answer,
we
will
Robert
and
I
we're
not
the
we're,
not
the
experts
on
everything,
but
we
try.
So
if,
if
we
don't
know
the
answer,
we
know
the
people
that
probably
do,
and
so
we
can
reach
out
internally
and
hopefully
get
your
question
answered.
B
B
And
we
already
covered
what
is
this
and
who
I
am
so
we'll
skip
this
one
I
can
cover
openshift,
AI,
real,
quick
and
the
roads.
What's
the
difference
between
them,
the
Reddit
openshift
data
science
that
would
be
roads,
I'm,
not
going
to
say
Reddit
openshift
data
science
every
time
because
I'll
twist
my
time
but
so
openshift
AI
is
the
the
parents
name
to
it.
B
So
rather
tokenshift
data
science
is
one
product
under
openshift
Ai
and
with
more
products,
possibly
coming
out
in
the
future
or
looking
to
come
out
in
the
future.
But
right
now
they
use
pretty
much
synonymous
synonymously,
so
Let's,
let's
start
with
why
Reddit
is
doing
AI
because
rather
there's
not
not
well
known
for
being
an
AI
company
per
se.
But
what
we
do
really
well
is
infrastructure
and
the
platforms
for
that.
So
the
image
we're
looking
at
right
now
is
a
fairly
popular
image
from
all
the
way
back
at
2015.
B
Is
that
the
they
often
neglect
a
bunch
of
these
places
and
AI
very
early
on
had
a
almost
infamously
high
failure
rate?
A
lot
of
Enterprises
tried
to
use
Ai
and
it
didn't.
It
didn't
work,
the
the
they
didn't
get
the
applications
out,
and
this
is
one
of
the
reasons
it
might
have
been
that
they
they
wanted
to
build
a
AI
model
which
predicts
customer
churn,
for
example,
and
they
had
some
really
smart
people,
hacking
away,
building
this
model
in
a
notebook,
and
then
they
didn't
really
know
what
to
do
with
it.
B
They
maybe
they
threw
it
over
the
over
the
fence
to
to
some
I.T
person
who
had
to
try
to
get
get
this
thing
now
serving
and
running
or
say
that
they
actually
got
it
up
and
running.
They
now
have
a
one-time
of
model
living
somewhere
and
if
that
model
was
trained
on
data,
that
does
not
represent
the
real
world
or
if
the
real
world
changes
which
it
oftentimes
does,
the
model
is
no
longer
accurate
and
they
would
need
to
somehow
replace
that
model
and
train
a
new
model
on
the
on
the
new
data.
B
B
A
Know
got
it
tossed
over
and
I
had
to
get
it
deployed
and
and
maintain
it
and
stuff
like
that,
I
I'm,
it's
cool
to
see
that
this
Evolution
has
finally
happened.
Right
because,
like
you
know,
we
back
in
the
day
we
get
this
thing
going
like,
and
it's
like,
okay
I
think
it
works.
You
know,
so
it's
it's
kind
of
nice
that,
like
we're
we're
here
where
we're.
Finally
getting
this
workflow
kind
of
moving
along.
B
B
So
so
this
is
where
Reddit
saw
that
we
could
come
in
and
help
out.
You
can
draw
a
parallel
from
this
to
early
development
as
well,
where,
before
devops,
where
a
platform
such
an
open,
such
as
openshift,
came
in
and
helped
a
lot
to
take
care
of
these
components
which
which
are
not
just
the
model
or
the
application
code.
B
So
that's
that's
the
first
piece.
The
second
piece
is
okay:
if
we're
going
to
use
openshift
are
container
is
actually
good
for
doing
data
science
on
is
this
openshift
is
very
much
container
based.
So
is
this
even
a
match
and
turns
out
that
containers
are
quite
excellent
for
for
data
science,
I
mean
they're
excellent
for
most
developments,
but
what
makes
them
quite
nice
for
data
science
as
well.
Besides,
just
the
standard
things
is
where
the
house
Slim
slim,
they
are
and
with
how
easy
it's
just
quickly,
spin
them
up
and
down.
B
So
that's
one
reason
why
why
containers
are
quite
good
for
this?
The
the
second
one
is
how
isolated
they
are
in
their
environments.
If
you
spin
one
up,
as
you
guys,
probably
know
it's
going
to
run
more
or
less
the
same
every
time
you
can
of
course,
connect
it
to
external
factors,
but
ignoring
that
part,
it
will
work
the
same
way
and
with
how
experiments
have
a
data
science
is.
This
is
also
quite
fantastic.
