►
From YouTube: ''Hot Dogs or Not" - At Scale with Kubernetes [I] - Vish Kannan & David Aronchick, Google
Description
''Hot Dogs or Not" - At Scale with Kubernetes [I] - Vish Kannan & David Aronchick, Google
Kubernetes promises to be a multi workload platform. This talk will explore how Kubernetes can be easily leveraged to build a complete Deep Learning pipelines starting all the way from data ingestion/aggregation, pre-processing, ML training, and serving with the mighty Kubernetes APIs. This talk will use Tensorflow and other other ML frameworks to highlight the value that Kubernetes brings to Machine Learning. Along the way, key infrastructure features introduced to abstract and handle hardware accelerators which make Machine Learning possible will also be presented.
About David Aronchick
David Aronchick is the Senior Product Manager for the Google Container Engine, and leads product management on behalf of Google for Kubernetes. David has been helping to ship software for nearly 20 years, founding and being part of the management team for three different startups, as well as squeezing in time at Microsoft, Amazon, Chef, and now Google.
About Vish Kannan
Vishnu Kannan is a Senior Software Engineer at Google. Vishnu received his Masters in ECE from Georgia Tech. He has been a systems engineer ever since he graduated. He hacked on the Linux Kernel for a couple of years at Cisco. He then worked on Borg at Google. He is currently focused on Open Source Containers, spending most of his time on Kubernetes.
Join us for KubeCon + CloudNativeCon in Barcelona May 20 - 23, Shanghai June 24 - 26, and San Diego November 18 - 21! 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.
A
Everyone
have
a
great
lunch,
barbecue
fried
chicken
I,
don't
know
the
lines
are
brutal.
Anyhow,
people
on
the
live
cast-
probably
don't
want
to
hear
about
that.
Hi
I'm
Dave,
our
on
check,
I'm,
a
product
manager
at
Google
I
until
recently
was
a
product
manager
on
the
career
days
project
and
now
we're
working
on
something
brand
new
which
we're
really
excited
about.
This
is
fish
fish.
Would
you
like
to
know
yourself.
B
How
would
now
I
guess
they
believe
me,
everyone,
I'm,
bish,
I
work
at
Google,
I
work
on
Karina's
I've
been
working
with
givenness
for
the
last
three
to
four
years.
More
recently,
I've
been
focusing
on
like
enabling
high
performance
applications,
specifically
machine
learning,
with
gorillas
in
a
portable
fashion,
and
also
like
looking
at
more
of
the
application
stack
and
making
empowering
curative
uses,
essentially
yep.
A
Do
one
of
these
like
look
up
in
the
dictionary
things,
the
definition
of
machine
learning
nope,
so
the
idea
is
first,
you
start
with
a
question
and
in
this
case
I
don't
know
how
many
people
have
bought
a
house
recently
or
sold
the
house
or
something
like
that,
but
you
may
say:
well
how
do
you
determine
what
your
house
is
worth?
You
come
up
with
some
metrics
in
this
case
square
footage
and
overall
house
price,
and
you
start
drying
some
points
on
a
graph,
and
you
start
and
say:
okay.
A
Well,
here's
a
point
on
the
graph,
maybe
I'll
add
some
more,
maybe
I'll
add
a
lot
more
and
then
all
of
a
sudden
it
looks
like
a
pattern
and
you're
like
this
is
magic.
Then
you
answer
your
question:
how
much
do
you
sell
your
house
for?
Well,
my
house:
is
this
big,
therefore
I
go
up
to
the
chart?
I
go
over
and
I
get
a
number
presto
machine
learning
right,
accept,
Congrats
your
machine
living
expert
and
on
the
way
out
the
door.
You
can't
get
a
certificate
with
certifying
that,
but
things
can
get
complicated
right.
A
You
could
have
nonlinear
groupings
based
on
neighborhoods
and
environment
and
crime
rate
and
so
on
and
so
forth.
Things
can
be
multi-dimensional
here.
Obviously,
I
had
something
credibly
boring
just
two
features
in
a
linear
regression
or
things
could
change
over
time
and
so
at
a
high
level.
What
we
think
machine
learning
is
is
when
it's
you
a
way
to
solve
problems
without
explicitly
knowing
how
to
create
the
solution.
