►
From YouTube: Gitlab 15.11 Model experiments + MlFlow Integration Demo
Description
A demo of the recently released features Model experiments and MlFlow integration on GitLab
Documentation: https://docs.gitlab.com/ee/user/project/ml/experiment_tracking/#machine-learning-model-experiments
Training Code: https://gitlab.com/gitlab-org/incubation-engineering/mlops/mlflow_experiment
Feedback: https://gitlab.com/groups/gitlab-org/-/epics/9341
A
Hello,
everyone,
my
name,
is
Eduardo
I.
Am
the
incubation
engineer
for
my
labs
here
at
gitlab
and
today,
I'm
going
to
be
showing
you
a
little
bit
of
a
demo
of
model
experiments
and
our
ammo
flow
integration
running
at
gitlab.
All
of
these
features
are
already
available
on
gitlab.com
and
are
also
available
on
self-managed
customers
for
version
15.11
and
onwards.
A
The
most
common
or
most
popular
tool
for
this
is
ml
flow
and
we
are
releasing
a
an
integration
with
ammo
flow,
where
gitlab
becomes
a
backend
for
them
of
local
client
I'm,
going
to
show
you
a
little
bit
over
here.
So,
for
example,
you
have
here
the
ammo
flow
classic,
mufflow
client,
automl,
MFL
server
and
I'm,
going
to
track
a
a
series
of
models
through
the
script
that
are
going
to
be
make
available
as
a
on
the
link
below
it
will
track
a
series
of
experiment
of
of
runs
in
Channel
flow.
A
So
here
I
have
a
really
nice
demo
and
it
created
the
the
runs
with
with
all
of
the
metrics
and
parameters
and
everything
else,
and
for
the
data
scientists
that
wants
to
use
gitlab
to
track
those
so
already
created
here.
A
model,
a
project
so
I
can
just
change.
Instead
of
of
using
these
variables,
I
will
use
these
ones.
A
So
it's
a
it's
a
token
that
gives
permission
and
the
path
to
this
project
so
that
it
knows
where
to
log
things
at
so,
I
can
navigate
to
package
and
Registries
and
model
experiments.
I
see
that
I
don't
have
any
experiment
over
here,
so
I'm
going
to
create
a
new
and
start
logging
them
to
gitlab.
A
So
once
I
refresh
this,
so
there
was
no
change
to
the
code
itself
for
the
training.
Just
where
to
save
things
to
I
can
refresh
I
can
see
that
this
the
experiment
was
created,
I
can
click
on
it
and
now
I
see
all
of
the
candidates
that
were
created.
This
is
still
being
logged,
so
I
can
just
refresh
and
it
will
show
all
of
the
the
the
the
parameters
and
the
metrics
over
here
as
well.
A
It
already
also
locked
the
artifacts,
so,
for
example,
I
can
click
on
artifacts
and
it
will
show
the
express
the
the
it's
part
of
the
experiment.
It's
the
the
candidate
number
18
and
I
have
all
of
the
all
of
the
model
files
that
were
logged
through
ammo
flow
that
without
any
configuration
necessary
by
the
data
scientist,
so
I
can
come
back
to
the
model.
A
Experiments
I
can
also
see
all
of
the
information
for
a
specific
candidate
over
here,
and
we
are
naming
here
candidate
because
a
little
bit
different
than
mflo
MFL
numbers,
it
has
runs,
we
name
as
candidates,
because
candidates
are
candidate
to
become
a
model
version
and
that's
why
the
choice
of
the
naming.
We
also
want
to
add
model
registry,
and
this
will
tie
into
the
model
registry
later.
So
I
have
all
of
the
information
that
was
logged
to
the
specific
candidate.
A
I
can
delete
the
candidate
if
I
want
to,
and
over
here,
I
can
also,
if
I
want
to
explore
this
data
to
create
a
report
or
anything
like
that.
I
can
download
this
data
as
a
CSV
file
and
it
will
open
the
CSV
file,
and
that
was
a
quick
demo
of
the
current
state.
We
are
working
on
more
features
working
on
tying
this
in
integrating
to
the
pipelines
working
on
adding
the
model
registry
and
manual
internet
features,
but
this
is
the
current
version
as
15.11.
Thank
you
for
watching
have
a
great
day.