B
This
makes
it
really
easy
to
reproduce
results
and
reproduce
experiments
which
is
we'll
see
it
a
little
bit
when
we
jump
into
the
product
as
well,
but
yeah.
Another
thing
from
back
in
the
past
is
you
would
have
this
model
which
are
trained
on
and
you
try
to
give
it
to
someone
else
and
they
would
have
the
wrong
dependencies
and
everything
would
run
wrong.
Nothing
new
the
same
in
development,
but
it's
very
much
the
same
here
as
well.
B
Then
containers
buy
themselves
are
often
not
enough,
so
you
probably
want
to
connect
them
together
into
clusters.
That's
where
kubernetes
comes
in
coming
in
with
the
automatic
rollouts,
rollbacks
and
so
on,
and
the
really
nice
part
here
is
how
extensible
is
again.
This
is
great
in
the
data
science
field,
because
there
are
so
many
open
source
Tools
in
data
science,
so
you
can
really
take
your
Cube
flows,
airflows
ml
flows.
A
Yeah
I
think
once
you
start
doing
things
at
scale
right,
that's
where
openshift
really
kind
of
shines
like
when
you've
got
a
cluster,
maybe
two
clusters,
a
handful
whatever
it
is
right
and
a
couple
of
applications
and
you're
doing
things
like
okay,
DIY,
it's
fine,
it's
totally
legit,
it's
a
perfect
way
of
doing
work,
but
then
I
think,
once
you
start
scaling
out,
you've
got
multiple
applications:
multiple,
like
dozens
of
clusters,
hundreds
of
clusters,
whatever
it
is,
and
you're
really
starting
to
like
scale
and
Branch
out
like
big
time.
That's
where
openshift
is
just
it.
B
B
A
Yeah
so
on
my
team,
we
created
a
validated
pattern
based
off
of
medical
diagnosis
and
they
use
open
data
Hub
behind
the
scenes
like
so.
Essentially
they
had
a
Jupiter
notebook.
They
built
a
model
out
and
then
built
took
that
model
and
built
it
into
an
application
that
gets
deployed
as
part
of
the
pattern
and
then
another
one
uses
like
I
think
it's
odh,
but
yeah
but
yeah.
We
we've
talked
about
it
a
long
time.
I
could
probably
actually
find
the
episode
awesome.
B
A
B
And
we
also
so
it
you
can
run
it
more
or
less
where
openshift
runs
and
we're
we're.
Currently
we
have
the
micro
shift
as
well
right
for
Edge
devices,
so
we're
currently
building
a
slimmed
down
version
for
serving
which
can
go
out
to
these
microchip
devices
and
help
you
just
have
a
serving
layer
and
serve
models
on
these
Edge
devices.
B
So
that
that's
where
openshift
data
science
comes
from
and
it's
since
it's
a
curated
version
of
open
data
Hub,
it
doesn't
contain
everything
open
data.
Hub.
Does
we
take
a
few
of
the
components
and
we
put
them
into
a
nice
little
ux
and
it
essentially
gives
a
very
good
Baseline
for
your
mlops
platform
and
the
cool
part
about
this
one
is
you
can
also
grab
in
other
components,
so
say
that
you
you
want
to
use
airflow
for
your
pipelines.
B
For
example,
you
could
integrate
that
as
well
and
use
those
pipelines
and
then
use
openshift
data
science
for
serving
and
model
experimentation,
for
example,
but
before
I
keep
getting
ahead
of
myself.
What's
actually
inside
of
openshift
data
science,
so
we
have
model
development,
environments
and
I'll,
show
I,
guess
briefly
going
through
this
now
and
we'll
very
soon
see
in
the
demo,
how
they
actually
looks
like
and
how
they
work,
but
there's
environments
for
model
development
that
their
Baseline
notebooks.
A
B
You
do
with
them,
then
there's
model
serving
and
monitoring.
So
after
you
experimented
with
your
model
and
you
build
something
which
you
want
in
production,
you
probably
want
to
be
able
to
serve
it
somehow
and
that's
where
model
serving
comes
in
just
points
to
the
model
which
you
build
and
it'll
pull
it
in
start
serving
it
and
give
you
metrics
monitoring
on
it
awesome
then
lifecycle
management
gives
you
pipelines
for
data
science
if
you've
done
devops,
you're
well
familiar
with
pipelines
and
the
benefit
of
those
work
works
very
similar
here.
B
It
gives
you
at
the
very
least,
continuous
delivery
of
the
pipelines,
so
you
can
trigger
trigger
them
at
say
that
your
data
is
Shifting,
for
example,
so
the
data
you
keep
getting
in
starts
changing
watch
it.
