GitLab MLOps, 20 Jan 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: MLOps SEG Update Jan 20th 2022

Description

- Text Version and Links: https://gitlab.com/gitlab-org/incubation-engineering/mlops/meta/-/issues/43
- All updates: https://gitlab.com/gitlab-org/incubation-engineering/mlops/meta/-/issues/16

A

Hello, everyone and welcome to another update for envelopes on the sag on this incubation engineer here at gitlab this week is going to be a very short one, I'm going to talk about what I did last week and this week um so nothing much uh following up on the next. On the on the previous uh update that we had where I shared uh what I wanna, what I'll be working on in the next few months, I would say there are four areas that we are mainly looking at.

A

First is render drupal notebook, devs, glitter, uh well, runny pipelines for data science within gitlab, uh increasing awareness of the data science use cases within gitlab and analytics repository. So these are the four areas. I'm gonna start uh splitting my updates on this one. So for the first one uh render notebook diffs. What I've been working on is mostly adding the capacity or the ability for users to change to swap between the raw and the rendered diff or declinative for now. So we are moving forward with this.

A

So, for example, on this commit here I have these buttons, but they will be working later, uh just swap it in between the versions and on this one we can already see that we have both the uh classic diff, but also the cleaner def. I have to implement now the hiding between one and another, but at least this part is done. uh We have both of them coexisting and now we need to move on and implement, sharing the for sharing the comments, for example.

A

Here uh the comment is in online 67 of the classic of the raw, but it shouldn't be on line 67 of the the cleaner div. So this is something that we need to work on. uh Moving on, we have glitter, which we didn't do anything uh this week. um Yeah nothing should be out over here.

A

uh Next, we have internal awareness, awareness of data science use cases on this. One.

A

What we're trying to do here is to help the company understand the use case that we're talking about uh it's not enough to have one large stage, I believe, to be working on envelopes. uh Of course, this is great and will really push the project forward, but in addition to that, it's great if the stages that already exist, the people that we already have start assimilating these use cases as well.

A

So what I did uh what we did on this, I recorded a session with uh with nadia, uh which is a project designer on a verify stage, uh just explaining on my perspective, what how data scientists use gitlab, we'll be working on more version or more additions to this style uh coming soon, but uh it just start I'll, be linking every uh I'll be adding the links into my uh issue update um and the fourth uh area. We a little bit more exploratory that we to work on is analytics repository.

A

Just a reminder of what it is analytics repository is a place to help data. Scientists discover share comment on the research or on the studies. They already do so like I mentioned before, but I don't know 50 40, of the work of a data scientist to create machine learning products, there's entire site of creating business, uh business value or assisting business and creating decisions which are done mostly as report like the final output of their work is a report, uh but the notebook or a decoder was used to create.

A

A report is also very important, and some and often on large companies- this is completely lost. uh Analytics repository is something uh to solve that um think like a cms for a data scientist.

A

So on that note, what I did look this week was an idea of uh it was a tutorial shared on twitter on how to run jupyter light on with github actions, I'll change this to get lab uh pipelines later or gitlab pages later. But this is something very cool, because this is running on the browser.

A

um There is no server here, so I can open a python notebook and this uses a webassembly and compiles the the python into uh webassembly. So the entire thing here is just running python on the uh on the browser which is really cool.

A

If you think about it, because one of the use cases that you had for the analytics repository uh when I implemented this before on organizational workforce was tutorials, so people would just create tutorials entrepreneur, books and add to that analytics repository that others could try out, and you could have an entire tutorial running on the browser rather than depending on the server or anything like that, which is a pretty cool idea. You could also have, for example, run books running like this.

A

I don't know it's very, very interesting and something we can explore down the road. um So those were the four uh points that I had for this week. Very short. uh Well, thank you all for uh watching this one. Oh one thing, one last thing that I forgot you might have been missing. uh I was talking before about motor registry and I'm not talking about motor registry anymore.

A

Why? Because motor registries, we feel that it's just such a certain use case of gitlab for machine learning that it's not even.

A

Useful to have a poc a profile concept here and I myself, as an incubation engineer, cannot maintain an entire model registry uh as a feature, so I am not going to implement uh moderator, I'm not going to be working on moderators. That doesn't mean model register isn't coming to gitlab, it's just not coming through me and that's the update that I had for today. Thank you all for joining and see you the next time.