Nathaniel Simard
@nath_simard
Founder & CEO @Tracel_AI Building Burn, an open-source Deep Learning Framework in Rust 🦀
🔥 New video drop! Parts 2 and 3 of my deep dive into how Burn (a Rust 🦀 DL framework) fuses tensor operations into efficient GPU kernels ⚡ We move beyond scalar ops into the real machinery of fusion:
Burn is the #1 trending Rust repo on GitHub today and among the top 10 across all languages. Proud of the team and the community who contributed to the latest release!
That's also the point when AI will have the biggest positive impact on the world.
My bar for AGI is far simpler: an AI cooking a nice dinner at anyone’s house for any cuisine. The Physical Turing Test is very likely harder than the Nobel Prize. Moravec’s paradox will continue to haunt us, looming larger and darker, for the decade to come.
This is probably my favorite release since the first one of Burn. Huge milestones achieved, and we're just getting started! The release also comes with an updated website, don't hesitate to take a look at burn.dev
🔥 𝗕𝘂𝗿𝗻 𝟬.𝟭𝟴.𝟬: 𝗜𝗺𝗽𝗼𝗿𝘁𝗮𝗻𝘁 𝗣𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 𝗠𝗶𝗹𝗲𝘀𝘁𝗼𝗻𝗲𝘀 𝗔𝗰𝗵𝗶𝗲𝘃𝗲𝗱 1. State-of-the-Art Multi-Platform Matrix Multiplication Kernels 2. Dynamic Graph Flexibility with Static Graph Fusion Capability Full release note: github.com/tracel-ai/burn…
🔥 𝗕𝘂𝗿𝗻 𝟬.𝟭𝟴.𝟬: 𝗜𝗺𝗽𝗼𝗿𝘁𝗮𝗻𝘁 𝗣𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 𝗠𝗶𝗹𝗲𝘀𝘁𝗼𝗻𝗲𝘀 𝗔𝗰𝗵𝗶𝗲𝘃𝗲𝗱 1. State-of-the-Art Multi-Platform Matrix Multiplication Kernels 2. Dynamic Graph Flexibility with Static Graph Fusion Capability Full release note: github.com/tracel-ai/burn…
This is exactly why I don't like some people labelling Rust as system programming language when it plays role in many areas. Nice article x.com/matthiasendler…
Since a PyTorch-free 3DGS engine was posted here and seems to have hit a nerve, here's another one that's quite popular: github.com/ArthurBrussee/…
The work on fusion was very hard, but I'm glad I made it! The performance improvements speak for themselves, and it also paves the way for even more optimizations in the future. 🔥
1/ Burn 0.17.0 released: Up to 5x faster neural network training & inference with a new Metal compiler, expanded hardware support, and optimized kernels. Full Release Notes: burn.dev/blog/release-0…
Deep learning
Without Googling, name something Canada invented
Hey, great video, by the way! It's a very technical post, we could have gone into more detail on how to use tensor cores, which is central to the problem.
it was a great article! i just couldn't understand it well, felt like i was missing some great context here since i don't do GPU stuffs!
AI will become a national expense at the same level as energy, defense, roads, education, and healthcare
Every country and company should build the muscles to train their own models (based on open-source) or run the risk of being left out and dependent on others!
🚀 Burn 0.16.0 is out!🔥 - Even better GPU utilization with batched sync operations/transactions and simultaneous reads in CubeCL runtimes. - Performance gets another boost with mixed precision support for GEMM and convolution, alongside notable GEMM improvements.