Okay,
do
we
believe
at
Google?
This
is
like
our
biggest
passion
in
the
world.
A
I,
don't
know
if
you
know
we
have
some
large
data,
centers
and
they're
pretty
expensive
to
run
the
number
one
price
ends
up
being
power
most
of
the
time,
and
we
have
some
pretty
smart
data
center
engineers
and
we
said:
hey
data
sent
engineers.
Do
you
think
you
could
wire
up
ml
to?
You
know
the
way
that
we
run
our
data
centers
and
improve
things,
so
it
did
and
for
those
who
don't
know
the
the
benchmark
for
MA.
Excuse
me
for
data.
Centers
is
PE
which
stands
for
power
usage
effectiveness.
A
You
ideally
want
it
to
be
1
1
watt
goes
in
one
goodness
comes
out,
nobody
ever
runs
at
one,
so
at
Google.
It
looks
like
this.
This
is
how
our
ml
arexx
is
me.
Our
data
center
efficient
power
usage
efficiency
was
working.
We
literally
hooked
up
the
data
center
to
this
ml
thing
that
ran
the
water
and
fans
and
so
on
and
so
forth,
and
it
looks
like
this
and
then
we
turned
it
off
and
it
looked
like
that.
A
It
literally
was
like
flipping
a
switch
and
that's
the
power
of
machine
learning,
and
because
of
that,
we
think
machine
learning
should
be
for
everyone.
That
would
be
a
great
thing
except
ml
needs
DevOps,
and
by
that
what
we
mean
is
DevOps
needs
composability
portability
and
scalability
and
I'll
get
to
these
in
just
a
second
there.
Any
way
we
could
turn
this
down.
Just
a
little
bit
even
holding
it
away
from
my
face,
it's
do
feedback
good!
That's.
B
A
So
by
devs,
like
I,
said
by
composability,
I
mean
the
following:
a
lot
of
people
look
at
ml
and
they
say
well
great.
You
know,
I'm
gonna
go
I'm,
gonna,
download,
PI,
torch
or
tensorflow
or
CNT
k,
or
something
like
that
and
I'm
gonna
go
build
a
model
who
thinks
they
are
done
when
you've
finished.
Building
your
model,
no
hands
go
up,
good,
you're,
all
very
smart.
The
reality
is
that
real
ml
looks
a
lot
more
like
this.
A
A
So
we
go
back
to
this
path
over
here
and
you
look
at
all
those
services
and
each
one
of
these
may
be
on
a
different
system
and
every
one
of
those
systems
has
an
opportunity
for
change
conflict,
something
going
wrong,
in
particular
in
the
ml
space
you
see
here
like
you'll,
have
containers
those
containers
on
premises
may
run
against
CPUs.
But
then,
when
they
move
to
the
cloud
there's
GC
GPUs
of
various
size,
they
could
be
a
six
or
FPGAs
or
you
name
it.
A
Even
ml
frameworks
often
operate
differently
in
different
environments,
and
you
need
that
stability
as
well
and
then
finally,
scalability
I
think
you're
going
the
opposite
direction
on
the
volume.
Is
there
any
like
I'm
now
like
about
two
feet
from
my
face
down
here?
Okay,
so
as
far
as
scalability
is
concerned,
same
deal
you,
you
often
want
to
be
highly
scalable.
So
you,
you
start
your
ml
and
one
minute
later
you
realize
you're
not
going
to.
You
know
converge
for
the
next
450
years.
A
B
That's
should
be
clear,
it's
clear,
it
is
it's
already
doing
all
of
this,
and
we
just
like
we're
just
going
to
talk
about
like
how
it
helps
with
machine
learning
specifically
here.
So
here's
like
a
really
high-level
picture
right,
like
I,
want
to
give
this
message
across,
and
it
may
be
obvious
to
some
of
you,
but
I
just
want
like
to
learn
it
a
little
bit
more
because
I
want
people
to
think
of
cameras.
This
way
in
that,
like
Kiera's,
is
abstracting
out
a
whole
lot
of
things.