It
doesn't
look
like
it
looked
like
yesterday,
for
example,
you
might
want
to
retrain
your
model,
so
you
can
then
kick
off
a
training
Pipeline
and
get
a
new
model
into
production
automatically.
A
B
It's
your
single
purchase
pipelines,
so
it's
keep
flow
Pipelines
and
you
can
use
it.
The
full
stack
is
elyra
at
the
very
top
if
you
want
to
use
it
and
that's
a
drag
and
drop
pipeline
Builder
using
python
files
and
python
notebooks
that
sends
it
to
keep
flow
pipelines
which
then
sends
it
to
openshift
pipelines.
So
when
we
kick
off
a
pipeline,
we'll
see
it
run
in
openshift
pipelines
as
well.
Okay,
awesome.
A
B
Course
so,
and
this
models
AI
models
in
general,
you
fit
in
some
data
and
you
get
a
prediction
right
and
to
build
one
of
these
models.
You
feed
it
a
lot
of
data
and
you
tell
it
what
prediction
each
data
sample
should
give
it.
It
should
give
for
this
one,
so
you
send
in
a
dog-
and
you
tell
it
this
should
be
a
dog
very
simplified,
but
you
do
that
a
lot
and
the
model
starts
learning.
B
A
So
to
continue
it's
a
flow
right,
so
that
way,
it's
always
like.
It's
always
cute.
It's
keeping
up
with
everything.
That's
going
on.
You
know
so
that
way
like,
like
you
said,
if
there's
a
dog
with
five
legs,
now
it'll
see
that
and
they'll
say:
oh
okay!
This
is
a
five-legged
dog
now,
but
it'll
still
call
it
a
dog
exactly.
B
So
that's
the
that's
the
magic
with
the
pipelines
right.
You
can
gather
this
new
data
and
you
can
just
train
the
model
automatically
or
semi-automatically
whenever
you
need
to
gotcha
going
a
little
bit
deeper
in
into
Rabbit
Hole
as
well.
We
also
have
something
called
fine
tuning
which
so
retraining
just
trains
the
model
from
scratch
and
fine-tuning
takes
the
existing
model
in
your
stitches
hit
only
on
the
new
data,
so
two
different
two
different
ways
of
modifying
the
model.
B
Yeah,
so
some
models
are
quite
large
in
chatgpt,
for
example,
and
you
might
want
not
want
to
retrain
it
every
time
you
want
to
change
it,
so
a
very
good
way
to
teach
it
new
data
is
to
fine
tune
it
and
fine-tuning
can
also
so
you
can
have
these
models,
which
are
really
good
at
say,
recognizing
a
bunch
of
different
dimension
objects,
and
then
you
fine
tune
it.
So
it
now
is
really
good
at
recognizing
a
specific
object
which
you're
interested
in
so
the
two
application
areas
of
fine
tuning.
A
Nice,
so
we
do
have
a
question
and
it
looks
like
Frank
answered
it
right,
so
one
of
them
was
with
data
science.
Is
it
possible
without
a
data
Lake
and
Frank
answered
like
yeah,
it's
possible,
but
it's
essentially
you're
going
to
run
into
issue
sooner
or
later
and
just
to
read
out
what
he's
saying
he
says.
Database
are
not
necessary
per
se,
but
they
provide
a
more
flexible
steam
schema
mechanism
than
traditional
data
stores,
and
then
Dwayne
also
asked
does
the
edge
version
of
the
roads
deployment?
A
B
B
B
B
B
Collaboration
gives
you
projects,
which
are
you
have
access,
control
and
all
of
those
good
things
in
it.
A
project
is
essentially
an
openshift
namespace,
so
you
can
they're
very
separated,
and
this
is
one
of
the
main
enhancement
enhancements
we
do
when
we
pull
in
front
of
example,
sorry
swap
to
Swedish
there
for
a
second
when
we
pulled
him
from,
for
example,
from
Cube
flow
pipelines
and
environments
are
not
necess
are
not
natively,
don't
have
projects
natively
in
there.
A
A
B
So
we
provide
we
provide
environments
and
Tools
around
it,
so
you
can
get
models
into
production
if
you're
serving
in
pipelines
and
so
on
the
tools
you
want
to
use
for
the
models
themselves
to,
and
we
do
have
a
few
things
to
help
with
bias.
Detection
and
things
like
that
coming
in,
but
largely
the
tools
you
want
to
use
for
the
model
is
to
to
improve
them
and
to
validate
them
is
very
much
up
to
what
your
preference
is
yeah.
