►
Description
wasmCloud Weekly Community Call - this week featuring a demonstration of wasi-nn, .15 roadmap discussion, building for arm64 and more!
A
Okay,
all
right
so
welcome
everybody
back
for
the
january
sixth
meeting
for
wasm
cloud
happy
new
year,
where
we're
trying
to
enable
developers
to
build
their
functions
and
services
and
webassembly
and
run
them
everywhere,
and
I
think
we've
got
a
couple
new
people
joining
us
today.
Milan
and
felix.
You
guys
want
to
take
a
second
and
maybe
introduce
yourselves.
B
Sure
you
can
go
first,
felix
thanks!
Very
much!
Hey
everybody
felix
here
from
shopify,
I'm
part
of
shopify's
automation,
group,
there's
another
one
of
my
colleagues
he's
on
the
call
as
well
jeff
charles.
I
recently
joined
this
group
and
we
are
using
a
web
assembly
to
run
low
latency
scripts
and
I'm
looking
forward
to
meeting
everybody
and
working
with
this
group.
Those.
A
D
A
A
That's
great
and
in
our
tradition
of
our
tradition,
we're
trying
to
run
is
to
kick
off
each
meeting
with
a
demo
and
I'm
really
happy
to
turn
the
meeting
over
to
andrew
and
min
queue
from
intel
who
are
going
to
do
a
demo
of
wazi
nn
andrew.
Are
you
ready
to
share
your
screen?
I
can
give
you
permissions.
F
A
A
A
If
anybody
wants
to
look
at
it
on
topics
before
demo,
it's
down
at
the
very
bottom
here,
while
you
guys
are,
let
me
know
when
I
can
stop
filling
airtime
here,
a
whittle
with
phil,
caddy
and
wapc
with
phil.
I
think
he's
gonna
try
to
do
one
or
both
of
those
next
week,
including
the
some
of
the
new
wapc
code
generation
stuff.
We
have
code
generation
itself,
we
will
have
the
wasmcloud.15
release,
which
is
hopefully
sometime
here
in
january.
A
We
have
the
website
work
that
jordan
and
I
are
working
on
and
we
have
tutorials
crosslit
wasn't
three
on
small
devices
and
possibly
a
talk
through
on
the
yeoman
wpc
generator
stuff
that
the
microsoft
folks
are
working
on.
If
there's
any
other
demos
just
feel
free
to
add
them
right
to
that
agenda
in
chat,
and
we
can
go
from
there.
G
G
So
I
think
we
don't
need
to
introduce
the
water
here,
because
everybody
knows
that
right.
So
you
know
this
was
a
presentation
that
we
prepared
for
the
web
machine
learning
workshop.
You
know
end
of
last
year
ish.
So
so
we
are
part
of
a
buy
code
alliance
which
is
a
you
know:
open
source
community,
that's
dedicated
to
creating
a
you
know,
secure
new
software
foundation.
I
think
some
of
the
members
of
this
group
also
involved
in
the
alliance,
so
we
are
part
of
our
founding
members.
G
G
G
G
G
So
web
assembly
provides
this
ideal
portable
format
of
deployment
and-
and
in
addition
to
that,
you
know,
there's
also
usually
performance
requirement
for
web
assembly,
and
it
could
be.
You
know
some
special
feature
of
a
cpu
like
a
avx
512
and,
potentially
you
know,
gpu
and
special
accelerator,
like
a
tpu
support
on
a
variety
of
devices.
So
that's
the
reason
that
you
know.
G
G
You
know
the
reason
for
that
is
that,
because
machine
learning
is
still
evolving
rapidly,
with
new
operations
and
network
topologies
and
under
some
number,
like
you
know,
machine
learning,
operation,
increased
20
percent
every
year
and
and
there's
no
end
of
a
no
no
no
side
of
ending
the
happening
anytime.
Soon.
Next
slide.
G
So
it's
a
simpler
api
and
it
actually
allows
for
better
eyepiece
protection
because
it
doesn't
really,
you
know
the
models
that
they
involve.
You
know
just
load
the
model
and
you
don't
need
to
have
the
internal.
You
know
knowledge
of
the
of
the
model,
construction
and
the
the
weight
of
the
model,
for
example,
exposed
to
the
api
level
so
we're
closely
working
together
with
another
effort.
F
Let
me
let
me
share
this.
This
slide
real,
quick,
so
at
a
high
level,
if,
if,
if
you're
into
block
diagrams,
what
what
this
explains
is
that
if
you
want
to
run
machine
learning
models
using
you
know,
this
was
end
thing.
Here's
what
it
would
look
like.
