Victor Guerra

@marksaroufim @vitransformer Witnessing what you did with the CUD…, 😮‍💨 sorry, GPU mode community was great, wishing you all the best on what comes next!

After 5 amazing years, I’m leaving the PyTorch team at Meta. I did my best work there and got to work with some of the smartest, most OSS pilled engineers in the industry. More soon on what’s next: still systems, still OSS (but not everything), a smaller team with a lot of GPUs

@zaachsiieegeel Awesome 🚀 We are rapidly improving: github.com/modular/skills and you'll see a lot more here in the next few weeks! Contributions welcome 🤗

What’s the best model you can train in a day if someone hands you a pile of Blackwell GPUs? You can try out yourself On April 9 in Paris, @GPU_MODE + @verdacloud + @sestercegroup are hosting a GPU hackathon with a bunch of GPUs to run on and even more of them for the winners.

Mojo 🔥 just added a powerful new reflection capability that can synthesize trait conformances for you. In other languages, this is only available to compiler authors. Maxim Zaks breaks down how Mojo makes metaprogramming explicit and accessible: mzaks.medium.com/when-magic-bec…

Mojo in Jupyter is here 🙌 @jeremyphoward released a new Jupyter kernel that lets you run Mojo directly in notebooks. It works great on macOS, supports recent Linux versions, and is easy to install via pip or uv. Give it a try and let us know what you build! github.com/AnswerDotAI/mo… #MojoLang #OpenSource #DeveloperTools

Released pyncd/tsncd today. We now have a way to manage deep learning algorithms algebraically! The code on the left renders into the diagram on the right. Mismatched domain/codomains are automatically managed under the hood.

The journey that started a long time ago as sparse compilation, enriched with new ideas by so many others along the way, continues with the introduction of the Universal Sparse Tensor (UST) type. developer.nvidia.com/blog/establish…

If you work with MLIR and like programming with agents, try the Context7 MCP. LLMs have historically been abysmal at MLIR (probably due to lack of training data), but Context7 helps the agent discover APIs and use them correctly. The same goes for any relatively niche library.

Aw snap @UnslothAI and @danielhanchen ! @Modular is coming for you! modular.com/blog/how-i-bea…

Oh, you're writing CUDA kernels? Everyone's on Triton now. Just kidding, we're all on Mojo. We're using cuTile. We're using ROCm. We have an in-house DSL compiler targeting the NVGPU MLIR dialect but wait, Tile IR just dropped so we're going to target that instead. Our PM is on TileLang. The team lead was on CuTe but now she's back to handwriting PTX. If you're not on Pallas, you're ngmi. Our intern is building on TT-Metalium for our Wormholes. Our CFO approved an order for some big chungus wafer-scale chips so now we're porting our kernels to CSL. Our CTO is working on a kernel-less graph compiler so we won't need to write kernels anymore. Our CEO thinks we're talking about the Linux kernel. We're building Claude for dogs.

@JeffDean Bookmarked! Another good resource is Algorithmica en.algorithmica.org/hpc/

@krismicinski Not sure how much you've tinkered with MLIR, but if you're 100% new, @jeremyjkun's blog post series is great. The GitHub repo is here with links to each of the blog posts: github.com/j2kun/mlir-tut…

From an idea to powering world-leading AI models and cutting-edge accelerators — Mojo has come a long way. Now, Mojo 1.0 is on the horizon: stability, open-source plans, and new tooling for developers everywhere. Read our latest blog post to explore the road to 1.0 ⬇️

@TheZachMueller @huggingface @GuggerSylvain @StasBekman @mervenoyann congrats Zach on such a good run at @huggingface and thank you for all your contributions to the ecosystem. Good luck on what's coming next for you!

After almost 4 years at @huggingface, it's time to move on. When I joined, I had only ever trained on Google Colab and was shellshocked and in awe at the master-class of people I'd be working with, @GuggerSylvain, @StasBekman, @mervenoyann, and so many more. 4 years later and accelerate is now the backbone of the entire Hugging Face ecosystem, and it's been a gateway to allow for anyone to do distributed training. I couldn't be more proud of my time, and my team, as we tackled the fun problems of raising up the entire community into the modern age of distributed training. I won't say what's next for me yet, but I won't be going too far in the world of AI, and especially the wonderful communities I've been able to be a part of over the nearly decade of being in OSS.

We know that one of the biggest barriers to programming GPUs is access to hardware: "Code you’ve written for NVIDIA or AMD GPUs should now mostly just work on an Apple🍎 Silicon GPU, assuming no device-specific features were being used." Preview here:👇 forum.modular.com/t/apple-silico…

In order to celebrate the release of the print version for the Ultra-Scale Playbook (of which I have no affiliation with and love deeply), I'm going to be giving away 5 copies! To enter, simply like + retweet this tweet. Winners will be selected at random 10AM EST on the 13th

It was fun to nerd out with Richard of @sw_unscripted about programming languages in the modern era. At timestamp 46:01 I answer a FAQ: Why care about programming languages in the age of AI coding tools? Let me know what you think!

Check out Modular's GPU Programming comics - such a fun way to portrait a complex topic. comic.modular.com @Modular #gpucomic

The Future of Compute Portability x.com/i/broadcasts/1…

Learn hands-on GPU programming in Mojo🔥 by solving puzzles🧩! Start here: builds.modular.com/puzzles