►
From YouTube: Weekly Sync 2020-07-31
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.q3lrykz1ifcw
B
Okay,
so
I
looked
up,
I
started
to
look
at
the
distributed
orchestrator
and
I
think
that
it
would
be
well.
I
think
that
it
would
be
good
to
go
through
it
together.
I
saw
you.
B
That
this
isn't
what
you're
intending
to
have
it
finished
bias
right.
So
I'm
trying
to
figure
out
okay
where's
it
at
where's
the
gaps
and
then,
where
are
you
trying
to
go
versus?
What's
actually
going
on
right?
So
I
think
in
the
future,
in
general,
more
more
comments
would
be
good.
Okay,
I'll
do
that.
D
Okay,
yeah,
maybe
like
till
the
code,
is
merged
I'll
also
come
in
like
a
water
display,
yeah
yeah.
B
Exactly
yeah,
so
if
we
could
maybe
have
like
if
we
could
have
like
a
document
going
as
well,
you
know
like
in
the
documentation
with
what's
going
on.
That
would
be
really
good.
So.
C
B
Look
at
the
code,
we
can
say:
okay,
what's
actually
happening
right
because
I'm
yeah,
sometimes
I'm
thinking,
I'm
not
sure
whether
I'm
going
I'm
I'm
looking
at
this
and
I'm
going
okay.
Well,
this
could
be.
You
know
this
is
probably
intentional,
but
but
you
know
why
is
it
intentional
and
then
is
it
or
is
it
intentional
right?
So
mainly
things
like
I
thought
we
had
talked
about
or
we
had
talked
about
like
the
operations
that
the
nodes
are
instantiating
and
stuff
and
it
looks
like
you're
giving
where
the
hell
did
it
go.
B
It
looks
like
down
here
you
give
you
give
you
know
when
you
instantiate
the
worker,
you
give
the
operations
right.
You
can
also
provide
operation
implementations,
but
it
doesn't
look
like
this
is
some
sort
of
it
looks
like
this
is
exactly
what
operations
they'll
be
running
right.
B
Yeah
so,
but
but
it
looks
like
this
is
the
one
that
it
will
be
running
right
like
not
necessarily
not
all
of
them,
but
they're
subset
of
them.
Okay,
that
is
okay,
so
I'm
just
I.
The
other
thing
was
the
reason
why
I
say
this
is
because
I
didn't
see
where
what
what
I
was
thinking
we
were
going
to
do.
There
was
make
that
the
like
the
allow
list
of
of
a
loud
demon
of
allowed
operations
right.
D
Those
are
the
list
of
allowed
operations:
okay,
oh
okay,
okay,
okay,
yeah!
This
is
where
they
see
the
comments
without
it
yeah
yeah.
So
those
are
the
list
of
allowed
operations
and
whenever
a
data
flow
comes
in
the
prima,
the
primary
node,
the
data
flow
node
sends
in
like
whatever
operations
it
needs
and
whatever
nodes
has
those
set
of
operation
response
back.
Okay,
the
data
flow
node,
which
tells
okay,
which
allocates
the
indexes
which
we
talked
about.
Okay,.
B
B
B
D
So
the
primary
node
has
a
list
of
operations,
operation,
names
and
indexes
it
is
allocated
to
so
from
the
list
of
indexes.
You
know
how
many
workers
there
are:
okay,
oh
from
a
list
of
indexes.
Okay,
yeah,
like
how
many
active
occurs
there
are
there
like
there
might
be
other
nodes
which
are
ready
to
accept
rules,
but
these
are
the
ones
which
are
associated.
A
B
All
right
yeah,
I
think
I
mean
it,
looks
so
and
then
okay
and
then,
where
was
that
so
we've.
B
B
Okay,
so
the
point
we're
at
now
is:
we've
got
the
worker
and
the
orchestrator
and
the
orchestrator
will
create
the
orchestrator
creates
one
of
those
round
robin
cues
or
the.
What
do
you
call
it?
I
think
you
had
a
bra
yeah
circular
q,
yeah
yeah.
It
creates
one
of
those
and
so
yeah,
so
we
have
for
each
instance.
We
have
a
circular
queue
right
for
each
operation,
yeah
for
each
option.
D
B
B
B
B
D
Like
this,
with
the
earlier
implementation,
the
cli
was
not
actually
outputting
properly.