So
you
have
full
access
to
to
environments,
to
experiment
in
independencies.
B
A
B
I
I
think
so,
but
I'll
say
it
would
say:
Frank
has
another
answer
cool.
So
all
of
this
put
together
gives
you
a
baseline
platform
for
our
Baseline
tools
to
kind
of
build
mlops
platform
on
toppo
openshift
and
the
really
cool
part
with
it
is,
as
I
mentioned
before.
It's
also
hybrid.
So
it
works
across
the
hybrid
cloud
and
because
it's
on
openshift
it
kind
of
brings
everyone
together.
So
you
have
the
it
Ops
data
scientists
and
application
developers
now
all
on
the
same
platform.
So
it's
easier
to
manage.
B
Shows
a
very
simplified
version
of
an
analogs
workflow
and
where
you
can
pull
in
different
tools.
So,
for
example,
openshift
data
science
covers
model
development,
it
has
servings,
you
can
integrate
it
in
the
application
development
workflow
and
we
have
monitoring
and
management
already
a
little
bit
and
more
coming
out.
B
But
we
don't
tell
too
much
on
the
Gathering
prepare
data
side.
So
that's
where
you
can
plug
in
someone
like
Starburst
or
the
amq
streams,
and
even
on
the
parts
where
we,
where
openshift
data
science
do
cover,
you
can
still
plug
in
things
such
IBM
Watson
for
automl
or
intelope
and
Vino
for
Accelerated
Hardware,
and
that's
that's
kind
of
how
we
want
to
build
it
out.
So
it's
modular.
So
you
can
pull
in
these
tools
with
you
which
you
need
and
which
are
well
fitted
for
your
ml
Square,
the
one
you
need
to
build.
A
B
B
And
the
first
thing
you
see
when
you
enter
it
is
going
to
be
these
enabled
applications
in
this
environment.
I
only
have
Jupiter,
but
if
you,
if
you
add
more
things
such
as
the
GPU
add-on
or
things
like
that,
it's
going
to
show
up
here
as
well
and
it'll
act
as
a
little
as
a
little
portal
for
you
to
get
to
different
applications.
That's
connected
with
openshift
data
science.
B
So-
and
this
Jupiter
specifically,
is
quite
nice.
We
just
quickly
launching
an
environment
and
playing
around
with
so
here.
You
can,
for
example,
select
your
notebook
images
with
different
dependencies
in
them
you
can
select
the
deployment
size
and
these
are
t-shirt
sizes,
but
they're
customizable
as
well,
so
just
making
it
easy
to
give
a
few
different
options:
environment,
variables
and
if
I
had
GPU
enabled
here,
you
would
see
a
drop
down
for
that
as
well,
but
this
just
spins
up
a
single
environment
and
what
you
might
want
to
do
instead
is
create
a
project.
A
B
Have
two
different
ones
here:
I
have
one
for
object,
detection
and
one
for
customer
churn,
so
I
figured
we
would
speed
through
them,
both
real,
quick
to
show
two
different.
One
showcase
is
serving
and
the
other
one
showcases
pipelines.
So
let's
jump
into
object
detection,
and
this
is
how
it
looks
like
inside
a
projects.
B
A
B
A
A
And
for
those
that
don't
know,
Min
IO
is
a
it's
an
object.
Storage
emulator,
essentially
right
like
it,
takes
storage
and
presents
it
to
your
cluster
as
as
object
storage.
So
it's
a
pretty
cool
little
project.
Yeah,
don't
know
if
it's
production
ready
but
like
it's
definitely
something
if
you're
kicking
the
tires
on
something
and
you
need
Object
Store
and
you
don't
have
S3
or
an
object:
storage,
local,
then
Min
iOS.
It's
an
awesome
tool.
B
It's
very
handy
to
spin
up
and
get
something
quick
running,
so
let's
create
a
workbench
for
this
object.
Detection
so
enter
and
name
I
spelled
wrong
there.
We
go
we're
going
to
select
the
image
just
like
we
saw
before,
and
the
images
have
different
versions
as
well
deployment
size,
I'm
going
to
leave
it
on
small,
so
don't
break
anything.
B
B
So
this
is
how
the
data
connection
looks
like
when
you
set
it
up,
so
you
provide
it
with
a
name
with
an
access
key,
a
secret
X,
access,
key
and
point
a
region
and
a
bucket
and
behind
the
scenes
it's
nothing
magical.
All
it
does
is
just
save
these
environment
variables,
so
it'll
be
easy
for
you
to
reference
throughout
your
projects.
