The blog post provides an overview and personal commentary on a book published by Google DeepMind about scaling large language models (LLMs) using mathematical modeling. It discusses the importance of understanding performance met...
About:
Eric writes a Substack about building software outside of work, focusing on systems/pl, interaction design, and open-source.
Website:
Specializations:
Interests:
Outgoing Links:
Subscribe to RSS:
The author discusses the development of a deep learning compiler in pure JavaScript, inspired by JAX and its use of XLA for optimization. The post outlines the challenges faced in implementing numerical computing and GPU compute i...
jax-js is a machine learning library for the web that enables efficient numerical programming and model training directly in browsers using WebGPU and WebAssembly.
The author describes their weekend project of creating a hypervisor that spins up virtual machines (VMs) upon SSH login. The project utilizes Firecracker microVMs and aims to provide a whimsical user experience. The author reflect...
5.8jax-js devlog feb 17
2025-02-18 • programming software development github jax tinygrad
The author reflects on their progress in a side project related to software development, particularly focusing on GitHub activity and the understanding of compiling operations into kernels. They discuss insights gained from a visi...
The author shares their experiences working on software development tasks at the Val Town office, discussing the implementation of various functions related to Jacobian-vector products and vectorization in programming. They expres...
The post discusses the concept of a 'tutorial mindset' for teaching array and functional programming using JAX and NumPy in a browser environment. It outlines the steps to create a code editor/REPL using Vite, enabling users to ru...