GitLab Incubation Engineering, 20 Dec 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Incubation Engineering MLOps - Update 2022-12-20

Description

Jupyter Diffs Stats and progress on MLFLow Integration

- Request for testing: https://forms.gle/pALYkCFbz9jZmfXU7
- Feedback Issue: https://gitlab.com/gitlab-org/gitlab/-/issues/381660
- This Update: https://gitlab.com/gitlab-org/incubation-engineering/mlops/meta/-/issues/64
- All Updates: https://gitlab.com/gitlab-org/incubation-engineering/mlops/meta/-/issues/16

A

Hello, everyone and welcome to another update for mlops uh incubation engineering at gitlab. My name is Eduardo and today we're going to talk about how my progress on machine learning, experiment, tracking and uh some stats on the Jupiter diffs that we worked before so uh since our last update. The features that we got merged is now that the Emma flow tags are now tracked.

A

So if you have tags on your uh tracking code, it is saved on gitlab, although it's not displayed yet the the UI is coming soon. We added a detailed page for candidates, so more details for your candidate or what would be an ml4 run, and additionally, we are now recording Run Names.

A

So if you, if you give names to your runs um so that you can search for them better on the on mlflow, we're now recording them on gitlab and we'll add soon the UI for displaying this on cute lab candidates, we are going through the internal testing feedback as well, so we already had our internal users here, gitlab using it, and we got some good feedback. So I was focusing more until now on the ux. On the user.

A

Experience of just getting it set up and making it work with minimal uh effort by the user and that works well. What is missing right now with feature, uh and the biggest feature that is currently missing, is better to search and uh featuring on the on the candidates. So this is what I'm going to work uh now. uh This is my next to add on the mlflow integration on machine learning, experiment tracking from gitlab um some interesting metrics, that uh from from the last few weeks, November was the month.

A

uh This is for the Jupiter book depths. If you remember correctly, this was what I was working on before uh mflo integration, and it is very interesting that November was the month with the highest number of comments on Jupiter notebooks ever we- and this is interesting because we haven't put any effort on this on the past six months. So the effort that we put a long time ago on enabling uh diffs particular books is consistently getting value out of it.

A

So users consistently come keep coming back, even though we are not adding more effort into this until now. So that's very cool to see um and yeah. So, what's next, we will continue working on machine learning, uh expert tracking and the next big feature we intend to add is searching and browsing candidates.

A

So when you go to your uh or to the experiment page and you display, it shows that all the different candidates, we want to add the search filtering, uh better uh information being displayed over there, so it just improves General ux and that's it. What I have for today. Thanks have a good one.