You'd
have
user
application
code,
which
would
say
you
know
hey.
I
want
to
execute
this
model.
F
F
F
Currently
perhaps
at
some
point
in
the
future,
it'll
be
a
runtime
flag
and
it
will
be
compiled
in
by
default.
But
this
is
definitely
an
optional
feature.
I
mean
not
not
everyone.
Who's
using
watson
time
is
going
to
want
to
use
machine
learning
capabilities,
but
so
that's
that's.
At
least
we
get
that
here's,
the
code
that
actually
you
know,
runs
the
model.
F
And
so
here
you
see
me,
you
know,
reading
the
model
files
in
I
load
in
those
model
files
using
this
ynn
call
that
russ
knows
how
to
you
know,
translate
into
a
wazzy
call
down
here.
We
so
here's
we
have
the
tensor
we're
setting
the
the
input
here
we're
doing
like
the
computation,
so
this
will
be
the
inferencing
and
then
we'll
retrieve
the
output
and
and
we'll
sort
the
results
so
that
we
can.
You
know
print
something
that
makes
some
sense
to
the
user.
F
So,
let's
see
here
so
that
script
does
a
bunch
into
downloading.
So
I
pre-downloaded
all
that
stuff
here
what's
happening.
When
I
run
this,
is
it
it
compiled
my
example
main.rs
to
a
wasm
file
with
those
wazzy
calls
in
it,
and
now
it's
running
it
and
you
can
see
that
actually
loading
the
model
is
sort
of
slow.
That's
sort
of
an
artifact
of
currently
wowzi
doesn't
like
memory
map
mapping
files
when
it
reads
them,
it
like
reads
them
byte
by
byte.
So
it's
doing
the
whole
byte
buffer
transfer.
F
F
It
executes
the
inference
and
then
it
sorts
the
results
in
and
you
get
these
class
ids
with
their
probability.
So
it's
saying
that
there's
a
53
probability
that
that
image
was
class.
Id
963
turns
out
that
class
id
963
is
a
pizza,
but
you'll
say
wait.
We
didn't
even
see
the
image
yet
so
if
to
look
at
the
those
test,
fixtures
that
I've
created
there's
a
the
whole.
The
open,
vino,
rrs
repo,
contains
those
texts.
F
Yeah
test
fixtures,
and
so
this
this
image
was
what
was
converted
into
the
tensor,
and
here
here
you
see
the
the
alexnet.
You
know
weights
and
stuff
like
that.
That
need
to
be
converted
for
openvino
to
consume
that
so
that
that
stuff's,
all
there
there's
more
links
to
all
this
stuff
in
in
the
blog
post
that
I
posted.
F
F
I
haven't
done
any
work
on
this
since
then,
because
I've
been
on
vacation,
but
if
you're
interested
in
sort
of
more
details
about
what
I'm
talking
about
right
now,
here's
here's
one
of
the
links,
so
it
would
be
using
wisely
and
then
and
was
in
time
and
then
here's
implementing
westing
and
in
wasm
time
the
first
is
as
a
user.
You
know
how
would
I
use
wasn't
wasn't
in
currently,
and
the
second
is
okay.
If
I
want
to
write
my
own
wazi
proposal,
what
would
I
need
to
do
to
implement
that?
F
F
A
I've
got
a
question:
what
is
the,
what
is
the
thought
or
is
there
been
any
discussions
with
any
other
engines
on
incorporating
wazing,
and
you
know
more
broadly
or
you
know,
building
it
out
or
is
the
idea
that
contributing
it
to
wasm
time?
Other
implementations
should
look
here
and
then
try
to
pull
the
work
in.
F
F
Yet
right,
the
the
ml
stuff,
you
know,
is
sort
of
down
the
road,
so
we
have
a
chance
to
evolve
the
api.
Without
you
know,
someone
breathing
down
our
neck,
saying
hey
when's
this
going
to
come
out
in
that
sense,
wazzy
itself
is
very
early,
but
wazzy
and
nnn
itself
is
very
early
right.
The
api
we
need
to
evolve
it
and-
and
you
know,
we've
talked
about
switching
over
to
a
model
builder
api.
Where
you
could.
You
know
control
the
the
nodes
in
the
graph
directly
from
within
wazi.
F
That
would
that
would
maybe
look
a
little
bit
more
like
what
you
know
tensorflow,
and
these
guys
can
do.
We've
talked
about
other
implement
ever
other
backing
implementations.
Besides
openvino
right,
you
could
have
tensorflow
or
or
something
like
that,
you
know
implementing
yzn
and
we.