A
A
I
can't
remember
the
exact
amount,
but
it's
I
think
it's
a
couple
terabytes
or
something
like
that
of
of
storage,
and
you
can
use
you
know,
block
file
or
object
storage
right,
so
he's
just
making
the
point
that
you
get
all
of
that
with
your
odf
openshift
platform,
Plus
subscription
and
then
rope
nine
just
asked
as
we
always
like
end
users
that
can
resource
that
can
use
resources
responsibly.
Is
there
any
simple
way
for
the
end
users
to
understand
how
much
their
workload
is
consuming?
So
is
there
through
the
monitoring?
B
Well,
first
of
all
they
get
to
choose
how
much
resources
it
should
use.
It
won't
say
how
much
it
is
from
a
total,
but
as
an
admin,
you'll
probably
set
it
up,
so
they
they
have
access
to
not
not
larger
t-shirt
sizes.
That
then
you
would
expect
someone
to
use
right,
so
they
kind
of
limited
in
the
T-shirt
size.
B
It's
not
sizes,
and
the
other
thing
you
can
do
is
you
can
have
cuddling
on
this
on
this
pods
for
the
workbenches,
so
you
can
automatically
turn
them
off
if
they
are
active
too
long
or
sorry
if
they're
inactive
too
long
or
you
can
just
shuttle
down
at
midnight.
For
example,
it's
a
free
up
resources
that
way
gotcha.
B
B
It's
we
have
One
customer
Boston
University,
who
very
much
utilizes
stats
where
they
they
have.
They
go
from
knitting,
zero,
workbenches
or
zero
resources
up
to
300
people
using
them
all
at
once,
and
then
they
have
a
so
that
they
have
some
tools
for
spinning
up
all
these
workbenches
really
quickly,
and
they
also
have
tools
for
culling
them
down
after
the
lecture
is
done.
B
A
B
Okay,
so
this
is
how
your
environment
looks
like
if
you
want
to
customize
it
too
much
and
in
here
you
have
it.
If
you
ever
used
a
Jupiter
notebook.
This
will
look
very,
very
similar
and
familiar
to
you.
So
what
we're
gonna
use
now
is
Justice
GitHub.
This
git
feature
to
quickly
pull
in
a
GitHub
repo.
B
B
The
other
thing,
so
we
have
this
data
connection
right
and
we
connected
this
to
our
workbench.
So
what
we
can
do
is
we
can
take
that
that
model
which
we
stored
in
the
buckets
and
we
can
run,
and
we
can
run
inference
on
that
one.
So,
actually
let
me
back
back
up
a
little
bit.
I
jumped
forward
one
too
many
steps
so
we'll
before
we
start
running
or
inferencing
on
this
serving
model,
I'm
just
going
to
simply
pull
down
the
model
and
run
a
locally
inference
on
that.
B
A
Nice
and
the
cool
part
about
this,
too,
is
like
with
the
notebooks
and
everything
and
then
oh
and
I
know
we
keep
going
back.
I,
keep
bringing
it
back
to
openshift,
it's
not
on
purpose
mostly,
but
like
with
you,
can
you
can
build
an
image
stream
that
will
import
the
notebooks
into
the
image
stream,
and
then
you
can
redeploy
that
image
out
of
out
of
your
workbench
here
right
so
like
so,
you
had
some
Jupiter
notebooks
that
you
guys
that
you
know
all
right.
A
This
is
this
is
my
core
thing
that
I
want
to
use
right.
This
is
the
base
of
my.
You
know
that
everybody
needs
to
use
or
whatever
you
can
build
those
into
an
image
stream
and
then
that
image
stream
becomes
available
as
one
of
the
images
in
your
in
your
workbench
that
you
can
use
so
say,
I
I
can't
remember
the
one
you
selected
right.
You
have
one,
that's
like!
Maybe
it's
just
Jupiter
or
you
have
like
the
yeah
Cuda.
It
would
actually
be
part
of
this
list
here.
A
As
long
as
you
had
access
to
it
right
and
you
could
have
like
the
you
know-
kitten's
core
or
whatever,
and
then
it
would
just
import
like
that
base
those
those
base.
Notebooks,
so
I
think
that
that's
pretty
awesome,
you
can
have
all
that
stuff
integrated,
because
it
goes
back
to
that
story
that
you
were
telling
in
the
beginning.
Right
like
you,
it's
the
reusability,
it's
the
repeatability
and
stuff
like
that,
without
going
back
and
rebuilding
your
Dev
environments
and
stuff
right.