Then
I
had
to
like,
because
I
was
returning
a
dictionary
not
yielding.
D
B
D
B
Be
great
that
will
save
me
time
then,
when
I
review
great
okay
and
then
I'm
also
planning
on
doing
layer
support,
probably
tomorrow
so
I've
been.
I
had
a
reason
why
I
was
slightly
late
today
because
I
was
just
talking
to
my
manager
and
I
have
another
meeting
with
her
at
11
and
I'm
slightly
I'm
trying
I'm
trying
to
finish
up
that
colonel
patch
thing
that
I
was
telling
you
guys
about,
and
so
I'm
I'm
I'm
I'm
under
the
gun
on
that
one.
B
So
so
that's
why
that's
that's
why
things
have
been
lagging
lately,
so
I'm
hoping
I
can
get
that.
I
think
I
think
I
found
all
the
right
fields
and
stuff,
so
I
think
I'm
hoping
I'm
hoping
I
can
get
that
done
this
weekend
too
and
then
and
then
I've
got
some
more
time
freed
up
here.
I
also
want
to
let
you
guys
know
that
I'm
gonna
be
I'm
gonna,
be
like
I'm
taking
vacation.
Obviously
you
know
no
one's
going
anywhere,
but
I'm
taking
some
vacation
yeah.
D
B
Right
right.
D
B
So
and
that's
gonna
be
from
the
because
it's
gonna
be
my
wife's
birthday.
So
well,
you
know
our
wedding
got
postponed,
but
she's,
basically
my
wife,
so
the
fifth
or
the
tenth
so
basically
I'll
be.
I
think
I'm
going
to
try
to
find
it
and
I
may
be-
I
don't
know,
but
the
11th
I
may
find
so
I'm
trying
to
I'm.
I
think
I'm
going
to
try
to
get
sudarsana
or
yash
to
lead
on
the
7th
and
the
11th
or
we
may
just
drop
one
of
those.
B
Let's
see
we'll
see
we
may
we.
It
depends
whether
somebody
somebody
wants
somebody
else
wants
to
lead
the
meeting
or
not,
and
then
I
have
to
figure
out
how
to
help
to
get
everyone
to
have
a
meeting
because
of
google
meet
and
be
having
to
allow
people
in
oh
yeah.
We
used
to
use
hangouts
and
it
didn't
do
that.
That's
right!
B
Maybe
we
could
just
switch
back
to
that.
Who
knows
okay,
anyway,
so
and
then
saksham
you're
also
waiting
you're
still
waiting
on
that
layer,
support
example
and
that's
one
of
those
things
that
I'm
hoping
to
catch
up
on
yeah.
C
B
So
I
think
that
you
should
show
the
editing,
so
we
have
mnist
where
we
show
editing
in
flight
right
in
this,
like
by
in
flight.
I
mean
like
before
we
go
into
train
with
the
pre-processing
source
in
this
one.
Do
you
save?
I
don't
remember
seeing
you
s
yeah,
so
you
don't
you
don't
cr,
you
do
it
in
flight
in
this
one
too.
Don't
you
you.
B
So
I
think
this
because
this
one's
bigger
right
you,
it
might
be-
it's
probably
worth
it-
to
show
how
we
could
edit
it
and
save
it
off
as
something
else.
And
then
that
way,
you
can
show
the
comparison
between
scikit
and
the
cnn
right,
because
I
think
that
that's
still
a
cool
thing
to
show
basically
like.
B
So
you
are
using
this
source
you're
using
the
preprocessing
source
every
single
time.
Let
me
let
me
sorry.
Let
me
take
some
notes
here
because
we're
just
this
is
not
gonna.
Okay,
so.
B
Right
use
instead,
so
what
I'm
saying
is,
instead
of
so
basically
disregard
everything
I
said,
and
now
now
I'm
going
to
restate
it
and
hopefully
a
clear
way
so,
instead
of
using
or
instead
of
using
the
preprocessing
source
for
each
ml
command,
so
dfml
train,
slash
accuracy,
slash
predict,
let's
use
dfml
edit,
to
create
a
new
json
file
containing
the
pre-processed.
B
Data
and
I'm
going
to
go
ahead
and
say
that
you
should
do
like
dot
json.gz
or
something
I
mean
just
play
around
with
that.
