►
From YouTube: CDF - SIG MLOps Meeting - 2021-02-25a
Description
For more Continuous Delivery Foundation content, check out our blog: https://cd.foundation/blog/
B
Good
good
to
be
back.
A
Yeah
yeah
first
first
first
session
of
2021,
hopefully
they've,
been
treating
you
better
than
last
year,
did.
B
B
Good,
so
do
we,
I
didn't
really
think
of
anything
for
an
agenda.
Did
you
have
any
anything
to
cover?
I
could
see.
A
B
I'm
saying
hearing
a
weird,
weird
sort
of
echo
from
you.
I
don't
know
if
it's
my
end
or
not
echo
like
like
it's
a.
A
A
A
A
Yeah,
so
what
I
thought
we
should
probably
talk
about
today,
really
is,
is
just
what
the
outcome
are
for
this
year:
yeah,
what
we
think
the
key
pieces
of
work
are
that
we
need
to
do
and,
and
then
yeah
start
starting
to
think
about
we're
going
to
execute
on
that
and
what
we're
going
to
need
to
plan
for.
B
Yeah,
I
I
think
at
the
end
of
last
year
it
was
well
the
end
of
last
year
that
when
we
sort
of
took
a
recess,
we
were
starting
to
think
about
how
to
sort
of
get
the
word
out
there.
So
I
imagine
that
would
be
a
big
part
of
the
ames.
Here
would
be
a
more
wider
awareness.
People
are
discovering,
discovering
this
they're
discovering
that
the
road
map
and
things
and
are
very
interested
in
it,
but
yeah
there
could
be,
could
be
more
of
that.
So
I
guess
that
would
be.
A
Yeah
and
I
think
there's
this
two
element
that
we
will
consider
really
one
piece
is
the
the
sort
of
general
communication,
which
is
just
letting
people
be
aware
that
the
roadmap
exists
and
that
encouraging
people
to
understand
it
and
contribute
to
that,
and
obviously
that
probably
needs
a
lot
of
outreach
to
to
get
to
a
broad
audience.
A
The
other
element
is
actually
a
cdn.
The
negation
level
make
sure
that
we're
communicating
clearly
with
all
of
the
teams
who
are
working,
devops
related
science,
making
sure
that
they're
fully
engaged
and
then
considering
they're
in
the
road.
My
part
of
the
the
work
that
they're
doing
doing
on
their
own.
B
A
Yeah,
so
it's
a
it's,
it's
the
need
to
do
that
across
as
many
projects
both
within
the
the
cf
and
the
broader
learn
explanation,
and
because
we've
got
we've
got
a
spliff
project
x
at
the
moment
in
terms
of
most
of
the
machine,
learning
activities
have
gone
in
the
lex
foundation
and
more
of
the.
B
So
with
that,
like
one
possibility,
there
is
identifying
what
those
projects
are
just
taking
a
note
that
springs
to
mind
because
of
argo,
like
I've
thought
about
argo
a
bit
lately,
it's
got
a
little
bit
of
traction
in
more
switched
on
developery
data
science
teams.
B
Have
this
founder
argus
found
a
bit
of
a
nation,
that's
a
cncf
project,
so
that
would
be
one
that
it
would
be
relevant
to
do.
You
agree.
B
And
then
there's
the
what's
the
sub
calculation
to
do
with
ai
in
linux
foundation.
It's
just
called
the.
B
There's
the
lfai
foundation,
open
source
innovation,
afterwards,
machine
learning.
B
Yeah
lf
lf.
Why
can't
I
paste
that
just
lf
ai
data
dot
foundation,
I'll
paste
it
in
the
chat
here
is
the
sub
foundation.
B
B
Otherwise,
they're
not
ones,
I'm
that
familiar
with,
but
yeah,
there's
yeah,
there's
a
bunch
of
things
within
the
linux
foundation
that
are
interesting.
