►
From YouTube: LILYPAD: Why Distributed Compute is the Next Frontier - Ally Haire - Filecoin Unleashed Paris 2023
Description
LILYPAD: Why Distributed Compute is the Next Frontier for Filecoin & Web3 - presented by Ally Haire at Filecoin Unleashed Paris 2023 - https://filecoinunleashed.io/
Link to the video mentioned in the presentation: https://www.youtube.com/watch?v=aK12PRx8V0k
A
I'm
a
developer
relations
engineer
with
the
filecoin
foundation
and
today,
I'm
really
excited
to
share
a
super
impactful
and
valuable
project.
That
I've
been
had
the
pleasure
of
working
on
for
the
last
few
months
and
it's
called
the
lilypad
project
or
well
the
lilypad
network
now
so
yep.
A
That
is
a
hint
just
want
to
take
a
moment
to
recognize
the
team
behind
This,
brilliant
work
and
who
I
wouldn't
be
able
to
give
this
talk,
without
so
shout
out
to
our
amazing
engineering
team
in
Luke
and
Kai,
and
our
headbreed
Searcher
Levy,
who
I
believe
is
here
and,
of
course,
to
David
for
his
leadership
here
as
well.
So
quick
agenda
for
today,
I
want
to
run
through
where
this
fits
into
the
filecoin
vision.
A
Then,
of
course
fill
you
in
on
the
aim
of
the
lily
pad
Network
and
there's
that
word
again
and
then
I'll
hop
straight
into
a
code
demo
to
show
you
how
to
use
lily
pad
and
finally,
I'm
going
to
finish
off
with
some
thoughts
on
why
I
think
distributed.
Compute
networks
like
lilypad,
are
the
future
of
filecoin
and
of
web3,
and
some
of
the
applications
I
think
it's
going
to
empower
so
just
to
set
the
scene.
A
This
decentralized
internet
is
one
that
we
want
to
allow
for
the
democratization
of
data
and
access
to
data,
so
non-dependent
on
restraints
of
geography
of
resources
or
of
level
of
access
to
services
and
one
that
not
only
solves
some
of
the
big
social
and
ethical
challenges
of
the
kind
of
web.
2
Cloud
stack,
but
also
some
of
the
engineering
challenges
so
including
pricing,
resilience
and
distribution.
A
So
we've
so
far
built
out
the
largest
decentralized
storage,
Network
globally,
an
impressive
feat
in
itself
and
we've
also
layered
on
then
this
powerful
programmability
with
fbm
and
I
know
the
filecoin
Saturn
retrieval
Market
is
well
underway
to
allow
for
faster
movement
of
data
across
the
globe.
I
know
they're
here
too
I
just
saw
them
so
welcome
and
the
fourth
pillar
of
this
mission
is
computation
and
to
me
this
is
kind
of
where
data
becomes
truly
useful.
So,
after
all,
an
llm
is
nothing
without
the
ability
to
kind
of
train
and
compute
on
data
sets.
A
So
it's
really
my
strong
view
here
that
distributed
computing
is
going
to
unlock
this
next
generation
of
really
web
scaled
decentralized
applications
and
bring
this
like
power
of
robust
compute
to
web3,
and
this
is
where
projects
like
bakuyao
and
lilypad
live.
So
in
case
the
name
bacquiao
isn't
familiar
to
you,
though
I'm
probably
is
facial's
a
peer-to-peer
and
open
computation,
Network
or
a
network
of
nodes
which
provides
this
platform
for
public,
transparent
and
optionally.
A
Verifiable
computation,
I
won't
say
too
much
about
this
one,
though,
if
you
want
to
hear
more
about
it,
I
recommend
you
stick
around
for
David.
Aronchick's
talk
later
he's
the
expert
here.
He
might
also
talk
about
cheese,
one
of
his
favorite
things
and
we
are
in
France
after
all.
So
if
bakuyao
is
a
distributed,
compute
Network,
though
what
is
lilypad
well
lilypad
leverages
the
power
of
balkia
and
then
it
combines
it
with
on-chain
on
chain,
sorry
guarantees
and
crypto
economic
incentive
models.
A
So
with
lilypad
V1
we're
really
aiming
bigger.
