Building beautiful things like Mojo🔥 and MAX
@Modular
, lifting the world of production AI/ML software into a new phase of innovation. We’re hiring! 🚀🧠
It was a joy catching up with you Lex, you're one of my favorite inquisitive people because of your depth of knowledge in AI, programming, and willingness to really explore complex topics. 🔥+❤️=❤️🔥
Here's my conversation with Chris Lattner (
@clattner_llvm
), a legendary engineer, his 3rd time on the podcast. We talk about Modular AI and Mojo, a new programming language that is a superset of Python and can achieve 35,000x+ speed ups over Python.
Mojo🔥 is here! Many, many folks want to download and run the Mojo compiler locally instead in notebooks... but languages aren't just a compiler, you need a full ecosystem of developer tools to build things!
Mojo is still a young language, but today is a huge step forward! 🚀❤️🔥
Mojo🔥 is now available for download locally to your machine! ❤️🔥🚀
Beyond a compiler, the Mojo SDK includes a full set of developer and IDE tools 🛠 that make it easy to build and iterate on Mojo applications. Let’s build the future together!🔥
Mojo🔥 is no threat to Python🐍: it lifts it up and gives Python programmers superpowers🦸.
If anyone should be scared, it should be C++ and hard to use accelerator languages. Python is what developers love: C++ is mostly a pragmatic necessary evil for when you need performance.
Can Python survive this?
Modular, the company behind Mojo just raised $100 million dollars to fix AI infrastructure for developers.
That's a lot of money!
Mojo is a programming language for AI developers and is 35,000x faster than Python (Modular's benchmarks.) It combines the…
I am super excited to join
@SiFive
today, leading the software team that builds tooling to enable 'silicon at the speed of software'. Chip design is challenging, and needs open tools that are well designed, easy to use, and state of the art. More at:
I'm super excited the "Actors" proposal was accepted by swift-evolution last week. This is a major step forward for improving memory safety / design patterns for task based concurrency on CPUs. Swift is the first mainstream language to embark on this! 💪
Amjad: no joking, my son was telling me last week that we need to get Mojo into Replit so he can use it at his school. I'm pretty sure we can figure this out?
Chris Lattner of LLVM and Swift fame just announced a new programming language for ML that is high-performance and backwards compatible with Python (works with Python libraries). Could be a game changer.
Super excited to be part of the team that's bringing Machine Learning supercomputers to the world with Google Cloud TPUs. An immense number of scalable FLOPS can completely change productivity - and hopefully unlock entirely new avenues for ML research and production.
A big step forward for Mojo - it's not just open source, but open development and contribution model with a progressive Apache2 LLVM license. 🏆
We ❤️🔥 the Mojo community and are thrilled to continuously open the language and its development as little Mojo is growing up!
👩💻 We're excited to announce that we've open sourced the Mojo 🔥standard library! 📚
Building Mojo🔥 in the open will lead to a better result and open sourcing the standard library is our next step in the journey.
🚀 We're also dropping MAX 24.2 today!
After spending years working on AI/ML infrastructure, Modular AI is finally going to build it right. It is time for the best SW architects, engineers and product leaders to come together to lift the world’s ML compute. Learn more at
#modular_ai
#AI
#ML
My previous tweets about Rafael's email confused the matter. Here is a longer-form statement of facts about the situation, and a personal opinion about various aspects:
I am not a web designer.
Congrats to the Swift team for a big update. It's exciting to see a mainstream language bring distributed actors for IPC/RPC (on the path to memory safety, almost completing the 2017 Swift Concurrency Manifesto) and native /regex/'s integrated with pattern matching! 💯🏎
This is a profound and inspiring vision for Swift 6: multi platform investment, concurrency, move semantics, rust-style ownership, Automatic Differentiation and investment in developer joy 💖🚀
So true. Too many apparently "simple" techs merely shift the complexity to other places (higher level tools, frameworks, pkg managers, wrappers, syntax extensions, etc). Well designed systems are simple to learn and use end-to-end, while permitting experts to build amazing things
Mojo🔥 0.4 just dropped today with a MASSIVE number of updates across language, library and tools built in the last two weeks. 🥰 Mac support didn't quite make the train but is looking promising for release soon! 🧑💻
Sigh internet🧐. Clarifying: all of the projects I was involved in & driving at Google (incl. S4TF, MLIR, an incredible new TF runtime, XLA, TPU SW, etc) are in very good hands 💪🙌, have exec commitment and strong roadmaps. I c/wouldn't leave if uncertain about that!