A
Yeah,
I
think,
there's
a
fair
work
that
we
still
need
to
do
within
the
cdf
projects
as
well.
A
In
fact,
we've
now
got
a
new
sig,
which
is
specifically
looking
at
sort
of
platform
on
artifacts
and
obstructions
the
the
the
events
which
is
you
know
intended
to
try
and
line
all
of
the
the
various
tooling
again
against
a
standard
set
of
ways.
Thinking
about
managing
dev
activities
he's
in
indian
program.
A
So
clearly,
we
need
to
be
given
guidance
in
those
areas
around
what
the
specialist
needs
are
for
them.
Ops
within
the
broad
abroad,
devops
workflows.
B
So
it's
basically
it's
like,
for
example,
for
technology
making
sure
it's
possibly
suitable
for
ml
ops.
Workloads
like
it's
that
it
like
an
example,
would
be
that
it's
people
are
running
very
long
running
things
handling
things
that
deal
with
large
amounts
of
data
and
so
on.
Whatever
the
specifics
are,
is
that
the
idea
is
that
it.
B
A
Takes
a
special
case
because
obviously
we're
using
tekton
on
within
in
jenkins
x
and
and
have
been
doing
ml
ops
capabilities
on
top
of
that
through
gently
x.
A
So
there
are
things
that
we
can
suggest
in
those
areas
to
help
help
improve
our
abilities
to
manage
larger,
datasets
and
yeah,
where
we've
run
up
against
some
of
the
boundaries
of
what.
What
techdon
do,
because
working
with
chunks
of
data
and
or
very
very
resourcefully
means
that
they're
trying
to
work
work
with
it.
B
A
B
What
about
any
changes
to
the
roadmap
itself,
any
things
that
have
happened,
sort
of
in
the
last
few
months?
That
might
be
interesting.
If
anything,
you
just
think
some
of
the
some
of
the
points
in
there
are
even
more
relevant
than.
A
I'm
not
expecting
large
amounts
of
change
this
year,
yeah,
I
I
don't
think
they're
they're
they're
hugely
leaps
forward
to
date,
but
but
hopefully
the
the
activity
that
we
do
in
terms
of
promoting
the
roadmap
will
encourage
teams
just
to
start
working
on
some
of
the
challenges
that
we
identified
last
year.
A
There
are
some
refinements
that
we
make,
in
terms
of
you,
know
better
communicating
the
the
scale
of
some
of
the
issues
and
making
sure
that
it's
clear
what
the
what
the
cost
involved
are,
and
why
why
you
would
need
improved
tooling
against
areas.
A
There's
more
data
coming
out
where
there
is
a
respect
to
actual
costs
of
of
running
some
of
the
larger
projects.
B
Like
like
gpt3
was
in
the
news
a
lot
about
how
many
millions
of
dollars
it
costs
to
trade.
A
B
That's
as
I
guess,
that's
an
exceptional
case
and
sort
of
a
singular
thing,
because
it's
trained
once,
but
I
don't
know
what
the
cost
of
inference
is
with
it.
I
kind
of
mentioned
it's
terribly
high
compared
to
that,
but
yeah.
It's
certainly
real
dollars
and
real
world
impacts.
A
Yeah
yeah,
so
I
I
think,
starting
to
provide
a
better
view
around
those
sorts
of
metrics
would
be
helpful
for
and
we
may
be
able
to
reach
out
to
some
other
people
and
and
and
get
some
suitably
sanitized.
This
sort
of
guideline
metric
to
to
help
people
to
understand.
B
Yeah,
so
that's
possibly
a
new.
A
new
thing
is
the
as
things
get
used
more
as
the
impact
of
things,
the
it's
not
no
longer
theoretical,
but
it's
more.
These
things
are
expensive
to
train
or
can
be.
A
So
you
know
we've
identified
a
number
of
real
world
challenges
and
you
know
if
we
can
back
up
with
with
with
sizing
information.