You
know:
try,
maybe
adjacent.gz
to
see
if
or
just
like
yeah
try,
try
just
make
it
a
json
file
and
then
like
look
at
the
size.
So
look
at
the
size
now
using
you
can
use
t
t-u-h
file
name
and
if
it
is
really
large,
then
you
can,
like
you
know,
g-zip
it.
So.
B
The
merge
command,
I
mean
oh
yeah,
I
guess
you
would
use
yeah,
you
would
use
the
merge
command.
Yes,
you
would
good
good
good
good
catch.
Yes,
merge
nice
yeah
because
we're
creating
a
new
source
very,
very
nice,
and
we
need
to
add
that
to
the
I
think,
that's
still
a
thing
that
oh
actually
yeah.
Well,
we'll
I
don't
we
won't
really.
B
C
B
All
right
well,
we'll
just
try
it
again
and
let's
see
what
happens,
because
that
should
work
and
if
it
doesn't
work
then
we
should.
We
should
definitely.
B
If
it
doesn't
work,
then
we
should
definitely
figure
it
out
and
at
least
make
an
issue
so
doc's
cli
merge.
I
need
to
document
so
this
is
just.
B
We
just
need
to
do
this
at
some
point,
and
so,
if
you
feel
like
throwing
it
in
here
great,
if
it
doesn't
work,
then
create
a
new
issue
and
and
yeah.
So
let's
go
with
that.
So,
okay,
cool,
yeah,
okay,
so
then
we'll
that'll
that'll
show
how
we
do
the
merge
command.
It
should
save
us
a
bunch
of
time
running
each
of
those
commands.
The
train
accuracy
predict
and
then
we
can
also
maybe
showcase
how
we
can
gzip
the
sources
or
do
the
part.
C
B
C
C
B
Okay,
do
you
want
to
share
your
screen.
B
B
Okay,
so
we
need
to
check
that
export
function.
We
should
be
let's,
let's
pop
open,
let's
open.
B
B
Yes,
okay,
so
where
is
let's
scroll
up
here
export
value?
I
believe
yeah,
the
one
where
we
have,
that
collections
abc
callable
and
stuff
okay,
so
that
was
how
we
found
if
it
was
the
list
right
or
whatever.
C
B
Sweet
okay,
so,
let's
just
say
okay,
this
is
like
this
is
where
we
want
to
add
the
magic
crate.
Okay.
So,
let's
see,
let's
see,
let's
see
yeah
I
mean
I
think
what
we
should
do
is
we
should
expand
that
little
alif
to
just
be
like
you
know
that
alif.
The
first
thing
is.
C
B
Have
sure
it
should
because
well,
let's
see
if
it's
not,
then
we
have
a
problem,
because
I
believe
it
should
call
record
dot
export.
Let's
see,
I
believe
that
where
is
it
basically
yeah?
So
if
it's
not
it
should
because
because
the
thing
is
that
custom
json
the
custom
json
encoder,
that
we
have
is
not
the
best
way
to
do
this,
because
so,
let's
see
because
the
problem
is
somebody's
gonna,
just
not
use
it
at
some
point
right,
especially
like.
B
If
you
look
at
the
http
api,
you
can
make
there's
another
place
where
this
happens
so
but
yeah,
it's
probably
the
case.
Where
is
that?
B
B
B
B
Put
it
into
the
main
export
value
function,
so
util
cli
command
is
where
we
want
to
go.
B
B
B
Yeah,
this
stuff
needs
to
be
properly
combined
at
some
point:
okay,
so
yeah.
So
this
is
what
then
yeah,
let's
put
it
in
the
body
of
that
right
and
then
then
we
can
make.
Let's
see
is,
is
was
flattened
for
the
images
or
was
it
for
arrays?
Is
it
the
same
thing
like
what
is
the
difference
between
two
list
and
flatten?
Do
you
guys
know.
B
D
B
B
B
So
yeah,
I
don't
know
if
we
always
want
to
flatten
it.
B
B
Okay,
so
what
what's
going
to
blow
up,
if
we
do
to
list
instead
of
flat.
B
So
yeah,
let's
see
that
would
be
well!
No
now
you
want
that
to
be.
You
would
want
that
to
be
two
lists
right.
So
so
here
one
second
section.
So,
let's,
let's
one
second
here
so
let's
make
the
alif
yeah.