What
we
really
want
to
do
is
we
want
to
see
people
how
people
would
use
this
or
what
changes
they
would
need
in
order
to
use
this
in
in
production
right.
This
is
currently
at
the
proof
of
concept
stage.
You
know
we
can
do
this.
A
Got
it
so,
my
I
I
I'll
close
with
that
question
around
like
what
is
it
that
you
need?
You
know
like
the
call
to
action
for
us
here
at
the
end,
but
in
the
interim
I
think
you
touched
on
another
great
question:
how
much
collaboration
or
cooperation
was
there
with
other
engine
providers
like
tensorflow
to
ensure
that
the
api
really
kind
of
is
a
truly?
You
know
good
middleware
between
you
and
lots
of
was
there?
Was
there
proactive
engagement?
A
G
So,
okay,
so
thank
you.
Let
me
take
on
this
one.
So
so
we
are
really
working
very
closely
with
you
know
with
the
webinar
group,
which
is
a
really
a
consortium
for
like
a
in
the
collection
of
all
those
top
machine
learning.
Practitioners
from
you
know,
for
example,
microsoft
from
tensorflow,
google
group
and
and
intel
and
other
another,
a
bunch
of
companies
that
were
involved,
so
we
worked
really
close
with
them
and
there
is
a
similar.
G
G
You
know,
web
assembly
is
supposed
to
be
near
native
performance,
but
in
terms
of
a
high
performance
you
know
high
performance
computing
right,
such
as
the
machine
learning.
You
know
the
gap
is
still
pretty
big.
We
are,
you
know
talking
about
implementing
sim
d
428
bit
right,
but
you
know
avx.
512
is
a
common
practice.
A
F
I
would
say
issues
in
wasm
time:
I
will
respond
to
those
pretty
quickly
any
issues
to
the
wasan
and
repo
as
well.
You
know
I'll
I'll,
see
and
we'll
respond
pretty
quickly
if
you
want
to
like
discuss
something,
maybe
not
necessarily
an
issue,
I'm
on
zulip
pretty
much
all
the
time
the
bike
coated
alliance
zoo
up.
So
if
you
find
me
on
there
I'll
send
you
the
links
later
liam
and
you
guys
can.
A
Yeah
awesome
I'll
make
sure
that
those
are
linked
in
our
notes
here,
and
I
have
your
articles
linked
in
great
job.
Thank
you
so
much
for
the
for
the
great
demo
today.
I
really
appreciate
it
and
let's
slide
on
down
the
agenda
here,
real
quick
to
next
topic,
which
is
just
a
community
news
tomorrow.
Speaking
of
awesome
time
is
the
wason
time
community
meeting
thursday
at
one
o'clock
meeting
agenda
is
posted.
I,
and
if
you
I
need
a
an
invite,
you
can
ask
for
one
on
zoola
board.
A
A
A
There
are
some
early
early
shots
here
and
I'll
link
in
if
people
want
to
start
giving
us
feedback
on
the
on
the
design-
and
I
went
ahead
and
posted
up
in
the
community
notes
here
kind
of
the
discussion
that
we
had.
If
you
just
want
to
look
for
reference
in
community
meetings
here
under
the
23rd,
actually,
here's
the
link
to
the
netlify
dev
branch.
A
If
you
want
to
give
live
feedback
and
there's
the
link
to
the
decomposed
issues
here,
sort
of
like
you
know
where
we're
going
with
this
ending
with
like
a
nice,
interactive
online
experience,
new
docs
and
so
forth,
and
then
the
sort
of
notes
and
links
to
all
the
artifacts
in
here,
like
the
new
website
map,
the
new
website
proposed
view
the
templates
we're
using
some
of
our
thoughts
around
color
theory
and
logos.
And
you
know
all
the
sort
of
details
are
all
linked
in
there.
A
Awesome
any
of
the
microsoft
folks,
I
don't
know,
speak
for
ralph
if
there's
any
progress
on
the
months
of
interest
here,
which
is
sort
of
a
standing
topic
of
of
links
to
topics
and
ideas
that
are
around
webassembly
I'll
hit
ralph
up
next
time
early,
because
I
think
he's
got
a
conflict
at
1,
30.
and
then
on
to
dot
15
stuff
kevin.
Did
you
want
to
maybe
pick
up
on
a
discussion
around
the
sort
of
stuff
that
we've
still
got
open
here?
I
I
A
E
Yes,
so,
as
far
as
I
noticed,
homebrew
is
there
so
homebrew
should
be
set
that
one's
kind
of
a
bit
more
of
a
manual
process.