In
terms
of
you
know
what
the
painting
point
that
is
really
really
costing
now
is
that
that
starts
to
paint
a
picture
of
what
the
opportunity
in
in
that
space
actually
building
building
the
solutions,
and
you
know
clearly,
people
want
to
be
working
on
on
projects
that
that
other
gonna
find
useful,
and
I
ideally
would
do
so.
A
You
know
in
a
way
that
will
generate
rates
of
revenue
to
support
those
teams
working
on
those
projects.
B
One
other
thing
I
thought
would
be
interesting
is:
if
we're
gonna
encourage
more
people
to
be
involved
in
more
discussion,
where's
the
best
place
for
doing
it
in
text
form
or
asynchronously.
Is
it
the
slack
channel?
Someone
was
asking
whether
we
should
use
github
discussions
like
do
you
have
any
thoughts
there.
A
So
I'm
keen
that
we
we
get
the
traffic
up
a
bit
in
the
in
the
slack
channel
and
start
to
encourage
of
more
regular,
ongoing
conversation.
A
I
I,
I
think
we
need
to
stay
focused
on
making
sure
that
the
capabilities
for
doing
it
are
there
there
in
the
first
place,
there
are
lots
of
good
resources
out
there
already.
A
A
I
think
we
certainly
need
to
make
sure
that
you
know
our
first
priority.
Is
everyone
know
that
we
are
starting
to
work
again
and
get
get?
People
excited
excited
motive,
to
contribute.
B
B
A
B
Yeah
and
and
we're
doing
it
monthly
at
this
time,
is
that
right
perform
for
the
next
month
or
two
yeah,
and
so
I've
got
it
as
the
well.
I
put
it
in
my
calendar,
so
it's
the
for
now
it
was
the
fourth
thursday
in
the
month
or
something
yeah
monthly.
On
the
fourth
thursday,
is
what
my
calendar.
What
google
calendar
tells
me
it's
on,
which
is
okay
with
me,
yeah.
A
Also
I'll
clone
a
2021
draft
roadmap,
so
let's
just
start
doing
requests
again
again.
B
B
Yeah
that'd
be
good,
so
in
terms
of
general
awareness
and
and
outreach
did
you
have
any
thoughts
on
blogs
or
or
sites
to
approach,
to
mention
it
and
how.
A
So
I
I
think
the
the
most
powerful
tool
will
be
trying
to
get
as
many
conference
slots
as
we
can
across
a
range
of
differences.
B
I
think
there's
probably
some
broader
devopsey
ones
that
might
be
interesting
too.
To
that
would
draw
attention
to
it.
I
think
it
would
be
relatively
easy
sort
of
pitch
to
talk
about,
because
you
could
just
take
a
subsection
of
it
and
find
the
most
relevant
things
to
talk
about,
because
a
lot
of
it's
quite
interesting
to
people
things,
people
don't
think
about
and
the
either
the
ethical
angle
or
the
you
know
something.
B
That's
in
the
news,
so
yeah
I'll
definitely
come
up
with
a
think
of
someone
come
up
with
a
list
of
things
and
maybe
even
submit
a
few
things
that
doesn't
sound
like
it
would
be
a
huge
amount
of
work.
It's
easy
to
talk.
A
A
As
you
say,
if
we
start
doing
regular
blog
work
interviews
and
things
like
that,
I
did
a
slot
a
month
or
so
back
the
the
the
cdf
sort
of
video
broadcast.
B
In
other
areas,
is
you
know,
since
we
talked
last
there's
there
been
any
interesting
developments
sort
of
in
the
field
like
it
to
me?
There's
lots
of
interesting
applications
of
it.
There's
different
companies
have
started
and-
and
things
like
that,
but
I
haven't
seen
any
sort
of
earth
shattering
developments.
It's
just
progress
continues.
