►
From YouTube: wasmCloud Working Group - Machine Learning 03/23/22
Description
wasmCloud is a platform for writing portable business logic that can run anywhere from the edge to the cloud, that boasts a secure-by-default, boilerplate-free developer experience with rapid feedback loop.
https://wasmcloud.com
A
Hello,
this
is
christopher
prowing
from
munich,
germany
and
today
I'll
show
you
around
our
machine
learning,
inference
capability
provider
and
the
application
we
build
around
it.
So
in
that
architectural
drawing,
you
probably
see
here
on
the
screen,
you'll
see
a
picture
of
the
application
there's
in
the
center
in
the
big
rectangle.
That
illustrates
our
wasn't
cloud
runtime.
A
There
are
multiple
artifacts,
the
smaller
rectangles
that
are
capability
providers.
We
have
got
an
http
server
and
that
is
also
an
endpoint
to
the
application
on
the
lower
left
and
on
the
top
right.
We
have
the
capability
provider
which
does
the
inference
processing,
so
the
machine
learning
processing
and
it's
marked
ie
that's
for
inference
engine.
However,
that's
misleading
a
bit
because
we
already
have
more
than
one
inference
engines
implemented
in
it.
However,
now
you
know
what
it
represents,
then,
in
the
vasencloud
runtime
in
form
of
circles,
youth
actors-
we
have
two
of
them
here.
A
The
one
is,
I
api,
so
inference
api,
that's
directly
connected
to
the
http
server
and
the
other
one
marked.
I
you
can
imagine
it
not
being
there
it's
not
implemented.
We
intended
that
for
first
iteration
of
the
application.
That's
not
important
outside
of
that
wasn't
cloud.
Runtime
you've
got
the
bundle
server.
The
bundle
server
is
a
kind
of
a
blob
store
and
that's
where
we
used
to
store
the
ai
models
as
well
as
some
metadata.
A
So
what
happens
is
if
you
process
requests
against
that
wasn't
cloud
runtime
they
come
into.
The
http
server
are
routed
via
the
inference
api
to
the
capability
provider,
who
does
the
processing
and
the
response
going
back
via
the
inference
api
and
the
http
server
so
that
you
will
receive
hopefully
a
200,
okay
response
with
the
result
of
the
inference
processing.
A
However,
as
I
said
before,
the
processing
can
start
the
capability
provider.
Ie
has
to
download
all
the
models,
and
that
is
done
during
startup
of
the
application.
A
A
Don't
worry
about
the
repository
url,
you
can
ask
for
it
in
the
select
channels,
and
I
think
you
will
get
a
quick
response.
So
one
of
the
preconditions
is
that
you
have
installed
bundle
server.
As
far
as
I
know,
you
can
install
any
version
node.
However,
that
versions
later
than
0.7.1
you
have
to
deal
with
keys,
so
they
harden
the
security
aspects.
A
So
I
for
development
purposes,
I
use
0.7.1
and
you
do
not
have
to
worry
so
much
for
security.
What
else
do
you
have
to
install?
Is
that
docker
and
docker
compose?
A
I
refer
to
the
page
of
wasm
cloud.
They
have
got
a
good
guide,
what
you
have
to
take
care
of
and
what
you
have
to
install.
Otherwise
we
may
directly
jump
in
and
start
the
application
now.
Let
me
see
if
I
have
it
ready
for
us
so
before
you
can
replay
the
application.
As
I
said,
you
have
to
load
the
bundle
server
with
the
models.
So
luck
for
you
for
us,
the
models
are
pre-configured
and
the
repository.
A
So
let's
see
how
you
load
them
up
into
the
bundle
server.
Once
you
have
cloned
the
repository
go
into
the
deploy
directory,
and
then
we
have
a
start.
Script
called
run.
This
is
your
friend.
If
you
just
type
it,
you
get
all
the
sub
commands
and
in
order
to
upload.
A
A
You
can
now
you
have
to
do
it
only
once
so,
once
you've
uploaded
the
models
and
the
meter
data
to
your
bundle
server.
You
do
not
have
to
do
that
again.
So
whenever
you
start
or
stop
the
bundle
server
next,
you
can
do
it
with
the
other
sub
commands.
You
see,
bundle
start
and
bundle
stop.
So
that's
started
now,
it's
time
to
start
the
application,
and
that
is
done
via
run
all.
A
A
You
can
follow
up
the
logs,
so
while
the
application
is
starting,
we
may
have
a
look
at
the
washboard,
which
is
getting
more
and
more
complete.
What
do
we
see
at
the
washboard?
We
see
the
application
starting
up
with
all
its
technical
stakeholders,
so
at
the
top
right
we
already
have
the
ml
inference
capability
providers,
you
see
its
status
is
healthy,
so
it's
up
and
running
on
the
lower
left.
You
see
two
links,
so
one
link
is
between
the
http
server
and
the
inference
api
actor.
A
A
Is
a
request
against
the
onyx
inference
engine
its
model
is
called
identity.
What
it
does
is
so
the
model
is
configured
such
that
the
output
of
the
model
is
always
the
input,
so
it
yields
what
it
gets.
It's
not
particularly
interesting
for
real
world
use
case.
However,
in
case
that
works,
you
know
that
the
application
is
up
and
running
and
works
well,
and
if
you
trigger
that,
you
see
we
get
a
200
back
and
also
we
get
a
result.
And
if
you
compare
that
result,
you
see
a
field
has
arrow
is
false.
That's
good!
Already!
A
If
you
compare
the
content,
you
see
that
the
content
is
identical.
In
fact
to
the
request.
That's
fine!
So
that's
a
request
against
the
onyx
engine.
Now
we
also
support
tensorflow
engine,
so
we've
got
a
second
model.
It's
also
pretty
say
simple.
However,
not
another
identity
model.
It
is
a
plus
three.
A
A
So
this
is
exactly
what
we
got
back
and
if
you
deserialize
that
into
f32-
and
you
see
that
it
incremented
now
it's
down
here,
so
it
incremented
every
input
by
three.
That's
why
it's
called
plus
three
yeah,
that's
it!
That
was
the
two
examples.