►
From YouTube: Gitlab Model Experiments Tracking - MVP Demo 20230206
Description
A demo of the implemented features for Gitlab Model Experiment Tracking and the MLFlow integration.
Epic: https://gitlab.com/groups/gitlab-org/-/epics/9341
A
Hello,
everyone,
my
name,
is
Eduardo
and
I'm
gonna
show
today.
What's
the
state
of
the
ammo
flow
integration
at
gitlab,
I've
been
building
this
for
a
few
months
and
I'm
gonna
give
an
overall
view
of
what
it's
capable
of,
and
evolution
I'm
not
going
to
go
into
sales
or
what
model
experiment
tracking
is.
There
are
other
sources
for
this
just
about
the
features.
So
on
your
right
side.
On
your
left
side,
you
already
have
the
UI
for
the
list
of
the
experience
currently
available
for
this
project.
A
On
your
on
your
right
side,
you
have
an
experiment
or
a
script
that
will
train
a
model
and
will
we
will
use
this
just
to
store
candidates
or
runs
into
gitlab.
This
is
taken
from
the
ml
flow
tutorial.
A
It's
a
very
simple
script.
Nothing
Fancy
was
done
with
it
and
we're
going
to
show
how
this
works.
So,
for
starters,
you
we're
gonna,
create
a
new
experiment
and
for
running
the
only
thing
that
we
need
to
do
differently
than
save
into
ammo
flow
itself
is
on
our
environment.
We
need
to
set
up
two
environment
variables.
One
is
the
track
URI,
where
we're
going
to
save
the
information,
the
second,
the
token
that
for
the
user,
that's
going
to
be
saving
this
for
authentication
purposes.
A
So
now
I
can
run
this
and
it
will
start
creating
some
some
ex
some
candidates
over
here,
I
can
refresh
my
page
on
this
site.
You
created
the
experiment
on
this
side
on
gitlab
and
it's
starting
to
show
the
candidates
over
here.
So
if
I
just
now
go
this
way,
I
can
already
see
that
I
have
a
bunch
of
different
candidates
appear
over
here
and
I.
Can
this
is
I,
can
now
search
for
the
name?
All
of
them
have
the
same
name,
but
you
get
the
gist
of
it.
A
You
can
search
for
it
and
you
can
also
you
can
also
sort
by
symmetric
or
some
that
you
have
so
this
in
this
case
R
square.
It
shows
the
the
name
of
the
user
that
created
this
this
candidate,
the
date
it
was
created.
You
can
also
see
if
there
is
an
artifact.
If
the
artifact
was
logged
to
the
to
the
candidate,
you
could
come
over
here
and
then
it's
going
to
display
the
all
the
files
related
to
that
candidate.
Remember
a
candidate
for
gitlab
is
what
a
run
is
on
ammo
flow.
A
So
here
you
have
the
Pico
file,
you
have
the
requirements
and
everything
else
it
was
generated
by
amazonflow,
and
you
can
also
you
can
check
the
details
of
a
specific
candidate.
You
should
go
to
details
of
it
and
you
come
over
here.
It
shows
that
at
experiment,
but
it
also
shows
now
that
we
are
logging,
the
the
metadata
for
that
run
or
for
that
candidate.
A
So
it
has
the
the
commit
the
hash
and
everything
that's
necessary,
the
the
original
user,
whether
it
was
local
or
not-
and
this
is
the
current
feature
set
of
experiment
tracking.
Thank
you
for
watching.