We
want
to
build
this
trustless
distributed
compute
Network
that
really
enables
this
internet
scale,
data
processing
with
AI,
ML
and
other
arbitrary
compute,
and
this
is
especially
kind
of
interesting
when
you
think
about
the
storage-
and
you
know
large
data
sets-
we
already
have
on
filecoin.
A
It's
called
is
just
a
solidity
implementation
for
arbitrary
untrusted
nodes
to
join
networks
and
for
users
to
run
compute
jobs
via
lilypad
modules,
so
we've
Loosely
based
it
on
modicum,
which
is
a
research
paper
which,
for
both
this
mediator
approach
to
resolving
consensus
of
deterministic
jobs.
So
this
is
kind
of
the
current
infrastructure.
At
the
moment,
this
allows
kind
of
coordination
layers
to
match
the
kind
of
CPU
GPU
Demand
with
supply
of
jobs
or
users
that
want
to
run
compute.
A
A
So
let's
see
a
demo
because
that
always
speaks
louder
than
I,
possibly
could
so.
This
is
the
first
demo
which
this
is
the
basically
the
hello
world
of
Lily
Pad,
except
it's,
the
hello
cow
world
of
lilypad
and
this
basically
Works
via
a
clay
I'm
not
going
to
show
it
now
because
I
think
I'm
a
bit
over
time.
A
Look
something
like
this
when
you
open
it
up,
depending
on
what
prompt
you
put
in
in
this
case,
it
was
hello,
Lily
Pad,
but
you
know
maybe
utterly
fantastic-
is
a
good
one
to
go
with
I,
think
you'll
agree
and
that's
basically
it
and
it's
not
just
a
clear
implementation.
You
can
also
run
this
from
a
smart
contract.
Again,
our
next
module
in
the
lilypad
suite
currently
like
you
can
make
any
module.
You
want
so
long
as
it's
deterministic
to
run
on
our
Network.
A
The
next
module
we've
designed
is
our
stable
diffusion
demo,
and
this
is
on
SD
XL
version
nine.
So
this
is
the
clay
version
of
that
demo
and
we
also
have
a
smart
contract
version
which
hopefully
will
run
nope.
Okay,
well,
the
video
can't
be
located,
but
trust
me.
No,
we
do
have
a
video
it's
up
on
our
YouTube
and
I'll
link
to
that
a
bit
later,
but
basically
for
the
stable
diffusion
of
to
run
from
a
smart
contract.
You
implement
your
code
with
an
interface,
and
that
is
down
the
bottom.
There.
A
The
function
receive
job
results,
and
then
you
just
call
this
Lily
Pad
module,
like
I,
showed
in
the
clear
you
input
your
prompt,
your
string
for
what
you
want
to
get
back,
and
this
will
return
to
your
smart
contract,
a
lily
pad
well
an
ipfs,
sorry
CID,
and
when
you
open
that
you
will
have
a
you
know,
maybe
a
rainbow
unicorn
in
this
case
or
an
astronaut
riding
a
rainbow
unicorn
anything
you
put
in
there.
Basically
pretty
impressive
results
as
well.
A
Okay,
the
next
thing
we're
actually
making
is
fine
tuning
a
model
with
functioning
a
stable
diffusion
model
with
Laura.
So
this
actually
allows
for
input
data
to
go
into
your
model,
and
then
it
will
output.
You
know
a
result
from
that.
So
this
is.
This
is
a
pretty
big
step
forward.
We
think
for
distributed
compute
networks.
We
didn't
quite
get
around
two
getting
that
one,
but
it
will
be
available
check
out
our
docs
very
shortly.
A
You
can
also
obviously
run
your
own
node
on
this
network.
It's
a
network,
so
you
want
to
be
able
to
get
paid
for
jobs
that
you
run
on
your
spare
GPU
or
you
know
CPU
at
home,
or
maybe
you
have
a
big
data
center.
Some
people
might
have,
so
you
could
run
this
there
and
earn
money
from
running
those
jobs.
A
Centralized
entities
should
not
control
access
to
AI
advancements,
or
at
least
that's
my
opinion
or
the
data
involved
in
them,
and
common
human
Goods
should
really
not
be
gate
kept,
and
this
is
where
decentralized
AI
comes
into
play
and
that
we
think
the
lily
pad
Network
or
distributed
compute
networks
like
it
will
be
able
to
help.