B
It's
abstracting
out
your
your
your
cloud,
whether
like
a
public
cloud
or
a
private
cloud,
you
don't
have
to
worry
about
that,
because
there
are
people
worrying
about
it,
and
so
you
have
like
separation
of
responsibilities
there,
and
then
you
have
a
variety
of
hardware
resources
that
you
might
want
to
consume.
You
may
not
want
to
consume
all
you
may
want
to
consume
some,
but
you
may
only
like
extended
to
consume
something
else,
so
curious,
abstracts.
B
All
of
that,
and
then
you
have
a
whole
bunch
of
like
operating
systems
and
like
kernels
and
versions
and
so
on.
It's
like
that's
another
whole
beast
and
like
hose
demon
and
curious,
provides
you
primitives,
where
you
can
have
separate
teams
managing
that,
and
still
like
herbalist
continues
to
work
for
you
and
it
like
shields
your
end-users,
in
this
case,
like
data
scientists
or
say
your
ml
practitioners
from
all
those
churn,
that's
going
on
right
like
they
don't
want
to
know
all
of
that.
B
Why
should
they
know
that
and
then,
like
you,
have
the
cold
cubed
as
itself,
which
I'm
not
going
to
get
into
in
this
talk,
but
hopefully
like
people
have
some
idea
of
what
Carell
is
this?
In
addition
to
that,
you
also
get
a
whole
bunch
of
other
primitives
that
are
very
important
for
managing
infrastructure
at
scale
like
you
get
primitives
like
authentication
and
authorization
with
our
back,
you
get
like
built-in
monitoring,
that's
extensible
to
almost
all
of
your
applications.
B
You
get
like
built-in
login
code,
our
namespaces,
so
these
are
like
a
common
set
of
functionalities
that
you
can
apply
for
a
whole
bunch
of
workloads
right.
So
now,
given
that,
like
you've
gotten
to
this
level,
that
you
have
this
common
infrastructure
that
you
can
deploy
on
now,
what
are
you
going
to
do
like?
The
more
interesting
part
is
like
running
the
real
applications
right
like
that's.
That's
the
reason
why
all
this
infrastructure,
you
can
access.
So
you
can
go
and
like
choose
your
application,
we're
not
like
restricting
you
to
some
specific
set
of
applications.
B
B
Coronas
doesn't
restrict
you,
but
the
good
part
is
it's
pretty
easy
because,
like
all
the
rest
of
the
problems
are
solved,
so
you
can
have
like
individual
groups
of
people
focusing
on
just
that
and
they
don't
have
to
worry
about
the
rest
of
the
infrastructure.
That's
pretty
cool
right!
Like
you
have
you
have
a
large
organization,
and
now
you
know
how
they
can
all
like
cooperate
with
each
other
and
still
add
value
to
to
the
rest
of
the
direct
consumers.
B
So
in
this
case,
then,
consumers
are
most
likely
going
to
be
data
scientist
and
they
are
going
to
focus
on
their
workflows,
which
is
like
say
Jupiter
or
it's
like
10,
suppor,
cafe
or
whatever
it
is.
They
already
think
about
that.
They
don't
think
about
the
rest
of
the
infrastructure
right
and
that's
sort
of
the
Google
model
and
like
it's
totally
possible
with
curators,
and
we
are
starting
the
journey
now
so
like
specifically,
Cuba
dates,
already
supports
accelerators
in
an
extensible
manner.
In
that,
like
you,
don't
have
to
wait
for
a
team
to
support
everything.
B
Gpus
are
already
supported
and,
like
there's
more
and
more
enhancements
happening
along
that
and
in
2018
there
will
be
having
like
performance
optimizations
and
all
of
that,
but
oneness
and
there's
also
going
to
be
support
for
fu
GS
and
like
high-performance
snakes
and
so
on.
So
actually
is
like
enabling
all
sorts
of
extensions
that
unblocks
you
from
doing.
B
You
can
just
like
you
can
become
productive
in
a
day
or
two
with
a
quitter
with
the
pre-existing
cumulus
cluster,
rather
than
like
going
and
buying
a
machine
from
like
from
from
a
store
nearby
and
like
starting
to
install
iOS
and
all
then
spending
like
a
month
and
then
now
you're
starting
to
go.
