tech.lgbt is one of the many independent Mastodon servers you can use to participate in the fediverse.
We welcome all marginalized identities. This Mastodon instance is generally for folks who are LGBTQIA+ and Allies with an interest in tech work, academics, or technology in general.

Server stats:

2.9K
active users

Our machine learning weather forecast is public! 🌦️

Right on time for the weekend, too!

It's an alpha version for now, because the whole team is working hard to wrestle every iota of performance from our graph NN .

But even as an alpha version, it's playing in the big leagues.

I am so excited to finally share this with you! 🚀

I'll link our blog post and the forecast below. Make sure to share this with your colleagues and machine-learning-curious friends!

Still can't believe how far these people have made it in this short time, running four machine learning models operationally, with one developed in-house.

The Blog: ecmwf.int/en/about/media-centr

Your Weekend forecast: charts.ecmwf.int/?facets=%7B%2

More to come soon!

Public

@jesper congratulations.

It was way faster than I expected. Considering also GPGPU computing, another long expected performance boost, has not been integrated into IFS yet. Perhaps the code base being completely new played a role?

I don’t have the knowledge to comment on the core elements. Let’s stay light then and ask why did you choose to utilize the AI moniker? And not the more proper ML.

Thanks for the excellent teamwork.

Public

@WuMing2 Thank you!

Lots of good questions. Let me see which ones I can answer!

I agree that GPGPU computing has some great benefits for part of scientific computing. But in my experience, it's also not super easy to get right. Anytime you adapt a codebase, there's a lot of moving parts. For AIFS, the clean slate and the fact that we can work with Python instead of Fortran, C, and CUDA really is a huge advantage in speed of development. Also, we're in a bit of a nice position with AIFS, since we don't know how good it can get (only upsides), whereas with the IFS, we need numerical guarantees for an operational system. Overall a much harder situation, in my opinion.

Also important that AIFS currently is very much in an alpha state. Lots to do before it can reach its first cakeday.

The name for the AIFS was honestly chosen way above my pay grade. I do appreciate sort of worplay though.

Thanks for the lovely words!