Speedup Open-Sora's training by 3x and inference by 2x with our novel DSP (Dynamic Sequence Parallelism)! For 10s 512x512 videos, Open-Sora's inference time: 1xH800: 106s 8xH800: 45s 8xH800+DSP: 22s DSP can be seamlessly adapted to all multi-dimensional transformers, unlocking

Totality 2024!

It's been a wild ride. Just 20 of us, burning through thousands of H100s over the past months, we're glad to finally share this with the world! 💪 One of the goals we’ve had when starting Reka was to build cool innovative models at the frontier. Reaching GPT-4/Opus level was a

Llama3 8B and 70B are out, with pretty exciting results! * The ~400B is still training but results already look promising. * Meta's own Chat interface is also live at meta.ai * TorchTune integration is shortly going live: github.com/pytorch/torcht…

Kathy Yelick on Post-Exascale Challenges — hpcwire.com/2024/04/18/kat…

Real2Code -- translating real-world articulated objects to sim using code generation! With the code representation, this method scales well wrt the number of object parts, check out the 10-drawer table it reconstructed 😉

This seems like a cool research place to join mlcollective.org/?s=09

Hi all prospective grad students! Our Equal Access to Application Assistance (EAAA) program for UC Berkeley EECS is now accepting applications! Any PhD applicant to UC Berkeley EECS can submit their application for feedback by Oct. 6th 11:59PM PST forms.gle/rYVtXyaE9qE8ug…

I created a video to help explain how LLMs work. Hope you'll find it useful! youtube.com/watch?v=6o0Xnl… P.S. Creating videos is hard, but kind of fun experience

Easily the most impressive uncut autonomous video I've ever seen, from Physical Intelligence

The reality of the Turing test

Growing up in Sri Lanka coffee was not a regular drink, but after being in Seattle for a while, and trying out freshly roasted beans and brewing at home, I just don't want to miss it ... psst mine tastes better than starbucks :)

New paper: turns out you can train deep nets without normalization layers by replacing them with a parameterized tanh()

Kohl's has just become an Amazon return center only 🙂

#GrokImagine is absolutely mind-blowing for both the quality and speed of generation. Almost scrollable speed

He loves attention so much!

Lyrics I had written sometime back comes to life with Suno.ai. Enjoy! Pata Aran Akase | පාට අරන් ආකාසේ .... youtu.be/w1K-FEhcIW0?si… via YouTube

Nothing beats developer productivity than a desktop under your desk :)

When I created Claude Code as a side project back in September 2024, I had no idea it would grow to be what it is today. It is humbling to see how Claude Code has become a core dev tool for so many engineers, how enthusiastic the community is, and how people are using it for all

Some conversations make it obvious we’re in a once-in-a-generation shift in AI. Sid Sheth shared his take on the watershed moments that brought us into the #AgeOfInference — and why breaking the memory & bandwidth wall now defines all future AI infrastructure. Great to dig into