Do
the
cool
thing
trial.
You
don't
have
to
do
that
like
this
is
already
a
solved
problem
and
it
scales
to
like
thousands
of
nodes.
B
So
if
you
get
to
the
point
where
even
at
one
many
experiments
at
the
same
time,
then
you
can
actually
do
it
with
kubernetes
and
and
if
you
are
running
in
a
public
cloud,
for
example,
you
you
don't
have
to
provision
for
your
peak
utilization
like
you
just
go,
use
your
single
machine
and
when
you're
ready
to
like
go
and
launch
say
a
thousand
experiments,
the
resource
is
there
for
you
and
cumulus
still
works.
You
don't
have
to
ship
to
a
different
platform
and
container
packaging
is
already
a
standard
right
like
you.
B
Don't
again,
you
don't
have
to
reinvent
the
wheels
there.
So
all
these
things
are
common
problem
for
machine
learning
as
well,
and
criminals
would
solve
that
for
you
and
there's
actually
more
right,
like
I
like
auto
scaling,
so
auto
scaling
is
like
working
in
like
most
criminals
managed
services.
Today
we
have
support
for
like
priority
and
eviction,
but
I
like
say:
if
you
have
an
intern
and
you
it's
ain't,
our
hiring
season
and
you
like
hundred
interns
in
your
company.
B
Each
one
is
like
doing
an
m/l
experiment,
but
you
have
the
shared
resource
and
you
have
a
researcher
who's
having
a
conference
deadline
and
they
have
to
get
there,
get
their
training
and
like
get
their
models
built.
Are
you
gonna?
Have
the
interns
come
and
like
ruin
their
conference?
No
right
like
do
you
how
you're
gonna
create
a
separate
system
for
that?
B
Why
Kunis
provides
all
those
parameters
for
you
already
where
you
can
set
priorities
for
individual
users
and
and
make
sure
that,
like
a
low
priority
user,
it
does
not
go
in
and
like
take
away
all
the
other
resources
that
are
meant
to
be
consumed
by
more
important
peoples
or
more
important
things
that
are
happening
within
your
organization,
be
also
I'm
going
to
I'm
just
going
to
state
this.
But
if
you
have
more
questions
about
this,
I
will
talk.
I
can
unhappy
til
I,
explain
further
after
the
talk.
Kiran
is
assuming
adequate
network
bandwidth.
B
So
the
sort
of
a
fundamental
distributed
system,
design
assumption
that
Google
has
evolved
to
over
time.
So,
like
our
general
recommendation,
is
like
Oh
Trojan,
your
network.
Don't
try
to
solve
like
locality
issues
too
much.
So
that's
the
kind
of
model
like
upstream
is
moving
and
moving
towards
and
it
was
already
in
that
direction.
But
we
still
support
data
gravity
for
scenarios
where
you
don't.
B
So
that's
the
main
message
here,
but
right
you,
you
want
to
use
machine
learning
on
cubed
as
I
talked
about
all
the
awesome
infrastructure
stuff
that
like
cubes,
can
provide,
but,
like
I
said
earlier,
data
scientists
or
ml
practitioners
do
you
know
want
to
see
their
infrastructure
right
like
they
don't
want
to
learn
the
cumulus
api
is
like.
That's
that's
not
like.
That's
not
the
specialization
right
like
that's,
not
the
thing,
that's
most
fun
for
them.
So
are
you
going
to
like
tell
them?
B
Hey,
go
on
Canada's
goal
and
docker
Golan
containers
and
then
like
now
now
we
can
start
doing
whatever
you
want
to
do
like
look
at
the
sheer
number
of
things
that
you
have
to
learn
in
order
to
start
becoming
effective
in
using
cumulus
for
machine
learning.
It's
like
I
mean
it's.
It's
a
firehose
right,
like
you,
probably
have
to
do
like
ten
Coursera
courses
before
you
can
like
even
get
to
the
point
of
like
doing
some,
some
some
interesting
machine
learning
in
production.
You.
A
With
that,
we
are
proud
to
announce
cube
flow.
We
are
not
talking
about
this.
Incredibly
broadly
just
you
here
and
those
on
the
stream.