B
B
So
here
we
have,
we
currently
have
openvino,
but
you
can
just
like
custom
notebooks,
add
in
custom
serving
runtimes.
So
if
there's
something
missing
you
can
add
in
is,
for
example,
the
really
large
LMS
they
usually
use
their
own
custom.
Runtime
lrm
stands
for
Lord
language
model,
which
is
shatibility
or
similar.
B
So
I
created
one
with
openvino
here
and
then
I
deployed
a
model
to
it
and,
as
I
mentioned
before,
to
deploy
a
model
is
simply
just
selecting
what
type
of
model
it
is
and
then
selecting
where
the
model
is
stored.
So
in
this
case
it's
this
has
free
bucket
and
a
specific
path
and
then
I
will
pull
it
in
and
start
serving
it
and
everything
goes
well.
It'll,
look
something
like
this.
Where
you
get
a
little
endpoint
and.
A
B
A
B
B
B
A
B
Excited
it's
yeah
I
love
to
demo
this,
because
it's
so
it's
so
much
fun
to
play
around
with
it.
That
leads
us
to
the
the
last
feature:
I
haven't
really
showed,
which
is
the
pipelines
and,
as
mentioned
before,
it's
Pipelines
just
for
data
science
workflow.
Instead,
so
I'm
gonna
jump
into
our
other
projects.
B
I
have
a
very
tiny
one
created
here
and
what
this
does
is
it
downloads
some
data
and
process
it?
It
will
then
train
two
models,
one
with
I,
don't
remember
what
the
full
name
here
is
a
gradient
boost
right,
one
with
gradient
boost
and
one
with
Renault,
Forest
and
finally,
we'll
compare
both
of
them,
and
it
will
push
the
best
one
of
them
as
the
best
model
as
a
pickled
file
in
this
case
now,
I'm
just
pushing
this
model
to
to
our
Min.
B
The
workbench
I
ran
here
is
elyra
and,
as
I
mentioned
before,
Lyra
will
convert
this
to
a
is
your
keep
flow
pipeline,
which
then
goes
into
the
openshift
pipeline,
and
this
gives
a
really
nice
interface
for
data
scientists
to
just
easily
drag
and
drop
in
this
case
notebooks,
but
python
files
as
well,
but
also
giving
a
lot
of
access
to
other
settings
such
as
what
data
should
transfer
in
between
them.
What
environments
do
I
want
to
use?
B
B
We
can
also
see
the
random
force
model
and
the
gradient
boost
model.
So
you
have
access
to
all
of
this
and
they
also
give
a
I
think
this
is
from
elyra.
They
give
a
really
nice
little
interface
for
debugging
it.
So
if
I
I
don't
know
how
to
show
this
because
I
need
to
download
it
I
believe
yeah
I
can't
show
that
here.
But
if
you
download
this
and
look
at
it,
it'll
show
you
your
little
notebook
and
give
you
all
the
outputs
and
potential
errors.
A
Nice,
the
cool
thing
like
I'm,
just
I'm,
not
a
data
scientist
at
all,
like
you
probably
picked
up
on
that
right,
I'm
kind
of
dumb
but
like
the
cool
thing,
is
that
you
don't
have
to
be
an
open
shift
or
kubernetes
Guru
to
do
this
right,
like
you,
you
just
go
to
this
workbench.
You
do
your
thing.
It
looks
like
all
the
options
are
there
like.
If
you
need
to
add
all
your
connection,
points
and
stuff
like
that,
that's
that's
pretty
awesome
now
and
I.
I
assume
that,
like
anything,
I
can
do
in
the
UI.
A
B
A
Yeah,
this
is
awesome.
This
is
really
awesome.
So
if,
if
people
wanted
to
like
learn
about
this
right
like
are
there
any
resources
that
are
out
there,
that
they
can
go
to
and
just
kind
of
Kick
the
tires
with
it
or
do
they
have
to
have
a
subscription
or
can
they
use
like
the
developer
subscription?
Do.
B
B
We
have
a
few
really
good
videos
out
there
and
we
also
have
this
website
called
AI
on
openshift,
and
this
in
here
you
can
find
a
bunch
of
material
on
open
data.
Hub,
openshift
data
science
AI
a
few.
How
to's
different
tools
like
me
and
I,
O
I,
just
went
down
and
copied
copied
this
one
to
get
into
my
project,
but
also
other
things,
such
as
all
the
flows,
we're
talking
about
airflow.
A
B
So
how
that
works
is
open.
