Jannis Schönleber

PhD at ETH Zürich

[Jannis Schönleber @joennlae]

Thrilled to announce my latest work 🚀 We use the Maddness method in our latest #StellaNera project, a general approximate matrix multiplication acceleration with unmatched energy efficiency. Paper: https://t.co/jXxLSyLGLO (1/5)

[PULP Platform @pulp_platform]

We introduce Stella Nera, a fully parametrizable, open-source Maddness accelerator implemented in a 14nm. Check out @joennlae's "Stella Nera: Achieving 161 TOp/s/W with Multiplier-free DNN Acceleration based on Approximate Matrix Multiplication" on arXiv: https://t.co/u6xLxA2Pye

[Jannis Schönleber @joennlae]

@maximelabonne 8k context length could be a show-stopper for many applications

[Jannis Schönleber @joennlae]

@karpathy Very Nice :-) I can not offer full bare-metal, but GPT-like transformer training only using Numpy (heavily based on micrograd) in ~500 lines (also no PyTorch)

https://t.co/XXyqTgMVzZ

[Jannis Schönleber @joennlae]

@Underfox3 @LucaBeniniZhFe Thank you for sharing 🚀🚀
More information can be found in this thread:
https://t.co/31mTmJHK32

[Jannis Schönleber @joennlae]

Thrilled to announce my latest work 🚀 We use the Maddness method in our latest #StellaNera project, a general approximate matrix multiplication acceleration with unmatched energy efficiency. Paper: https://t.co/jXxLSyLGLO (1/5)

[Jannis Schönleber @joennlae]

@pulp_platform More information can be found in this thread 🚀
https://t.co/31mTmJHK32

[Jannis Schönleber @joennlae]

Thrilled to announce my latest work 🚀 We use the Maddness method in our latest #StellaNera project, a general approximate matrix multiplication acceleration with unmatched energy efficiency. Paper: https://t.co/jXxLSyLGLO (1/5)

[Jannis Schönleber @joennlae]

We're excited to share our work with the #AI community. All details, scripts, and implementation of Stella Nera are available on GitHub. Let's drive sustainable AI innovation together! 🤝 (5/5)
https://t.co/F4styp7C5a

[Jannis Schönleber @joennlae]

Thrilled to announce my latest work 🚀 We use the Maddness method in our latest #StellaNera project, a general approximate matrix multiplication acceleration with unmatched energy efficiency. Paper: https://t.co/jXxLSyLGLO (1/5)

[PULP Platform @pulp_platform]

We introduce Stella Nera, a fully parametrizable, open-source Maddness accelerator implemented in a 14nm. Check out @joennlae's "Stella Nera: Achieving 161 TOp/s/W with Multiplier-free DNN Acceleration based on Approximate Matrix Multiplication" on arXiv: https://t.co/u6xLxA2Pye

[Jannis Schönleber @joennlae]

Testing with ResNet9 on CIFAR-10, Stella Nera showed that high accuracy can go hand-in-hand with energy efficiency. It's a testament to the potential of our approach in practical AI applications. 📊 (4/5)

[Jannis Schönleber @joennlae]

@jenspyra Big fan of Jill Newslady and @colbertlateshow. Watching it from Switzerland since election '16. Made an app based on what I learned in this time😀https://t.co/Pt6svUDAVV #trumpyCrush

[FantasyO Ticker @fantasyOnews]

230 full teams are registered! Official season starts tomorrow evening with the Swiss night champs, continued with the 2nd Swiss National Event on Sunday. This means after this weekend we will have our first leader. Let's get this party started!

[FantasyO Ticker @fantasyOnews]

We're live! Create your fantasy team for this year's season now on https://t.co/xIK45L4Zei and share the hype with your friends! Deadline for all entries is Friday, 23rd of March.

[Jannis Schönleber @joennlae]

@stuetzpunktbern Eimol meh isch meh nid ganz de Basler noche choh. #derby #31joohr

[Pascal Buchs @pascal_buchs]

Follow me tomorrow for the final of the SwissOrienteering Grand Slam! 😊 Watch the live video here: https://t.co/48qPMEjcWL

[THOMAS CURIGER @thomascuriger]

This Saturday, I'm starting at the final of the Swiss Orienteering Grand Slam, where the best orienteers of Switzerland will fight in a mass-start race for the most important national trophy. Follow the event live on facebook at 15:00 CET: https://t.co/CSYpYu23my

[Jannis Schönleber @joennlae]

@ChamuelZbinden @remoruch

[Jannis Schönleber @joennlae]

@alaindenzler If I take a look at my calculations. You could run a 9:45 which would be worth my predicted 9:19 (at sea level and no wind). Go for it!

[Jannis Schönleber @joennlae]

@alaindenzler Wo isch de glaube an diä zweiti amtsziit bliebe? https://t.co/miKKEEUIC5

[Jamie McGeever @ReutersJamie]

Trump now odds on - 55% probability - to leave office before the ends of his first term, says @Betfair

[Jannis Schönleber @joennlae]

@riccardorancan My nutrition plan compared to yours 😉

[aj @ajlkn]

If only z-index: 99999999 could fix America too