►
From YouTube: ONNX Edge WG meeting 20190807
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
Okay,
so
just
the
record,
this
is
a
meeting
on
August
7th,
currently
present
darlin
our
Milan
Oh
JA
kami
and
Saurabh
dan
green
done
Greek
from
me
in
tow
and
yeah.
So
we
are
ready
for
more
participants
to
join
just
chatting
waiting
for
them
to
join.
Go
ahead.
Sorry,
you
see,
you
want
to
ask
something
so.
B
B
B
B
So
that's
part
of
it,
but
what
is
the
end
goal
is
the
end
goal
that
we
can
define
profiles
that
are
expressed
on
the
model
files,
the
onyx
model
file
and
then
an
off,
and
then
when
we
can
say
that
this
model-
and
this
particular
would
have
this
precision
been
random,
dystrophin
profile
a
and
if
it
is
done
in
profile
B,
then
we
will
be
getting
this
material.
Is
that.
A
Through
that
use
case
or
models
that
supports,
we
would
identify
the
sub
set
of
operations
that
are
necessary
to
be
implemented
right
for
that
profile.
So
you
can
say
the
body
could
said:
I
comply
with
this
traffic
right
and
for
that
profile
includes
the
attributes.
End
models
as
a
kind
of
example,
models
that
somebody
would
need
to
satisfy
the
in
terms
of
the
running
in
terms
of
the
accuracy
latency
or
the
those
are,
the
attributes,
we'd
identified,
accuracy,
latency
size,
power,
consumption
and
data
locality.
So
now,
yes,
ultimately,
you
define
a
profile.
A
A
That
Onix
profile
then
define
subsets
of
operations
and
certain
KPIs.
That
is
what
we
discussed
so
far
or
decided
on
so
I
know
that
you
missed
couple
of
last
meetings,
but
is
recently
I
mean
everything
is
captured
in
the
recordings
and
the
meeting
minutes,
and
that
is,
as
you
know,
captured
github,
and
last
time
we
discussed
one
profile
kind
of
prototype.
Let
me
maybe
share
this.
A
A
A
What
we
discussed
basically
was
this
this
document
here
we
put
in
discussion
material
just
copied
over,
didn't
play
that
offered
Pro
created
this
document,
like
as
an
example
and
how
we
will
define
a
problem.
What
is
the
problem
right,
and
so
this
was
plus
one
example.
We
discussed
last
time,
and
you
know
we
would
discuss
it,
didn't
define
it
completely
to
this
discussion
material,
hopefully
also.
A
We
use
this
as
an
example
potentially
to
define
as
a
how
would
find
the
template,
basically
how
we
would
go
ahead
and
define
another
profile,
and
so
the
structure
of
this
document
was
also
discussed.
Like
you
sections,
and
what
this
power
required
networks
is
an
onyx
or
a
shamanic.
Separations
are
coming
out
of
these
networks
really
right,
and
so
that
was
that
was
no
deep
discussion.
We
had
last
time
now,
if.
A
Of
and
then
this
is
how
we
decide
that
make
sense.
For
example,
if
we
decide
and
then
they
basically
leaves
Leafs
of
this,
this
topology
becomes
something
it
you
need
to
define
right
and
then
through
the
attributes
identified-
and
this
is
very
left
action
acted
was
for
me
to
panel
proposed,
limiting
in
this
suspect,
and
so
I
did
have
something
put
together
for
today
to
channel
it's
really
only
a
tough
much
time
but
thing
is
I
was
sure
how
much
of
that
map
you
know
makes
sense.
If
it's
only
me
and
you
and.
A
A
A
You
know
it
seems
like
what
do
we
say
that
in
the
workshop
right,
because
we
have
to
provide
update,
we
will
prepare
something
as
a
kind
of
presentation
for
the
workshop.
There
is
a
guidelines
that
was
sent
by
the
steering
committee.
I,
like
you,
have
like
ten
minutes,
I
think.
Similarly,
what
was
the
last
time
you
were
there
last
time
right
did
you
were
you
there
in
the.
C
A
B
A
A
A
A
B
A
B
A
Know
what
I'm
thinking
that
okay
sounds
good,
that's
official
statement
and
comment.
I
I
think
that
this
is
one
thing.
So
basically
what
I
had
for
today,
I,
don't
think
I
would
like
I.
Don't
know
I
feels
to
me
like
we
need
a
little
bit
more
audience
to
discuss.
I
feel
like
if
I-
and
you
just
continue
discussing
this-
is.
C
A
A
B
A
It
is
a
pushed
into
the
interview,
discussion,
materials.
This
was
what
we
discussed
last
time,
so
here
this
document
this
was
just
perfect,
just
like
considered
Oh
kind
of
also
work
as
defining
a
template
as
well
find
a
profile,
and
so
this
was
discussed
last
time.
It
wasn't
it's
not
done
by
no
means
it's
just
some
sort
of
proposal
right
of
how
to
think
you
know
the
form
and
structure
and
and
also
what
do
we
want
to
cover
what
kind
of,
and
we
debated
whether
does
we
need.