I
have
come
across
some
people
that
share
sort
of
my
opinion
that
that
some
of
us
think
it
will
be
easier
for
developers
to
learn
a
bit
of
data
science
than
data
science
scientists
to
learn
development.
B
So
it's
interesting
to
hear
some
other
people
have
that
view
sort
of
analogous
to
the
devops
thing.
There's
there's
always
a
shortage
of
data
scientists
in
the
world
and
then
not
not
that
there's
a
surplus
of
developers,
but
developers
are
usually
going
to
be
around
anywhere.
You
would
want
a
data
scientist
so
yeah.
I
think
we
might
hear
more
about
that
and
tools,
sort
of
specializing
in
one
or
the
other,
so
that'll
be
interesting
to
see.
That's
definitely
something.
I've
noticed.
A
B
Is
that
related
to
auto
ml,
or
is
it
the
other
name
for
that.
A
Well,
well,
I
think
it's
more
that
people
are
looking
at
some
of
the
the
sort
of
you
know,
service,
movement
and-
and
you
know
the
some
of
some
of
some
of
the
activities
in
the
code
space
switcher.
You
know
you
know
intended
really
dumb
down
very
standardized
operations
so
that
it's
all
figuration
and
no
coding
right.
A
A
Big
gap
between
a
very
specialist
tool
that
do
washing
learning
thing
very,
very
easy
and
and
then
a
huge
gap
gap
between
being
able
to
do
that
and
being
able
to
actually
build
a
end-to-end
product
includes
in
a
production
environment.
A
Yeah,
it
doesn't
look
like
things
are
improving
for
much
in
in
terms
of
the
quality
of
production
option
machining
asset
that,
if
anything,
we're
increasingly
looking
like
we're
heading
into
another
winter.
Due
to
the
the
the
number
of
failed
hell
products
out
there
that
yeah
yeah.
So
I
think
the
the
roadmap
maps
actually
going
very
important.
B
I
think
we'll
see
more
and
more
of
the
application
of
things
to
mundane
sort
of
solutions,
things
that
it's
not
machine,
learning
isn't
really
the
key
feature
of
it,
though,
but
it
just
adds
another
improvement
to
things
a
bit
like.
I
guess
when,
when
apps
started,
getting
web
app
front
ends
to
things
that
weren't
web
web
fronted
before
they
weren't
all
about
the
web
app,
but
it
was
just
one
one
piece
of
technology
in
the
mix.
B
It
might
be
like
that
with
machine
learning
for
some
time
where
it's
just
enhancing
things.
My
experiences
are
certainly
getting
easier
to
to
train
things.
B
As
you
know,
there's
more
power
on
on
tap
the
sort
of
models
that
I
deal
with
are
much
smaller
than
the
ones
that
sounds
like
you,
you
and
others
deal
with
at
the
higher
end,
especially
with
imaging,
but
for
the
more
structured
data
sort
of
enterprise
or
business
sort
of
data,
then
I
think
it's
definitely
getting
easier
to
apply
so
it'll
it'll
be
another
tool
in
in
the
tool
belt.
A
B
Yeah
yeah.
Definitely
there
was
some
of
some
of
that
floating
around
last
year.
There's
been
lots
more
in
the
shorter
term
about
where
things
around
facial
recognition
can
and
can't
be
used.
B
The
way
they're
being
abused,
there's
all
sort
of
obvious
ones
like
that
there
was,
I
think,
one
of
the
things
we
we
laughed
at
last
year
before
we
sort
of
took
recess,
was
the
the
law
of
laws
we
were
proposing
in
australia
around
regulating
the
linking
to
news
from
from
google's
and
facebook's,
and
so
on
that
got
amended
and
passed
here
in
some
useless
form,
but
it
did
at
some
point
the
legislation
mentioned
about
algorithms
and
how
the
the
the
news,
the
news
companies
were
supposed
to
be
notified
of
algorithm
changes,
which
I
thought
was
novel,
because
a
lot
of
those
algorithms
are
actual
machine
learned.