Data
Hub
has
a
fair
bit
more
components
than
we
have
in
openshift
data
science.
So
we
have
different
tiers
of
tier
zero.
Is
everything
that's
in
openshift
data
science,
and
then
we
have
another
tier
where
you,
where
we
incubate
the
different
components,
they're
about
to
go
into
openshift
data
science
and
a
few
which
are
only
an
open
data
hub
for
now,
and
you
can
find
all
of
that
on
the
open
data
Hub
website
nice.
B
A
A
All
right,
no,
this
is
this
is
really
awesome.
So,
like
my
experience
has
mainly
been
around
like
open
data,
Hub
and
open
data,
Hub
was
awesome.
You
know
like
it
did
everything
that
we
needed
it
to
do,
but
there's
still
some
like
there
there's
some
massaging
right.
This
it
looks
like
this
has
got
all
the
all
the
knots
worked
out.
A
You
know
to
stay
within
the
massage
conversation,
but
like
it's
cool
to
see
that
you've
got
this
interface,
you
can
go
through
and
you
can
work
all
your
magic
and
then
you
don't
have
to
know
all
the
back-end
stuff,
because
I
feel
like
that's
where
a
lot
of
like
hesitancy
comes
from
to
like
adopt
something
like
this
because
you're
like
oh
well,
I've
got
to
know:
I've
got
a
new
open
shift
or
I've
got
to
know.
Kubernetes
or
I've
got
to
know
all
this
stuff.
A
But
when
you
go
in
and
you
have
this
you're
like
Hey,
look
there's
this
UI,
you
just
go
click
the
buttons.
Add
the
connections
that
you
want
and
as
an
admin
right,
which
is
I,
I,
think
most
of
us
or
a
lot
of
us
hopefully-
and
you
know
those
that
aren't-
hopefully
you
don't
have
to
do
it,
but
as
an
admin,
I
can
also
control
like
those
levels
right
like
hey
I,
don't
want
I,
don't
want.
A
You
know,
Johnny
to
have
the
2XL,
because
I
know,
even
though
he's
a
medium
he's
always
going
to
take
the
2XL,
because,
like
databases
or
whatever
you
know
so,
I
think
having
that
kind
of
control
from
an
admin
perspective
and
then
having
that
kind
of
control
from
the
user
perspective,
because
it
still
feels
like
you're
empowered
to
do
your
job
without
being
limited
by
the
tool
right
and
I.
Think
that
that's
just
that
excuse
my
friend.
That's
awesome.
B
Yeah,
that's
the
self-service
part
of
it
for
the
data
scientists
and
that's
one
of
the
main
goals.
Openshift
data
science
is
to
very
easily
being
able
to
get
access
to
resources.
You
need
and
the
tools
you
need
so
making
it
easy
to
get
pipelines
and
get
the
workspaces
or
even
integrate
tools
outside
of
Roads
for
the
for
the
admin.
A
Yeah
I
think
that's
yeah,
that's
awesome
and
Frank
I
just
saw
a
comment
that
Frank
made
where
he
says:
data
scientists
are
the
rock
stars
and
I.tops
or
the
road
he's
like
that's
100
accurate
right
like
that
avrock
serves
without
rodeos.
You
can't
have
Road
just
without
the
rock
stars.
Yeah
I
totally
agree
with
that.
That's
awesome!
A
A
Your
your
data
center
has
just
you
know
the
temperature
just
Rose
by
like
30
degrees
or
whatever
you
know,
that's,
probably
terrible,
but
but
but
it's
taken
all
this
information
and
it's
actually
given
us
something
that
we
can
use
and
then
make
a
decision
off
of,
but
then
on
top
of
that
now
we've
got
all
this
Automation
in
there
right.
So
it's
like
okay!
Well,
look!
If
this,
if
this
is
what's
in
the
output,
then
go
do
this
thing
and
if
it's
really
bad
then
call
me
right.
A
It's
the
take
two
Tylenol
and
call
me
in
the
morning
pill,
or
you
know,
automation,
so
I'm
just
curious
like
how.
How
far
do
you
see
this
going
chat?
Gbt
is
awesome.
It's
become
a
household
name
right
like
it's.
It's
everybody
knows
what
chat.
Gpt
is
good,
bad
or
ugly
right
like
they.
They
know
what
it
is
yeah,
so
I'm
just
curious.
How?
Where
do
you
see
it
all
going
like
the
object
detection?
You
know
video
detection,
all
that
stuff.
It's.