A
So,
what's
interesting
is
that,
according
to
this
like
leaked
Google
article
as
well,
it
was
also
clear
that
open
source,
Ai
and
ml
well
excuse
me
is
already
out
competing
the
centralized
versions
and
that's
because
they're
able
to
Define
use
cases
and
fine-tune
them
more
easily
and
on
less
data.
So
it's
really
interesting
that
this
came
out.
I
was
like
so
engrossed
in
it
I'm
like
wow.
It's
really
cool
that
we
are
beating
these
big
players
that
take
years
and
years
to
compute
these
data
models
with
fine-tuning
of
them.
A
One
of
the
main
problems
for
this,
though,
or
one
of
the
issues
for
open
source
AI,
is
this
access
to
gpus,
and
this
is
a
key
problem
for
open
source,
Ai
and
ml
to
be
competitive
and
deep
in
or
decentralized
physical
infrastructure
networks
such
as
filecoin
can
really
help
to
solve
this.
So
we're
hoping
that
leveraging
filecoin
with
a
network
like
lily
pad
for
coordination
could
really
allow
this
distributed
compute
model.
A
The
other
problem
is
efficiency
and
pricing,
so
oligopolies
don't
make
for
efficient
use
of
hardware
and
they
don't
provide
good
pricing
models
for
Fusion
for
users.
So
lilypad
can
enable
coordination
layers
for
matching
these
CPU
and
GPU
with
Supply.
A
A
Okay,
sorry,
the
last
fly
from
me
is
distribution.
I,
really
love.
This
quote
from
Brooklyn
zelenka
who's,
a
fish
and
founder
who
said,
if
you
look
at
large
Cloud
providers,
their
data
centers
are
really
in
North,
America
and
Europe
almost
entirely
and
we're
leaving
behind
the
rest
of
the
world
and
I.
Think
that's
something
that
filecoin
and
compute
networks
on
top
of
it
can
really
help
as
well.
A
I
also
think
there's
so
many
applications
and
honestly
I
listed
these
just
off
the
top
of
my
head,
but
there's
probably
so
many
more
that
you
guys
could
come
up
with
as
well
or
you
know
even
teach
about,
but
web3
and
AI
have
a
really
good
relationship
and
I.
Think
web3
can
solve
some
of
the
problems.
A
We're
going
to
see
in
AI
as
well
so
web3,
for
example,
is
incredibly
good
at
unlocking
Network
effects,
and
we
this
is
what
we
can
bring
to
both
web
3
and
web
2
developers
like
the
power
of
Cheaper,
more
open,
more
efficient
compute,
combined
with
resilient
storage,
transparency
and
data
ownership.
A
So
we've
also
kind
of
brainstormed
some
application
and
use
cases
and
we'd
love
to
hear
from
you
what
you
think
it
could
be
used
for
and
I'm
getting
time
there.
So
I
am
going
to
hurry
up.
So
these
are
some
of
the
web
3
applications.
We
think
this
could
be
useful
for
a
lot
of
which
I've
already
talked
about
so
I'll
just
move
on
through
those.
A
We
also
think
this
could
be
really
useful,
specifically
in
the
filecoin
compute
Network,
for
some
of
these
things,
like
imagine
having
darted
hours
with
the
centralized,
Ai
and
ml
I
mean
like
the
abilities,
there
are
kind
of
crazy
and
obviously
unlocking
unlocking
GPU
power
in
our
system
as
well
would
be
awesome
all
right.
A
This
is
just
an
example
actually
of
using
something
like
a
distributed
Network
for
attribution
so
showing
how
web3
can
solve
some
of
those
AI
problems,
and
this
is
the
future
of
lilypad.
So
one
of
the
things
that
we're
going
to
do
next
is
deploy
this
to
the
IPC
Network.
A
For
those
of
you
that
haven't
heard
about
it,
we
have
a
talk
from
Marco
coming
up,
I
believe
who
will
tell
you
all
about
how
scalable
that
network
is
going
to
be
on
that
shortly
and
we'll
also
be
building
out
more
robust
testnet
and
more
use
cases
for
you,
guys,
alrighty,
and
if
you
want
to
get
in
touch
with
us.
This
is
our
details,
feel
free
to
hit
us
up
in
the
slack
we're
pretty
responsive
there,
and
otherwise
you
can
go
and
have
a
look
at
more
resources.
We've
got
a
light
paper.