So
the
the
lf
is
correct
right
now,
but
let's
just
take
that
that
that
stuff
you
commented
out
within
the
alif
and
make
it
its
own
if
statement
within
this
block,
like
so
at
the
indentation
level,
your
cursor
is
at
right
now.
B
Okay,
yeah
right
there
so
grab
that
stuff
and
make
an
if
statement,
yeah
yeah
all
that
yeah
right,
because
what
we
want
to
do
here
is
basically
say
if
it's
numpy
and
it's
an
iterable
and
it
can
be
too
listed,
then
to
list
it.
Otherwise,
if
it
looks
like
an
int,
then
make
it
an
int
if
it
looks
like
a
float,
make
it
a
float.
That's
what
we
want
the
body
of
this
thing
to
be
now
right.
B
Yeah,
so
let's
do
let's
do
two
list,
so
this
is
okay,
so
that
let's
do
flatten.
I
guess
let's
do
flatten
for
now,
because
you
know
it's
probably
gonna
break
things
right,
how's
it
going
sudhanshu,
so
we'll
do
flatten
for
now,
because
it's
probably
going
to
break
things
right
and
then
and
then
we'll
we'll
make
an
issue
that
we
should
change
it
to
to
list,
because
we're
losing
information
that
we
shouldn't
be
losing
the
goal
of
export
is
just
to
change.
B
C
B
So,
let's
just
do
flat
and
we'll
just
change
it
to
flatten
right
now.
Let's
see
okay
and
I'll,
make
an
issue
here
so
yeah,
and
then
you
want
to
get
the
rest
of
that
body.
Of
that.
If
statement,
I
think
you
might
need
a
let's
see
you
might
need,
I
think
it
it
did
the
type
name
at
one
I
think
in
the
in
the
json
within
the
encoder
it
was
looking
at
the
type
name.
B
B
B
B
Okay,
let's
see
yeah,
you
fix
the
indentation
and
then
remove
that
commented
stuff,
and
then
I
think
we
should
be
good
to
go.
B
Oh
yeah,
don't
return
yeah
yeah,
let's
see,
we've
got
to
do
object
key.
We
got
to
remove
that
last
lf
there
with
the
nd
array
and
then
un
do
the
yeah
indent
those
two
for
those
three.
B
B
Let's
see
it
looks
like
oh
yeah,
that's
true,
so
you
want
to
do
value.
Yeah,
oh
yeah,
oof
wow.
The
naming
of
these.
B
B
B
That
I
think
that
that
needs
fixing
there
yeah,
maybe
put
the
and
on
the
other
line
and
the
line
above
it.
B
All
correct:
well,
it's
doing
something
nope!
No,
it's
not
all
right!
Json,
decode
error,
wow.
What
the
hell!
Oh
because,
oh
because
the
previous
file
is,
is
just.
B
B
B
That
statement
well
it,
but
that's
what
I'm
saying
is.
It
must
not
be
being
run
because,
let's
see
the
error
message,
because
type
nd
array
is
not
json
serializable
right,
but
it
should,
it
should
run
it.
If
nd
array
has
I'm
looking
up
the
docs
right
now
in
the
array,
it
should
have
yeah.
It
has
a
flattened
method,
so
maybe
just
try
to
list
well.
Flattened.2
list
isn't
going
to
help
you,
because
it's
still
going
to
hit
that
it's
not
hitting
that
if
statement
so
try
get
at
her
to
list.
B
B
So
I
wonder
if
it's
not
hitting
that,
though
we
should
probably
just
stupid,
there's
no
goddamn
logging
in
here,
okay
yeah.
I
guess
let's
try
to
see
what
this
happens,
but
I'm
now
thinking
that
our
main
problem
here
is
for
some
reason
that
the
module
does
not
equal
numpy.
Let's
do
yeah,
let's
say:
oh,
let's
at
the
end
of
this,
let's
do
an
else
like
a
very
last
one
and
else
and
let's
just
like
raise
value,
error
or
something.
B
Because
we
don't
know
how
to
export
this
numpy
type,
right
so
say
a
no
numpy
type
or
something
and
then
add
the
type
yeah
put
the
type
so
yeah.
You
know
yeah.
A
C
Yeah,
that's
what
I'm
saying
like
if
it
is
an
empire
and
then
it
is
hitting
this
first.
