►
Description
Text version: https://gitlab.com/gitlab-org/incubation-engineering/mlops/meta/-/issues/21
All Weekly demos: https://gitlab.com/gitlab-org/incubation-engineering/mlops/meta/-/issues/16
A
A
As
usual,
we
have
this
issue
where
we
collect
all
of
the
weekly
demos
feel
free
to
subscribe.
If
you
wanna
know,
when
the
updates
come
up,
come
up
so
yeah,
so
for
this
week
we
will
continue
working
on
jupyter
notebooks
experience
and
exploring
how
to
connect
better
gitlab
to
the
envelopes
landscape,
which
tools
could
we
better
implement?
Which
opportunities
should
we
pursue?
So
this
will
be
talking
about.
A
So
before
we
start
working
on
the
div
itself
of
the
jupiter
notebook
to
get
a
little
bit
warmed
up.
We
are
working
on
small
small
issues
with
jupiter,
so,
for
example,
we
had
one,
mr
now
that
fixes
images
that
are
loaded
from
the
repository
itself
rather
than
the
url,
are
embedded
into
the
notebook.
A
So
this
is
one
of
the
things
we
worked
on
another
one,
small
improvement
as
well
images.
Output
images
were
not
being
rendered
well,
they
were
being
rendered
correctly,
but
they
were
not
being
resized
correctly.
So
we're
just
a
small
thing
here,
a
little
bit
of
a
larger
one.
A
lot
of
users
complain
about
how
we
render
or
how
we
don't
render
math
well
independent
books.
That's
because
of
the
library
we
use,
we
use,
ktex
and
kind
of
like
the
industry
standard
is
math
checks.
A
So
one
of
the
ideas
that
we
want
to
work
on
now
is
replace
one
by
the
other
and
start
using
math
tracks.
It's
a
lot
better.
That
has
a
lot
better
support
for
equations,
so
we'll
be
working
on
this
soon
and
regarding
the
jupiter
notebook
steph,
we
are
exploring
different
ways
of
doing
this,
of
implementing
this
this
stuff,
so
just
to
show
a
little
bit
of
the
the
challenge
that
we
have.
So,
for
example,
this
is
a
tripper
notebook.
A
It
has
errors,
it
has
lines,
it
has
everything,
and
this
is
the
file
the
json,
the
jupy.json,
and
it
has
each
cell
that
you
see
here
is
an
entry
over
here.
So
you
have
cell
types
with
a
lot
of
metadata,
so
this
cell
type
is
a
code.
It
was
the
seventh
executed.
It
has
one
id
the
output
was.
It
was
an
error,
so
you
would
see
it's
this
one
over
here.
A
It
also
has
a
trace
back.
It
has
the
source
here,
so
you
see
that
the
source
are
just
each
line
is
part
of
a
list,
and
now
we
have
to
ch
to
diff
this
thing.
So
one
of
the
ideas
that
we're
having
now
is,
instead
of
generating
text,
generate
a
jupyter
notebook
which
contains
diff
per
cell,
so
each
cell
has
its
own
depth.
We
can
use
the
ids
to
see
which
cells
were
removed
or
added,
and
then
for
the
cells
that
were
kept.
A
We
can
diff
and
see
what
were
the
changes
that
we've
been
done
and
then
do
a
little
bit
more
of
rendering
on
the
front
end.
So
this
is
other
way
that
we're
exploring
different
ways.
So
this
is
low
discovery
process.
We're
testing
different
ideas
on
how
to
move
forward
with
this,
but
I
think
this
looks
a
little
bit
promising
and
we'll
see
on
the
upcoming
weeks.
A
A
So
we
need
a
point
of
entry
for
that.
We
need
to
better
understand
and
better
quantify
what
are
the
needs
of
our
users
and
for
that
we
need
to
better
capacitate
our
sales
people,
the
contacts
we
have
with
directly
with
the
customer
so
that
they
can
have
discussions,
inform
discussions
and
be
able
to
understand
the
customer
needs
on
develop
side.
A
So
this
is
also
ongoing.
We
had
some
discussions
this
week
over
this
on
how
to
proceed
and
what
are
the
needs
of
our
account
representatives
as
well,
and
how
we
can
better
help
them
out
and
we'll
be
moving
forward
with
this
as
well
and
last
but
not
least,
we
have
also
on
this
line
of
exploring
how
to
integrate.
A
One
of
the
ideas
now
is
to
create
sample
projects
that
connect
with
all
of
the
major
that
we
can
mix
and
match
the
major
providers
on
envelopes
space
so,
for
example,
for
model
registry
with
model
registry
being
a
place
where
you
store
models
where
you
can
discover
models,
or
tomorrow,
artifacts,
where
you
can
track
them,
and
one
of
the
biggest
players
on
this
space
is
enough
flow.
It's
an
open
source
tool
by
maintained
by
databricks.
A
They
offer,
of
course,
a
a
complete
solution
with
a
lot
more
stuff
as
a
sas,
but
they
have
them
on
floors
open
source.
So
one
of
the
ideas
is
to
explore
a
little
better
how
to
integrate
them
outflow
in
gitlab
and
see.
If
this
is
a
way
we
want
to
move
towards,
how
can
we
add
more
flowing
as
a
backbone
to
the
cicd
ml
process?