We're
gonna
be
talking
a
lot
about
a
lot
more
shortly,
but
the
summary
is,
we
want
to
make
it
easy
for
everyone
to
learn,
deploy
and
manage
portable
distributed
ml
on
kubernetes
everywhere,
just
like
kubernetes.
We
very
very
strongly
believe
that
this
should
work
everywhere
in
the
world.
Not
you
know
it.
On-Premises,
on
your
laptop
on
your
cloud
provider.
A
A
Speaking
of
the
box
inside
the
box
at
the
start-
and
we
are
just
getting
started
here-
we
give
you
a
Jupiter
hub.
Thank
you
very
much
to
the
folks
at
Jupiter.
I
saw
you
be
back
there.
I
tensorflow
training
controller,
which
Auto
scales
depend
or
not.
It
doesn't
auto
scale
it
deploy
scales
based
on
what
you
have
available
to
it.
A
So
if
you
have
CPU
or
GPU
or
multi-gpu,
you
can
configure
it
very
easily
tensorflow
serving
deployment
and
the
wiring
and
service
endpoints
between
them
to
make
it
very
easy
to
get
going
out
of
the
box.
And
what's
in
the
box,
you
saw
this
earlier.
It
is
just
these
steps
in
it
over
time.
We
would
love
to
expand
it.
We'd
love
to
make
it
each
step
deeper,
offer
a
lot
more
options,
but
that's
up
to
the
community
and
the
people
who
want
to
take
it
down
using
cube
flow
will
ideally
look
like
this.
A
B
All
right
so
I'm
gonna
try
to
do
a
live
demo.
I'm,
not
sure
how
this
is
exactly
gonna
turn
out,
but
let's
see,
hopefully
the
demo
God
said
to
me.
So
what
I
have
here
is
a
mini
Q
cluster.
So
you
can
make
sure
that
it's
obvious
that
it's
a
make
your
pleasure
I'm,
going
to
what
I'm
gonna
already
have
the
cube
floor
upholstery
cloned
so
I
mean
I,
could
probably
like
show
the
cube
floor
upholstery
for
you
just
to
give
you
an
idea
of
like
what
we
mean
by
cube
flow.
B
Here
we
go
so
this
is
a
repository.
Have
cloned
this
I'm
at
the
base
directory.
What
I'm
gonna
do
is
I'm
gonna
apply
all
the
components
in
there.
Okay,
you
need
to
change
there.
We
go
so
I
just
went
ahead
and
like
deployed
cube
flow.
The
next
step
is
like
I'm,
going
to
show
what
quads
exits
in
here.
B
So
what
we
have
here
is
like
we
have
a
tensorflow
service
and
then
we
have
a
Jupiter
hub,
and
then
we
have
a
tensor
flow
operator,
I'm
going
to
go
and
like
try
to
show
how
we
are
going
to
use
each
of
these
things
in
just
a
bit,
but
I
just
want
to
show
this
is
one
is
in
the
Box
ice
part
of
cube
flow.
As
of
now
right.
The
next
step
is
like
I'm,
going
to
open
up
open
up
Jupiter
hull,
oh
god,
this
echo
is
like.
B
So
let
me
also
like
list
the
services
that
are
here
and
then
so
I'm
going
to
use
a
mini,
cube
command
for
opening
up
service,
which
is
mini,
cube
service
and
pointed
at
a
load
balancer,
and
so
here
we
go
I
I'm,
I'm,
opening
up
well
I
mean
cool.
Has
these
awesome
proxies
that
get
in
the
way
of
demos?
B
We
turn
off
those
proxies,
and
that
should
work
now
say
we
go
so
you
have
Jupiter
hub
just
working
out
of
box
for
you
writing,
and
by
default
we
have
like
a
dummy
authentication
system
which
you
can
like
replace
with
your
own
organizations.
Favorite
authentication
system
we'd
like
Google,
Earth
or
or
github,
and
so
on
so
let's
say
I
just
got
a
login
I'm
gonna
start
my
server
I'm
going
to.
We
have
two
options
now
with
ten's
flow,
you
have
a
CPU
and
a
GPU
image
since
this
mini
cube.
B
B
What
I'm
going
to
do
is
like
I'm,
going
to
I'm
going
to
run
a
benchmark,
there's
a
specific
reason.