Does
this
make
sense?
B
A
That's
a
good
question.
I
think
right
now
think
about
that.
Somebody
will
just
say:
oh
I
have
my
own
profile
that
I'm,
defining
and
I'm
gonna
publish
it.
I
didn't
consider
that
I
think
the
goal
of
this
exercise
is
that
mix
as
a
body
finds
defines
these
profiles
and
people
that
are
complying
with
onyx
become
compliant
with
these
profiles.
If
they
would
prefer
to
define
additional
profile
and
they
would
reach
out
to
the.
A
Onyx
either
steering
committee
or
I
think
infrastructure,
because
infrastructure
is
good
infrastructure.
Sig
is
the
one
that
would
probably
continue
defining
these
profiles
and
defining
right
because
working
group,
our
our
Charter,
is
done
once
League
complete
with
the
statement
of
work,
so
we
would
define
initial
set
of
profiles
and
say
this
is
what
we
have
and
recommend
is
to
six
and
they
accept
it,
and
then
this
for
their
life
of
this.
These
profiles
will
keep
going
and
and
being
being
updated
through
onyx
thick
body.
B
So
my
second
question
is
would
know
I'm
going
a
little
bit
in
the
implementation
space.
Then
we've
described
the
profile.
Is
that
something
like
the
attributes
of
different
profiles?
Would
that
be
part
of
the
same
onyx
file,
or
would
it
be
something
that
the
spec
has
defined
and
the
runtimes
know
about,
or
is
the
model
file
going
to
be
self-sufficient,
which.
A
No
model
will
not
be
self-sufficient.
I
mean
part
of
your
own.
Exact
profile
is
actually
identifying
a
set
of
models
that
you
need
to
run
and
be
able
to
support
right
as
a
provider
with
certain
KPIs,
so
the
profile,
the
the
model,
the
model
file,
our
files
is
a
part
of
the
profile
definition.
We
would
specify
the
profiles,
sorry,
the
models
that
one
needs
needs
to
support
like
Esper,
this
kind
of
example.
A
Here
proton,
you
will
say,
require
networks
right
and
so
now
we
can
identify
networks
here
and
now
in
terms
of
the
accuracy-
and
you
know
not
accuracy,
but
rather
you
know,
we
would
say
okay,
this
out
this
program
is
the
support
inception,
v3
Mohammad
v2,
whatever
we
we
agree,
what
models
B
should
be
supported
for
his
profile
and
then
and
accordingly,
set
of
operations
operations.
Now
we
didn't
go
to
discussion.
A
A
B
B
I
guess
what
I
would
I'm
just
thinking
that,
in
terms
of
sometimes
implementing
the
standard,
what
complexity
would
they
have
now
in
order
to
support
definitions,
if
it
was
all
described
in
the
onyx
model
file
itself,
hey
I'm,
this
model
complying
to
you
know
like
this.
By
the
way
same
way,
we
define
the
offset
we
say,
I
other
than
offset
I
define
this
file
this
device
profile,
and
these
are
the
attributes
of
the
device,
and
then
the
runtime
can
just
read
on
that
and
say.
A
Formalize
profile
through
some
sort
of
a
config
file,
or
something
that
somebody
could
say
if
I,
if
I
you
know,
is
a
part
of
a
compliance
right
this
this
this
file
describes.
All
you
need
describes
the
the
the
models
you
need
to
use,
drugged
the
reference
runtime
that
you
need
to
use
right,
or
you
know
not
actually
using
your
own
time.
Whoever
is
supporting
this
right,
but
the
thing
is
like
you
know
you
could
come
up
with.
A
A
You
know
you
can
get
all
the
information
you
need
to
validate
that
the
I
am
you
know
compliant
with
this
profile,
this
these
details
we
didn't
get
into,
and
this
is
something
that
we
need
to
discuss.
I
think
with
things
like
I
think
infrastructure,
sigur
whatnot,
but
for
now
I,
don't
I'm
not
too
stuck
on
it.
This
is
more
like
implementation,
I
guess
detail,
and
maybe
this
is
where
you
are
also
going
in
terms
of
how
this
will
be
kind
of
presented
is.
A
B
A
B
A
A
A
A
Maybe
we
can
start
preliminary
kind
of
sync
up
with
the
seek
I.
Don't
know
if
you
are
completely
ready,
but
definitely
yes,
I
think
some
some
preliminary
sync
up
or
start
discussion
with
a
sick
head
for
infrastructure,
I
think
we'll
be
okay
to
kind
of
flush
out
this
idea
and
so
or
how
do
we
proceed
with
implementation
defined?
Yes,
this
is
part
of
our
actually
statement
statement
of
work
eventually,
the
next
is
that
we
publish
these
kind
of
processes
or
recommend
actually
not
establish,
recommend
processes
in
collaboration
with
with
the
appropriate
see.