B
B
I'm
saying,
though,
is
it
must
not
be
hitting
the.
It
must
not
be
hitting
this
this
that
the
lf
module
equals
numpy,
because
or
else
we
would
have
hit
that
unknown,
numpy
type,
because
we
converted
it
to
a
list
and
then
it
should
be
a
regular
python
list
right
so
and
we
would
have
hit
the
value
error
if
it
was
a
different
kind.
So
we
must
not.
We
must
have
done
get
after
type
value
module
equals
equals
numpy
and
it
was
not
numpy
right.
So
what
I
wonder.
B
B
Let's
just
do
whatever
we
did
over
there
yeah,
let's
just
do
okay,
so
let's
just
do.
Instead
of
this
get
out
your
module
number
numpy.
Let's,
let's
do
the
it
looks
like
we
did:
alif
numpy
dot
in
type
name
lower.
B
B
B
B
B
B
Screen,
oh
okay,
yeah!
It's
not
it's
not
we're
seeing
vs
code
still,
maybe
turn
it
on
and
off.
Like.
B
Check
that
file
hey
all
right,
okay,
sweet!
How
big
is
that
file?
Let's
check
now
it
would
be
you
dash
h.
B
And
then
the
file
yeah,
this
is
a
3.7
m.
Okay,
that's
not
too
bad,
but
you
could
still
I
mean
you
could
still
add
the
dot
gz
and
then
that
way
it
would
be
that
way.
We
could
show
how
we
can
do
that.
Okay,
so
all
right,
great,
okay,
great!
So,
let's,
let's
add
this:
maybe
you
can
throw
this
up.
You
can
put
this
in
well,
I
guess
you
have
the
flower
pr
and
then
you
have
this.
B
So
let's
make
this
a
separate
pr,
but
okay,
oh,
but
the
other
thing
was
that
we're
now
changing
it
right,
we're
changing
it
from
we
changed
it
from
platinum
to
two
list.
So
let's,
let's
change
it
from
two
list
to
flatten
again
and
then
and
then
so
we'll
merge
this
one
and
then
you'll
have
it
in
the
flower
one
right
and
then
we
need
to
go
through
and
we
need
to
change
it
to
two
list
in
a
separate
one,
because
my
guess
is
that's
going
to
break
a
bunch
of
things
right,
I
mean.
B
B
Okay
looks
like
yo:
we
just
do
record.feature,
so
it's
not
going
to
export
it,
so
we
should
probably
be
fine
to
change
it
to
flatten,
or
I
mean
to
to
list
wait
what
the
hell.
C
B
B
C
B
Were
okay?
Well,
let's
keep
it
as
value.flatten
right
now
I
mean
because
it
doesn't
really
matter
whether
it's
a
tuple
or
a
list
right
like.
B
Okay
yeah:
let's
do
two
lists
great
all
right,
sweet.
Let's
do
that
and
then
and
then
let's.
Let's
call
that
good
and
let's,
let's
let's
push
this
up
here
so
and
as
a
separate
pr
just
this
change,
and
I
think
you
probably
need
to
run
black
sweet
all
right.
So.
B
B
This
this
will
be
a
separate
pr,
all
right,
so
so
sudanshu.
I
realized
that
I
was
definitely
the
wrong
decision
that
we
made
last
time
where
we
said
to
take
out
the
accuracy
I
mean,
and
you
probably
noticed
that
too
right,
because
then
we
now
no
longer
have.
B
Stuff
and
then
I
assume
you
saw
my
comment
too-
that
basically
we
should
just
move
it
back.
B
And
then,
but
you
know,
I
guess
the
the
good
news
here
is
that
we
we
did
we
did.
We
did
make
the
right
decision
initially
by
not
getting
rid
of
that
model
accuracy
method.
So
so
that's
the
good
news,
I
guess
so
we'll,
basically
just-
and
I
think
I
think
I
was
thinking
about
it
a
little
bit
more
and
I
was
thinking
you
know
for
this
case.
Let's
see
what
was
like.
Okay,
so
we've
got.
B
I
was
thinking
you
know,
we'll
have
the
log
message,
but
it
would
also
be
kind
of
nice
to
say
you
know
to
have
an
error
happen
as
well,
and
so
I'm
not
sure
if
you
have
any
how
many
ways
that
you
think
that
might
might
be
good
to
do
that.