B
You
know
sort
of
continuously
trained
models
based
on
you
know
daily
or
continuous
training
runs
or
weekly
so
like
when
you
have
some
legislation
that
says
something
like
an
algorithm.
Does
that
extend
to
a
model,
and
is
it
the
does?
That
then
include
all
the
data
that
the
the
private
data
that
derives?
B
The
model
and
so
on,
there's
a
lot
of
things
that
even
expert
practitioners
don't
get
the
subtleties
of
that
like
if
you've
got
a
piece
of
law
that
says,
google
has
to
tell
news
limited
when
they're
sorting
algorithm
for
daily
news
changes.
It's
like.
Well,
we
don't
know,
but
that's
likely
evolving
and
non-trivial
about
it,
machine,
learning
and
and
and
data
that
feeds
that
model.
It's
not
an
algorithm
that
someone
cooked
up
and.
A
B
B
B
But
when
it's
models
that
are
trained
from
a
significant
amount
of
data,
then
you
know
the
data
becomes
part
of
the
algorithm.
It's
not
just
static
code
that
you
can.
You
can
do
and
not
give
away
any
external
secrets
or
anything
like
that,
or
it's
just
yeah,
it's
hard
to
explain
it's,
this
sort
of
lay
understanding
of
computing
that
that's.
B
It
gets
ugly
when
it
gets
mixed
in
with
regulation,
and
it's
not
just
it's
not
just
lawyers
and
legislators,
it's
I'd,
say
90
to
95
of
developers,
wouldn't
know
that
you
know
google's
search
algorithm
or
something
like
that
is
fundamentally
machine,
learning
or
involves
machine
learning
and
and
a
bunch
of
things
they
use
every
day.
They
would
not
think
that
it
involves
it
so
yeah,
it's
it's.
If
it's
challenging
for
insiders,
then
how
can
we
expect
the
mainstream
to
be
able
to
understand
that
subtle
concept?
B
I
saw
in
the
chat
carsten
posted
a
a
suggestion
about
someone
who
he
was
in
there
he's
in
the
chat
room
who
could
in
the
city
in
the
cdf
slack
that
could
promote.
So
I've
taken
a
note
of
that
too.
So
she's,
one
of
the
persons
who
created
the
animals
or
yep.
A
Yeah,
oh
yeah,
I've
just
just
seen
that
yeah
so
larissa
is
he's
joining
us
this
this
year,
I
would
say
she
was
going
to
be.
We
had
a
10a
but
clearly
she's
had
another
commitment,
yeah
we've
we've
had
about
some
of
that
need
to
be
done
and
she
was
keen
to
get
involved.
B
Yeah
and
it'd
be
good
it'd,
be
good
to
encourage
people
in
general
to
pop
up
in
the
slack
and
we're
all
in
different
time
zones.
So
it's
always
nice
to
see
some
asynchronous
discussion
there,
and
even
if
people
just
post,
interesting
things
that
are
going
on
in
in
their
world,
then
that's
that's
useful.
I
think
we'll
get
more
value
out
of
it.
That
way.
So
I
think
that
that
should
be
something
we
try
and
promote
this
year
is
the
slack.
I
agree:
that's
a
great
way
to
chat
it's
fairly
low.
Friction.
B
You
can
sort
of
read
it
if
you
want
like
a
forum,
but
it's
seems
less
work
to
keep
up
to
date.
So
yeah,
hopefully
you
can
get
some
more
action
happening
there
and
that'll
be
good.
I
might
have
to
sign
off
now,
so
maybe
we'll.
B
I'll
start
looking
at
conferences
to
propose
to
and
as
I
find
things
I'll
just
throw
them
in
the
slack
room
and-
and
I
might
even
suggest
certain
certain
talks
and
things
so.
B
B
All
right
well
good,
catching
up
again
and
good
to
talk
and
see
you
next
time
and
on
the
slide.