Why
I'm
choosing
a
benchmark
for
this
demo?
It's
because,
if
I
had
to
like
go
ahead
and
like
try
to
train
a
real
model
like
Inception
or
rest
night,
it's
going
to
take
quite
a
while,
and
it's
probably
very
boring
for
you.
B
Although
the
benchmark
is
doing
is
like
it's
going
around
like
trying
to
run
a
wrestler
training
model,
if
you're
not
familiar
with
ResNet,
that's
you
can
just
totally
forget
about
it.
Just
consider
it
to
be
like
an
image
image
model.
It
can
predict.
What's
in
an
image,
and-
and
here
I'm
training
with
like
synthetic
data
like
pulling
data
from
the
cloud
in
a
demo,
environment
is
again
like
pretty
boring.
So
this
synthetic
data
here
is
all
like
part
of
the
other
standard,
tensile
benchmarks
and
what's
happening
here
is
like
behind
the
scenes
it's
it
has.
B
It
has
generated
some
some
synthetic
data
and
it's
starting
to
Train,
and
you
can
see
here
that,
like
the
training
rate,
is
like
literally
0.5
images
per
second,
so
it's
pretty
slow
because
I
got
like
a
laptop
which
is
not
even
like
connected
to
power,
so
all
sorts
of
like
awesome
things
are
happening
behind
the
scenes
to
keep
the
demo
going.
So
the
training
rate
is
like
pretty
low
here.
B
Don't
you
don't
have
to
like
worry
about
that
so
I'm
going
to
do
the
same,
cube
control
apply
here
right,
I'm,
literally
in
the
same
directory
on
my
laptop
and
I
did
keep
control
of
line.
The
next
step
is
like
I'm,
going
to
go
to
my
I'm
going
to
go
list.
The
services
here
and
I
have
the
same
pencil
flow,
Jupiter
hub,
load,
balancer
and,
as
you
can
see
here,
the
external
IP
hasn't
been
allocated
yet
so
we
go
to
like
wait
for
the
external
IP
to
be
allocated.
B
While
that
happens,
I'm
also
going
to
explain
to
you
about
the
operator,
which
is
the
other
part
of
of
this
right.
Like
Jupiter,
gives
you
this
interactive
workflow.
Where
you
can
train
interactively.
You
can
pull
down
data,
you
can
like
do
some
data
manipulation,
but
that's
not
great
when
you
want
to
take
it
to
production
right
like
when
you
want
to
take
it
to
production
like
auditing.
B
And
then
you,
you
use
this
declarative
API
to
go
ahead
and,
like
tell
curators
tell
this
operator,
they
like
go
in
and,
like
start
tense
flow
training
right
like
and
like
do
the
training
and
then
go
ahead
and
like
push
the
model
somewhere
else,
and
that's
all
like
automated
for
you
and
you
also
get
like
10
support
integrated
with
that
so
I'm
going
to
go
ahead
and
like
deploy
one
of
those
models
as
well.
I'm,
not
sure.
How
are
you
guys
able
to
see
this?
B
It's
probably
pretty
hard
to
read
during
the
demo
I'm
going
to
run
through
this,
but
like
please
go
ahead
and
like
take
a
look,
look
at
this
after
the
demo.
The
thing
that
I
want
to
highlight
here
is
like
I'm,
going
to
deploy
a
distributed,
tensor
flow,
which
is
considered
to
be
kind
of
hard,
and
that's
one
of
the
reasons
like
people
keep
playing
with
like
larger
and
larger
boxes.
B
So
here
you
have
like
distributed
tens
flow
where
you
are
configuring,
a
master
which
does
not
have
GPUs,
and
then
you
are
configuring,
three
workers
which
has
GPUs
and
then
you
have
a
parameter
server
right
and
you
are
expressing
all
of
this
in
with
a
declarative,
API
and
I'm
running
the
same
benchmark
like
the
one
that
I
ran
on
the
laptop
and
I'm
going
to
run
the
same
benchmark
and
I'm.
Also
like
going
to
go
ahead
and
like
deploy
this
in
my
GK
cluster.
