Introducing the LLM Inference Handbook: a single resource that explains everything you should know before you deploy an LLM, from how LLMs work to how to pick the right GPU. It explains topics such as tokenization, KV cache, quantization, speculative decoding, disaggregated inference, and much more.
https://t.co/LKqlKYVR21
@fernandezpablo De los ejercicios con peor relación riesgo/beneficio. Lo que podrías ganar en masa muscular no justifica las altas chances de joderte la espalda si no tenes la técnica perfecta
Behind every iPhone. His compiler.
Behind every Android. His compiler.
Behind every NVIDIA GPU. His compiler.
One American. Billions of devices. 🤯
Meet Chris Lattner 🇺🇸
> Started LLVM in late 2000 at UIUC as part of his graduate research.
> LLVM is a compiler infrastructure ~ the software that turns code into machine instructions.
> Apple hired him in 2005. He stayed 12 years.
> His toolchain now powers iPhone, iPad, Mac, PlayStation, Android NDK, and NVIDIA's CUDA.
> Also built Clang ~ the C/C++ compiler used by Google, Microsoft, and Sony.
> Built Swift in secret. Nights and weekends. While leading a 40+ person Apple team by day.
> Apple leadership was skeptical. He shipped it anyway. 🚀
> Swift now powers the vast majority of iOS apps on earth.
> Won the ACM Software System Award ~ same as Unix, Java, and TCP/IP.
> 2017 ~ Tesla VP of Autopilot. Worked in Elon's orbit. Left in 5 months.
> Joined Google Brain. Built MLIR ~ the compiler infrastructure behind TensorFlow.
> 2020 ~ joined SiFive to build open-source chips competing with Intel and ARM.
> 2022 ~ left Big Tech entirely. Founded Modular AI.
> Built Mojo ~ a new AI language that runs Python up to 35,000x faster.
> LLVM, Clang, Swift — all open-source. Mojo follows in 2026.
> Targeting NVIDIA's $4.8 trillion CUDA dominance. Raised $380M. Valued at $1.6B.
> Still writes code. Still answers GitHub issues himself.
He spent over 25 years building the compilers Big Tech is built on.
Now he's openly building the one that could break NVIDIA.
What a mind. Compiler GOAT. 🧠🐐
@abhi9u You might find this relevant: Druid meta-compiler, applies abstract interpretation over a bytecode interpreter to automatically derive JIT compiler templates. I’ve worked on it and was really fun
https://t.co/9NzBVDPmov