So
essentially,
when
we
do
a
github
release,
which
is
automated
in
wasmcloud
and
wash
we'll,
essentially
update
the
wasmcloud
homebrew
repository
to
point
to
the
checksum
and
release
file
for
that.
As
far
as
like
the
army
type
packages,
those
should
be
good.
Currently,
it's
kind
of
a
hacky
way
to
build
it.
E
It's
like
docker
and
docker
china
build
arm
packages,
we're
looking
into
purchasing
like
an
arm
instance,
possibly
in
aws
right
now,
just
to
speed
those
builds
up
and
make
it
a
little
less
finicky.
So
those
should
be
looking
good
and
currently
working
for
the
test.
Automation
for
wasm
cloud,
currently
working
through
a
few
little
issues
with
that.
So
hopefully
I'll
get
a
pull
request
in
today
or
tomorrow
to
get
that
moving
forward.
A
Does
anyone
have
any
suggestions
or
feedback
anyone
else?
That's
building
arm
packages
on
how
you're
doing
this?
I
think
the
wasm
time
team
ultimately
ended
up
just
getting
an
aws
box
themselves
that
actually
came
up
in
their
last
meeting.
Anyone
else
have
any
other
suggestions,
for
you
know
best
ways
on
you
know,
building
that
scalably.
H
Provider
for
questlet
and
what
we
had
to
do
effectively
was
on
the
linux
host
for
ci
instance.
We
had
to
have
the
entire
like
c
compiler
and
everything
else
cross
compiled
for
arm,
and
so
those
binaries
are
available,
like
you
can
get
the
c,
compiler
and
whatnot
and
c
lang
and
a
couple
of
other
things
basically
to
be
able
to
compile
into
llvm
on
on
a
forearm,
but
on
a
amd64
architecture,
and
it
works
it.
H
But
then,
if
you
have
to
deal
with
like
open,
ssl
or
anything
else
like
that,
like
you'll
have
to
basically
compile
that
from
scratch
to
make
it
work
with
arm
or
like
cross,
compile
it
for
arm.
So
it's
like,
it
really
depends
on
how
many
dependencies
you're
taking
on
at
that
time.
It's
really
not
scalable.
H
E
A
I
love
the
link
here
to
the
artisanal
way
of
doing
it,
as
well
as
the
feedback
on
there
be
dragons
here,
so
we
can
definitely
get
a
box
and
either
spin
it
up.
You
know
on
the
fly,
but
mvp
is
probably
just
turn
on
the
box
and
you
know
ssh.
You
know
like
configure
it
manually.
So
thank
you
for
that
awesome
guidance
on
that.
That's
huge,
I'm
gonna
link
that
into
the
notes.
A
E
Yeah
yeah,
so
basically
there's
a
homebrew
dash
or
wasmcloud
homebrew
repository
out
there
right
now
currently
only
has
wash
listed
in
there.
That's
been
tested,
jordan
rash,
helped
me
out
with
that
we're
just
I'm
just
basically
waiting
on
a
github
release
of
wasmcloud.
So
once
that
is
out
there,
we
can.
You
know
basically
point
the
homebrew
repository
for
wasm
cloud.
E
A
All
right
super-
well,
maybe
a
mini
demo,
I'll
put
us
down
for
is
once
we
hit
the
next
milestone.
You
can
do
a
like
a
quick
little
video
it'd
be
useful
for
website.
Anyway.
Absolutely
that's
awesome.
Thank
you
so
much
for
all
that,
that's
great,
so
we're
really
close,
as
kevin
mentioned,
on
the
new
wasm
cloud
experience,
which
is
still
linked
here.
A
If
anybody
wants
to
give
feedback,
and
last
week
we
had
a
little
discussion
about
bindle
and
I
think
there's
going
to
be
some
more
discussion
on
that
and
we'll
cue
that
up
for
a
demo
and
taylor,
do
you
want
to
do
a
quick
just
verbal
on
what
bindle
is?
I
know
it's
linked
out
right
right
here,
yeah
I'll
go
over
that
really
quick,
but
also
before
I
forget.
A
As
soon
as
we
cut
this
next
release
of
crescent
we're
gonna
be
moving
over
the
provider
formerly
known
as
wasc
to
the
wasmcloud
org
we
just
it
took
longer
than
we
intended,
but
we
had
a
bunch
of
refactors.
We
were
doing,
and
now
all
those
major
refactors
are
done
so
as
soon
as
we
cut
that
new
release,
hopefully
by
the
end
of
this
month,
we'll
be
moving
that
over
so
that
it
can
be
more
based
on
like
hey.
A
This
is
this
is
a
wasn't
cloud
thing
and
kind
of
centralized
like
we've
talked
about
before
so
that'll
be
going
over
there.