B
We
see
a
bunch
of
pods
being
created
right,
so
I
went
ahead
and
like
I
went
ahead
and
like
applied
a
manifest
and
that
manifest
ended
up
creating
that
manifest
ended
up,
creating
a
custom
operator
specification
and
this
object
and
the
controller
watching
this
object
went
ahead
and
created
created
the
the
tensorflow
entities
that
was
expressed
through
the
declarative
API.
So
you
have
a
master
running
here.
B
Here
we
go
so,
let's
switch
back
to
the
Jupiter
side.
Why
are
we
doing
interactive
training?
And
now,
let's
say
that
I
have
come
back
to
work.
I
did
something
at
home.
I've
done
some
training
now
I
want
to
like
actually
take
it
to
the
next
level,
where,
like
I,
want
to
continue
doing
some
interactive
training,
but
I
want
to
like
start
crossing
lot
more
data
and
at
this
point,
I
need
more
memory.
I
need
more
CPUs.
I
need
more
GPUs
right.
So
at
this
stage.
B
Okay,
so
here
we
go
so
I
am
right.
Now
I
am
in
cloud.
I
went
through
the
same
process
of
like
setting
up
Jupiter,
but
like
Jupiter
cache
as
my
identity.
So
it
put
me
in
a
notebook
because
the
notebook
already
exists
right,
I'm
going
to
go
in
and
like
run
the
exact
same
benchmark,
but
this
time
I'm
going
to
run
it
with
GPUs,
so
I've
changed
the
I've
changed
the
command
line,
the
the
benchmark
command
line
here,
a
little
bit
to
make
it
work
with
GPUs.
B
It's
mostly
about
removing
a
bunch
of
other
options
that
made
it
work
with
CPUs.
But
the
point
that
I
wanted
want
you
to
take
here
is
that
you
can
see
like
in
just
a
second
you'll,
see
that
the
the
training
rate
is
like
so
much
faster
right
like
this
is
not
something
cubed
as
it's
doing.
It's
like
it's
just
a
power
of
like
GPUs
and
the
optimization
between
intense
flow
GPUs.
But
the
point
to
take
away
here
is
that,
like
your
stuff
is
portable
right,
like
everything
is
portable,
your
workflow
is
portable.
B
Everything
remains
the
same,
and
now
here
you
go.
You
move
from
like
point
five
images
to
like
200
plus
images
and
when
you
move
to
the
distributed
setting
which
which
we
are
seeing
with
the
operator,
your
your
productivity
level
improves
even
further,
let's
see
if
the
if
the
distributed
job
actually
completed.
If
not,
we'll,
probably
have
to
wait
for
that.
B
So
I'm
going
to
go
ahead
and
drill
into
the
pods
view
and
I'm
going
to
look
at
the
logs
for
the
pod
okay,
so
come
on,
did
run
successfully,
and
it's
finally
me
that
it
ran
successfully.
But
the
training
rate
here
is
like
240
images
per
second
right.
So
so
you
can
start
scaling
linearly
and
you
don't
have
to
worry
about
like
Numa
and
like
with
I'm
on
the
same
PC,
a
socket
like
screw.
B
All
that
you
don't
have
to
know
all
that,
like
just
assume
that
you
have
some
fine
items,
some
usable
amount
of
network
bandwidth,
then
just
keep
scaling
like
improve
your
level
of
productivity
right,
and
so
this
is
not
the
only
part.
The
other
part
of
your
on
shows,
like
let's
say
the
whole
point
of
this
demo
was
like
I
want
to
know.
If
it's
a
hot
dog
or
not
right
for
folks
who
know
what
hot
dog
or
not
means,
let's
say
that
I'm
being
very
nave
they're
like
I,
haven't
I,
don't
know
any
machine.
B
B
No,
we
want
to
make
it
smarter,
and
so,
let's
say,
let's
apply
some
machine
learning
right
in
this
case
I'm,
using
like
pre-existing
models
that
have
been
trained
for
for
doing
this
kind
of
task
and
if
you're
curious,
it's
like
thing
its
inception
model,
so
I'm
going
to
do
the
same
thing
right
like
I'm
gonna,
go
ahead
and
like
deploy
and
go
ahead
and
ask
it
like.