@JeffDean
Rust experts say: "If Mojo is true, then I think Mojo will win just hands down. The reason Mojo will win, you don’t change the paradigm of already acclimated or proficient individual. Learn a bit more, and get amazing performance”
Good news: Mojo is true! Try it yourself today!
Don't miss
@ThePrimeagen
video covering our epic community blog 💪 comparing Mojo 🔥 and Rust ⚙️. Mojo 🔥 aims to be a superset of Python 🐍 over time, and we remain committed to our incredible community & open-source roadmap. Join Us! 👇🏼⬇️
We're super excited to provide an update on Swift for TensorFlow - things are getting real! Cloud TPUs are also making ML super computers available to everyone, come find out more at the
#TFDevSummit
!
Get ready to learn about the latest developments in:
@TensorFlowJS
Swift for
@TensorFlow
TensorFlow Lite
TPU Research Cloud
TensorFlow 2.0
TensorFlow Hub
TensorFlow Extended (TFX)
...and more!
Register for the
#TFDevSummit
today →
Very excited we get to talk about some of the work of the Google MLIR compiler team in a couple weeks. This new infra is a pretty profound rethink of compiler architecture and design, applicable to many domains. Abstract at the bottom of:
Only Lex could take the discussion through a whole lot of compiler geekery (llvm, gcc, clang, mlir), some of the internal challenges getting swift to market, my opinion of Elon’s secret to Tesla working, up to Google’s incredible lead in AI accelerators: hw/sw co-design on TPUs👍
Here's my conversation with Chris Lattner (
@clattner_llvm
) on the Artificial Intelligence podcast. He is one of the top experts in the world on compilers and powerful software+ML+hardware systems in general, w/ exciting projects at Google, Tesla, & Apple.
Epic work, the thing I love about this is how small and clean the code is - literally reimplementing everything down to the metal instead of depending on thick layers of magic.
The out-of-the-box features of
@Modular_AI
's Mojo are just incredible. We applied unrolling and now llama2.🔥 outperforms
@ggerganov
's llama.cpp by almost 20% in CPU inference speed.
The Mojo internals technical talk is now available👇. Dig in to learn more about how Mojo's compiler architecture is a step forward from other languages, about our quest to lift performance engineering out of compilers and into libraries, and Mojo for heterogenous compute. 🔥🔥
Our 2nd keynote, "Mojo 🔥: A system programming language for heterogenous computing", from the 2023 LLVM Developers' Meeting is now available:
#Mojo
#LLVMDevMtg
It is /such/ an exciting time in AI - where research is opening new capabilities on a daily basis!
Modular is driving complexity out of the system to allow more people to participate, enable new kinds of full-stack innovation, and to get that research to production. 🔥🚀
We are so excited to announce we have raised $100M to fix AI infrastructure for the world's developers, led by
@generalcatalyst
and filled by
@GVteam
,
@svangel
,
@GreylockVC
and Factory. We are humbled and so thankful to our incredible customers, developers & world-class team! 🚀
I have no idea how this is physically possible, but the amazing TensorFlow team has the session videos already available on YouTube! Check out the Swift for
@TensorFlow
video at
#TFDevSummit
2019 here:
IMO, the most important and transferable (to other domains) part of compilers is: the design, representation, validation and transformation of structured data. The mentality and designcenter crosscuts all of computing, from the simplest json payload to the most fiddly compiler IR
Starting to put together the content for my 2nd monthly newsletter - going out in one week! 🚀
Just like last month, it'll be packed full of Swift content, and I'm also going to throw in more tips for time management & productivity! 😀
Sign up here 📬
I often write code in the evenings so I can spend most of the day working with people. Rationale: “10x” leaders realize that the best products are built by diverse and collaborative teams, and great teams take a lot of work to get right. Invest in building up humans & code!