B
B
It
doesn't
have
text,
it
also
takes
images
which
is,
or
it
does
to
a
point
audio
and
can
output
those
different
things
as
well
and
at
some
point
I
mean
it's
already
predicted
that
the
majority
of
Enterprises
will
incorporate
more
and
more
AI
and
as
they
get
better
and
better
it's
it's
the
point
where
it
comes
to
your
imagination
right
how
far
it
will
go.
Yeah.
A
A
Indeed,
yeah
and
all
the
all
the
red
Hatters
are
going
to
be
like
dude,
you
got
light
speed,
right,
dance
will
light
speed
where
you
say:
hey,
write
me
a
Playbook
that
does
this
thing
you
know,
and
then
there
should
be
the
coop
stuff
that
comes
out
like
there.
It's
there's
just
so
much
it's
awesome
and
like
I
I've,
seen
some
external
projects
where
they're
doing
like
a
coupe,
GPT
or
something
like
that
or
check
I
can't
remember
what
they
call
it,
but
it's
essentially
the
same
type
of
cool.
A
B
One
more
slide,
which
just
shows
where
roads
are
setting,
so
we
talked
about
the
shot,
GPT
and
one
thing
which
we're
doing.
Obviously,
you
know
about
the
light
speed.
So
what
we're
getting
more
and
more
in
here
is
distributed
training
and
serving
for
these
large
language
models.
So
you
can
fine
tune.
You
can
prompt
tune
these
models
and
get
the
entire
lifespan
of
this
really
huge
models
as
well
into
into
your
mlo
system.
A
That's
awesome,
yeah
the
distributed
training.
That
is
that's
awesome.
Right
I
mean
you
think,
just
from
a
medical
perspective,
you
know,
if
you're,
if
you're,
trying
to
train
just
using
the
X-ray
story
again
right,
like
you've,
got
all
these
medical
facilities
everywhere
right,
like
pop-up
emergency
rooms,
you
know
Med
Clinics,
all
that
stuff
right.
They
can
take
an
x-ray.
They
can
send
it
into
this
model.
A
You
know,
make
their
determination
locally
or
whatever
and
then
send
it
into
this
model,
and
then
that
way,
next
time
or
maybe
even
that
time
who
knows
like
at
some
point,
they
can
use
that
and
say:
hey,
look,
okay,
nope
you're
good
to
go
doctor.
Had
it
right
or
you
know
hey,
maybe
if
we
need
to
review
this
with,
you
know
a
like
an
actual
doctor
or
not
not
a
radiologist
or
whatever,
or
what
the
radiologist
whatever.
A
A
A
A
So
I
I,
don't
have
anything
else.
I
mean
I.
Consider
and
just
ask
you
a
bunch
of
dumb
questions
because
I
have
a
bunch
of
them,
but
I
don't
know
if
there's
anything
else
that
you
want
to
cover
or
anything,
but
this
was
to
me
this
was
awesome.
I
really
appreciate
you
coming
on,
you
know
and
taking
the
time
I
know
it's
late
in
the
day
for
you
so
yeah.
B
A
A
So
if,
if
you
have
any
questions
or
if
you
have
anything,
you
want
to
talk
about
like
we're
over
there
as
well,
so
here's
the
workspace
to
openshift
Commons
so
go
ahead
and
hit
us
up.
Let
us
know
if
you,
if
you
want
to
get
a
hold
of
Andrew
and
you
don't
you
don't
have
his
email
or
whatever,
my
goodness,
I'm.
Sorry,
if
you
want
to
get
a
hold
of
Robert,
not.
B
A
A
hold
of
Robert
hit
me
up
and
let
me
know
and
I'll
connect
the
two
of
y'all,
so
you
know
you
can
ask
some
questions
but
yeah
other
than
that
we
got
Trulia
next
week.
It's
gonna
be
an
awesome
month.
August
is
going
to
be
great
Robert
thanks
again
for
coming
on.
This
was
super
informative
for
me,
I
I
really
appreciate
you
walking
through
the
the
UI
and
and
giving
the
presentation.
So
thanks
again,.
B
A
And
then
Frank
said
that
there's
oh
there's
an
AI
Channel
over
in
the
open
shift.
Commons
slack
workspace
I
will
so
what
I'll
do
is
I'll
update
the
I'll
update
the
video
description
with
the
workspace
to
the
to
the
channel,
oh,
my
goodness,
the
URL
to
the
channel,
and
so
that
way
we'll
have
open
shift
comments
in
there
and
then
we'll
have
openshift
Commons
to
the
AI
channel
and
then
that
way.
If
you
want
to
join
up
and
talk
to
these
guys
like
that,
they'll
be
out
there.