Is
this
a
hot
dog
with
a
real
hot
dog
image?
And
let's
see
what
it
says
the
same
thing?
B
Maybe
the
service
is
still
coming
up,
or
is
this
a
demo
fail
most
likely
a
demo
fail,
do
I
have
time
to
do
like
debugging
and
I
think
we're
all
I
guess
I
have
two
more
minutes.
So,
let's
see
so
let
me
show
what's
happening
behind
the
scenes,
while
the
demo
may
come
up
so
there's
a
model
service
here
right,
like
the
model
service.
What's
what's
going
on
at
the
mall
self,
is
like
it's
tense
flow
serving
and
it's
pre
set
up
for
you
there
with
the
model
server.
B
You
can
configure
it
to
to
make
it
point
at
like
your
own
model
in
this
case,
like
for
the
demo,
we're
making
it
point
that
a
model
that
is
hosted
on
GCS
for
us.
So
it's
it's
being
pointed
at
a
model.
That's
that's
stored
in
a
GCS
bucket,
but
imagine
that,
like
you
have
serving
available,
and
then
you
have
your
your
your
Jupiter
workflow
available.
You
built
a
model
right
and
the
model
is
like
stored
on
some
of
some
object,
store,
a
storage
or
like
your
local
laptop.
B
All
you
have
to
do
is
like
go
ahead
and
like
change
this
deployment
and
say
can
changing
the
command
line,
arguments
of
that
and
making
it
point
at
me
new
model,
and
then
you
got
serving
going
right.
You
don't
have
to
think
about
that.
You
have
to
do
that.
One
small
thing
and
that's
it.
Let's
see
if
let's
see
if
we
can
give
this
another
try
just
one
more
time,
David
yeah
the
still
book,
let's
see
if
the
smart
so
is,
actually
responds.
B
A
Okay,
so
we
are
incredibly
out
of
time,
but
I
will
be
very
quick,
so
oh
shoot
Thanks,
so
we
are
incredibly
out
of
time.
So
that's
it!
You
took
something
and
you
deployed
it
in
three
places,
but
but
to
be
honest,
that
the
message
here
is,
if
you're
a
data
scientist
you
didn't
have
to
think
about
any
of
that,
you
didn't
ever
think
about
kubernetes
or
learn
aid.
You
didn't
have
to
do
any
special
deployments.
We
care
of
it
for
you,
we
expanded
it
for
you.
A
You
began
to
really
decoupling
from
what
the
data
scientists
actually
do,
and
so
the
answer
is
yes,
for
now
we
are
just
getting
started.
We
would
love
for
you
all
to
come
and
help
contribute.
We
have
an
open
source,
repo,
github,
whack,
google
whack
cube
flow.
Thank
you.
We
we
really
are
just
getting
started.
We
want
this
to
be
a
community
where
we
come
together.
We
know
that
almost
everyone
here,
anyone
doing
ml,
has
written
some
custom
bespoke
solution
for
their
team.
A
A
A
B
B
So
on
on
an
on
a
laptop
like,
maybe
let's
ignore
the
laptop
on
a
workstation
there's,
one
part
pending,
which
is
enabling
installation
of
drivers
with
device
plug-ins
we're
working
on
that
so
you'll
be
working
with
upstream
community,
essentially
on
that,
so
that
part
is
spending.
But
if
you
move
to
any
cloud
providers
or
like
any
manish
cuban
is
it's
already
there
you
don't
have
to
you,
don't
have
to
think
about
it
right,
so
we're
working
Sabo.
A
Send
us
a
note
if
you're
having
problems
with
it.
Well,
we'll
find
you
the
right
person
in
the
future.
Yes,
based
on
the
idea
ear
here
and
again,
we
want
to
be
cloud
neutral
here.
The
question
was:
will
we
support
Tipu's,
which
is
for
this
dunno?
That's
Gus,
Google's
custom
TB
ml
chip.
As
you
know,
many
clouds
are
investing
in
various
custom
ml
chips
we
want
to.
We
want
to
create
a
real
abstraction
between
this
and
and
whatever
cloud
provider
would
like
to
participate
in
queue
flow
and
help
us
surface
those
custom
chips.