You
know
just
try
them
out
or
something
because
so
so,
basically
so,
okay,
let
me
open
the
pr
too,
and
now
I
don't
think
am
I
screen
sharing
again
or
am
I
not
I'm,
not
okay,.
B
All
right,
so,
oh,
would
you
look
at
that.
Somebody
just
put
up
next
boost:
okay,
distributed
nuts
orchestra,
okay,.
B
Okay,
all
right
so
basically
the
so
say:
okay,
so
we
have
the
accuracy
method
right
and
then
the
default
one
in
the
model
base
class
says
you
know.
I
know
how
to
do
this.
I'm
gonna!
You
know
if,
if
there's
one
predict,
if
there's
predictant
features,
then
I'll
just
pass
it
to
to
or
let's
see
if
there
is
features
right,
then
you
pass
it
to
the
score
right.
B
B
What
I
was
thinking
is
that
is
that
the
problem
is
okay,
so
say
you
have
that
use
case
where
you
go
through
and
you
say
well,
let
me
evaluate
all
these
models.
Well,
I
guess
you
wouldn't
be
evaluating
the
nlp
models
anyways,
so
this
is
kind
of
a
non-event.
B
I
don't
know
I
just
sort
of
I
was
just
I
was
thinking
about
that.
Did
any
more
thoughts
on
that,
or
did
that
sound
like
rambling
nonsense,.
B
All
right,
okay,
we'll
go
with
rambling
nonsense
and
no
thoughts
on
that.
So
oh,
no
much
who
can't
join
okay,
so
do
you?
Does
anybody
have
anything
else
they
want
to
talk
about
today,
then.
D
A
C
C
Like
I
should,
I
just
could
share.
B
Okay-
let's
see
so
sudha
yeah,
I
think.
Basically,
I
don't
know
when
you
got
disconnected,
but
basically
the
resolution
is
the
same,
just
sort
of
rambling
about
you
know
the
accuracy
stuff
yes,
and
I
think
that
you
know
just
keeping
them
in
that
method
is
probably
what
makes
sense.
B
Then
you
know
it
would
be
nice
to
have
a
way
for
them
to
raise
an
exception,
because
then
it
would
definitely
get
the
user's
attention,
but
the
user
will
probably
not
be
looping
through
a
bunch
of
different
nlp
models,
trying
to
see
their
accuracy
because
well
and
passing
them
an
accuracy
score
because
it
just
doesn't
work
like
that
right.
That
might
be
something
that
we
would
deal
with
with
like
a
regression
model
or
something,
but
there's
no
way.
B
F
Yes,
like,
like
two
of
the
models,
are
not
in
the
like
the
spacey
model
and
the
by
torch.
Sorry.
F
B
Space,
oh
yeah,
because
we've
added
new
models
now,
but
those
that
spacey
model
is
another
nlp
one.
So
I
don't
think
I
think
you
can
just
leave
that
one
too,
but
you
will
need
to
modify
the
method
signatures
now
right,
so
you're
going
to
need
to
have
them,
accept
the
accuracy
score,
so
so
basically
yeah,
instead
of
let
me
edit
this
right
now
and
and
we'll
make
this
so.
B
Oh,
and
I
think
we
also
gained
pi
torch,
didn't
we,
but
actually.
B
In
that
branch,
well,
you
can
you
should
oh
yeah
you
should
you
should
merge
in
the
other
branch,
because
yeah
yeah,
you
should
merge
in
the
master
branch
into
this
yeah.
You
should
merge
in
the
master
branch
to
this,
because
then
you'll
get
those
models
and
it's
going
to
show
up.
You
should
probably
do
that.
B
We
should
probably
we'll
we'll
hit
merge
on
this
pull
request
after
it's
done
and
then
we'll
do
the
same
thing
for
whatever
has
been
added
in
between
then
and
and
so
basically
take
take
the
nlp
models
that
are
in
there
right
now
modify
the
accuracy
method,
signature
to
accept
a
scorer.
B
But
just
log
or
but
log
warning
that
it
won't
be
used
then,
and
then
we
need
to
so
we'll
merge,
we'll
merge
this
bit
and
then
and
then
we'll
just
say:
phase
3.5,
because
since
we've
gotten
so
phase.