We've worked VERY hard to make sure our performance claims are fair, accurate, and have no "tricks". This is raw drag-racing against the mature systems that the industry already uses... and we're just getting started. o/c We support all the tricks as well, they compose on top!
You should checkout the world’s fastest unified AI inference engine, and its performance, on . Unparalleled performance across frameworks, models and hardware 🔥
Slides from our unveil of the new MLIR compiler infra are now up at bottom of . The team is very excited to get this out & there is a lot more to talk about later. We are focused on open sourcing in the next couple months! 🎈🎉
Jeremy pulls back the covers on what makes Mojo tick, why, and why that matters. I also love that he shares the backstory on building the demo and working with the Modular crew. Among his other virtues, he is a master story teller!
There's a new programming language in town - it's Mojo! I'm more than a little excited about it. It's Python, but with none of Python's problems.
You can write code as fast as C, and deploy small standalone applications like C.
My post is below, and a 🧵
We got tons of disbelief about how Mojo could be faster in practice than languages like Rust or C++ or Swift. This article lifts the hood a bit to peak at how this works from a technical perspective. 🏎⚙️
Inspired by
@ThePrimeagen
's epic video discussing Mojo 🔥 and Rust 🦀, and fueled by an electrifying community discussion ⚡️ - we have a new post up, and you wont want to miss this one! 😱 ⬇️
Mojo vs. Rust: is Mojo 🔥 faster than Rust 🦀 ? 🤔 ❤️🔥
The first crack at llama2.🔥 is here 🚀
A Mojo 🔥 community member - Mojician - did a simple port from Python to Mojo, and shows its already 20% faster than Karpathys llama.c implementation 😱 How much faster can it go? 📈
I'm looking forward to sharing some of my experiences building Open Source tools and communities (e.g. LLVM, Clang, Swift, Mojo and others) at the Accel | Open Source meeting next Wednesday in SF:
Join a great speaker lineup + amazing community! 🔥
1 week minireview of the Tesla Model 3: the hardware is truly great (a big step up from my Model S) but the software is unfinished and buggy.
I’m also sad how little progress HW2 Autopilot has made since I last drove it in June...
Tuesday is going to be wild. If you care about AI, developer tools, or software engineering, there is a good chance it will 🤯 your mind! Join us 9am pacific at
Life update: I started a new job at
@Modular_AI
! I was just about to accept an offer from Apple after leaving Google but after talking to
@clattner_llvm
and realizing how they planned to disrupt AI I knew I had to join. The public unveiling is Tuesday!
Building a programming language from scratch is a very long game. It’s the sort of thing best done when you walk into it knowing the challenges and understand what a huge effort it entails… this only makes sense when moving SoTA is worth it (to the world as a whole). 🔥
Apple has been working so hard on Swift, and I love their persistence in making it perfect. They never seem to do something hacky/short-term.
I think that this is exactly why Swift takes so long to be adopted. But I wouldn't have it another way.
Special shout out to
@mingshenghong
who figured out how to get Swift + TensorFlow to work (he built the Swift runtime for TF), and
@rxwei
+ Dan Zheng who built the Tensor API and the autodiff implementation. Such a great team!
The Google Brain team is hiring for a wide range of positions spanning compiler engineering, runtime and systems work, ML modeling and infrastructure, API design, Swift compiler engineering, data perf optimization, and more
See this page for more details!
The video of the unveil of the Google MLIR compiler infrastructure is now available. It was a lot of fun getting to present with Tatiana, who is the tech lead and manager of the project. The talk showcases the work of many amazing engineers, thank you! 🎉
Just to clarify, it is 10 years since the first lines of code written in secret. The public launch was in June 2014 at WWDC, and the open source release in December 2015. Time flies. Such a great team. So much left to do!
#linuxgate
is just 🤮🤮. I get that the LKML CoC was (apparently) rushed, but it is still a great step forward as a statement of intent. When did the principle of treating people with respect become so controversial? I'm so happy to work in progressive and positive communities!
@anatomisation
No, I mean it. I can disagree with someone and respect their beliefs at the same time. Intolerance of differing opinions is not good for anyone.
I am definitely sad to lose Rafael from the LLVM project, but it is critical to the long term health of the project that we preserve an inclusive community. I applaud Rafael for standing by his personal principles, this must have been a hard decision.