Maybe add a button for molab?

[koaning]

For a while now, it seems that whenever I try to run something with Binder on an open source documentation page that I get something that looks like this.

I also had a conversation with @twiecki this afternoon and the topic of marimo came up a few times. To be upfront, I work at marimo, so naturally I asked if maybe y'all would be interested in hosting a run in molab button in your gallery examples. Molab is effectively like colab but with marimo notebooks. It's a service hosted by the marimo team.

I'm certainly willing to help out with the PR and do some legwork for it, but before exploring it I figured it best to first check in to see if there are any concerns with this idea.

Let me know!

[koaning]

Oh, and to be explicit, I am talking about adding a button here:

[fonnesbeck]

Hey Vincent! The examples wouldn't work in Molab out of the box, since they are Jupyter notebooks, would they? Or are you proposing porting some of the examples to marimo notebooks?

In any case, its probably time to retire the Binder buttons, I agree.

[twiecki]

I think it's a worthwhile effort to make them compatible, even if just for folks that want to use the notebooks from marimo locally.
AI probably can do the porting mostly automatic.

[koaning]

We have some utilities on the marimo side of things to attempt an automatic conversion. We also have mechanisms to point to a file on Github in order to load that, so there should be plenty of options. Is there a preference on your end?

There are some scenarios, now that I think of it, where an automatic conversion might be hard, especially if you reuse the same variable across cells and redefine it. That's something that marimo typically doesn't like. But that's the main culprit that I can come up with.

[twiecki]

Probably that tool would not be able to fix the with pm.Model() as m reassignments? could just do m1, m2,....

If it does I think it would be better to update the notebooks directly, so they can also be opened by marimo directly.

[koaning]

I'm talking to folks internally now to see how we can make the intergration straightforward. I would imagine that being able to point to the ipynb files on Github would be preferable?