B
3.5
so
create
pr
that
merges
or
that
okay.
So
we
need
to
take
the
content
from
the
master
branch
right
and
we
need
to
put
it
in
the
accuracy,
score
branch
and
and
then
we
need
to
do
the
same
thing.
We
just
did
on
the
ones
that
weren't
there
right.
F
F
So
like
in
the
phase
3.5,
I
will
have
to
like
like
manage
that
by
torch
once
right.
These
space
events
are
not
compatible.
B
Let's
see,
oh
wait
a
minute:
okay,
the
space
yeah,
the
spacey
ones,
are
the
same
thing
as
tensorflow
and
transformers,
and
I
don't
know
why
I
wrote
pi
torch
here.
Why
did
I
write
pi
torch?
I
thought
pi
torch
is
the
computer
vision?
Do
we
have
pi
torch
models
that
are
that
are
nlp
ones?
Do
we?
I
think.
A
B
B
No
transformers-
maybe
let's
see
okay,
so
it
never
really
was
okay.
No,
he
hadn't
done
that
then,
okay,
so
I
don't
know
why
I
must
have.
How
did
I
I
don't
know
how
I
ended
up,
putting
it
there,
so
that
should
go
that's
basically.
B
This
is
like
space,
3.333,
3.3
repeating
and
then.
B
B
B
B
I
mean
not
predictions
prediction,
but
the
pytorch
models.
They
should
be
able
to
work
fine
with
them.
B
It
should
work
fine
with
the
with
the
existing
ack
like
the
new
accuracy
method,
because
it
should
just
run
predictions
and
then
you're
calculating
accuracy.
By
doing,
let's
see.
C
I'm
taking
the
prediction
and
then
just
checking
the
label
and
then.
B
F
Yes,
okay,
cool
all
right,
so
the
tensorflow
hub
is
actually
done.
Okay
and
the
transformers
like
I
have
reverted
it
back.
The
changes.
Okay,.
B
Great
sweet
all
right,
awesome,
awesome,
okay,
so
we
found
out
which
is
now
I'm
real.
I
you
know.
I
realized
now
that
we
had.
I
just
removed
that
bullet
point,
but
we
must
have
actually
known
what
to
do
and
then
and
forgot
about
that,
because
we
had
a
bullet
point
telling
us
that
we
should
have
just
modified
the
accuracy
signature.
B
We
found
out
that
we
should
just
be
modifying
the
accuracy
signature
for
nlp
models,
we'll
need
to
merge
new
models
in
master
into
accuracy,
score
branch,
okay,
so
and
then
oh,
no,
we
lost
himachu
again
he
joined
and
we
lost
him.
Damn.
B
Okay,
so
all
right
so
sutashu
are
you.
Are
you
feeling
good
about
path
forward?
Then?
Yes,
yes,
thank
you,
sweet!
Thank
you
all
right,
then
so
sakshan.
What
was
the
other
thing
you
were
saying.
C
B
C
Wait
which
stock
string
issue?
Oh
sorry,
data
flow
source
docs
too.
C
So
I
had
some,
I
wanted
to
ask
you
something
about
the
async
functions
and
so
yeah.
Let's
check
it
out,
I'm
sharing
my
screen
too,
so
that
you
can
see
what
I'm.
C
F
C
B
Let's
see
got
okay
or
wait:
god
expected
nothing!
Yeah!
Okay,
oh
so
maybe
you
need
to
put
in
blank
line
like
that,
so
maybe
copy
paste
that
blank
line.
No,
it's
not
expecting
anything.
C
B
So
the
reason
it's
not
expecting
anything
is
because
it's
expecting
to
see
some
output
right
there
right,
and
so
it's
saying
what
it
got
was
a
blank
line
and
then
this
stuff
right.
So
I
don't
know
what
the
format
is
to
expect
a
blank
line,
but
I'm
guessing
it's
putting
blank
line
right
there.
So
we
can
try
that.
B
B
B
B
C
E
B
E
B
That's
a
good
one.
I'm
excited
to
see
that
one
all
right
cool.
Well,
let's
see,
let's
see
if
I'm
watching
his,
I
will
say
something
in
the
chat.
B
All
right,
cool,
well
I'll,
talk
to
you
guys
later
and
thanks
everyone
and
have
a
great
weekend.