So
sorry
that
side
note
done
now.
I
can
move
on
to
bindle
for
those
who
weren't
here
before
the
holiday
break.
Bindle
is
a
new
aggregate
object,
storage
idea,
experiments
thing
we
have
the
idea
is
that
especially
in
webassembly,
and
we
think
there's
many
other
users
cases
if
you
have
any
use
cases
and
you
want
to
try
it
out-
please
do
but
the
main
thing
is
around
webassembly.
A
It's
immutable
and
then
it
has
a
list
of
the
different
things
that
are
called
partials
in
it
that
are
the
different
artifacts
that
are
just
it's
binary,
blobs
of
some
kind
in
our
case
here
talking
about
wasn't
modulus
they're,
probably
mostly
wasm
modules,
and
so
anyway.
This
is
something
we
talked
about
last
time
we
will
be
cutting
a
new
release,
probably
in
the
next
few
weeks,
and
we're
redoing
one
of
the
api
endpoints
to
make
it
a
little
bit
cleaner
for
everything
to
be
done
with
like
less
objects
needed
and
whatnot.
I
Any
other
things
yeah
one
of
the
things
that
was
on
our
list
to
do
to
explore
as
soon
as
we
got
back
from
vacation
and
once
we
hit
like
the
the
feature
complete
on
the
0.15
stuff,
was
to
take
a
look
at
replacing
our
par
files
or
provider
archives
with
bindles
and
see
what
that
would
entail
and
and
what
the
pros
and
cons
are.
But
we
haven't
started
that
effort.
Yet.
A
A
Now
that
we've
had
a
little
bit
more
experience
of
like
trying
this
out
and
as
we're
trying
to
apply
more
stuff
with
the
security
model
around
signing
and
things
so
yeah,
but
that's
all
there
we're
moving
forward
with
that.
So
if
there's
any
feedback
or
people
give
it
a
try
or
have
any
ideas,
please
let
us
know
on
the
repo
we're
in
this
meeting.
A
B
A
We
already
have
the
stuff
if
you
want
to
see
like
some
demos
and
examples
and
things,
but
it's
really
not
like
super
impressive
to
like
to
like
watch
in
a
demo
form
it's
more
impressive
and
it's
like
use
and
how
you
can
handle
stuff
and
filter
things
and
whatnot,
we'll
block
we'll
block
out
time
and
do
it
formally,
because
I
think
the
you
and
kevin
getting
through,
and
maybe
we
look
at.
If
wasn't
cloud
moves
to
it.
We
look
at
it
as
an
example
or
something
like
that.
A
I
think
there's
definitely
a
huge
value
in
it,
because
it's,
I
think,
one
of
the
things
that
we
all
have
in
common
here.
With
the
you
know,
we
have
a
lot
of
folks
that
are
around
the
ecosystems.
We
all
see
the
same
problems
and
I
think
us
just
agreeing
on
hey
we're
all
going
to
work
on
this
thing
and
not
three
different
things
is,
is
a
big
piece
of
the
battle,
the
big
piece
of
the
challenge
here.
A
So
if
it's
bundles
versus
you
know
par
files
like
let's
just
pick
something
and
and
then
agree
and
then
make
progress
together
is
huge.
So
on
our
agenda,
I've
got
blocked
out
and
kevin.
I
I
think
I
would
ask
you,
I
think
you
probably
gave
us
enough
overhead
here
or
enough
of
a
story
just
a
quick
glance
at
open
pull
requests
and
issues
for
the
project.
A
I
No,
I
think,
like
I
said,
the
0.15
is,
is
kind
of
our
priority,
and
so
getting
that
finished
is
is
the
first
thing
up
there
and
then,
after
that,
we're
going
to
start
looking
at
things
like
the
the
other
host
runtimes
like
swift
and
the
browser
and
and
some
of
those
but
yeah
we
haven't
really
done.
We
haven't
done
the
backlog
rooming
to
sort
of
put
those
in
in
order
of
completion,
so
we
haven't
figured
out
what
the
0.16
milestone
looks
like
yet.
A
Okay,
all
right,
but
that's
awesome,
kevin.
Thank
you
for
that
and
but
for
anyone,
that's
you
know.
Looking
ahead,
you
know
thinking
through
that
a
little
bit.
You
know
what
we'll
be
doing
later
this
year.
I
think,
would
be
great
as
we
as
we
go
through
there
and
I
think
we're
pretty
close
to
the
to
the
end
and
no
adls
or
anything
like
that
to
anybody
to
discuss
architectural
decision
logs.