GigaGPT: GPT-3 sized models in 565 lines of code
GigaGPT is Cerebras’ implementation of Andrei Karpathy’s nanoGPT – the simplest and most compact code base to train and fine-tune GPT models.
Every blog post that took off on Hacker News, ranked by the votes of its best submission. 31,689 posts.
GigaGPT is Cerebras’ implementation of Andrei Karpathy’s nanoGPT – the simplest and most compact code base to train and fine-tune GPT models.
It all started about two or three years ago, I walked into the office one Monday morning around 10am... and I set the alarm off. My initial shock that I was the first person in on a Monday at 10am was quickly overridden by the realization that I had no idea…
We are excited to announce the preview release of a new, advanced code optimizer for the Visual C++ compiler backend. It provides many improvements for
QWEN CHAT DISCORD Introduction The evolution of multimodal large models is continually pushing the boundaries of what we believe technology can achieve. From the initial QwenVL to the latest Qwen2.5 VL, we have made progress in enhancing the model’s ability to understand image content. Today, we are excited to…
Postgres 19 brings plan advice: the query hints the Postgres community held off on for 15 years. Shaun Thomas covers the 15-year debate, pg_plan_advice, pg_stash_advice, and how to use them.
DuckDB can attach MySQL, Postgres, and SQLite databases in addition to databases stored in its own format. This allows data to be read into DuckDB and moved between these systems in a convenient manner.
Every now and again you come across something that stretches your brain in a new direction. Last week I stumbled on a magic trick of a Choose Your Own Adventure: Al Leonardi’s Ace of Aces, a paper …
Stay up to speed on the rapid advancement of AI technology and the benefits it offers to humanity.
We are asking a court to declare the Allow States and Victims to Fight Online Sex Trafficking Act of 2017 (“FOSTA”) unconstitutional and prevent it from being enforced. The law was written so poorly that it actually criminalizes a substantial amount of protected speech and, according to experts,...
March 19th, 2019 Dear Members of the European Parliament, The companies signing this letter to the European Parliament are urging you to vote against Articles 11 and 13 of the proposed copyright directive. The text of the trilogue agreement would harm the European economy and seriously undermine the ability of…
Founded in 1920, the NBER is a private, non-profit, non-partisan organization dedicated to conducting economic research and to disseminating research findings among academics, public policy makers, and business professionals.
Powered by generative AI, Alexa+ is your personal AI assistant that gets things done—she’s smarter, more conversational, more capable, and free with Prime.
GitHub will require all users who contribute code on GitHub.com to enable one or more forms of two-factor authentication (2FA) by the end of 2023.
Support for using Rust in the Linux Kernel was recently merged by Linus Torvalds. This is important because Rust is memory safe, meaning that code written in Rust does not suffer from things like buffer overflows, use-after-free, and other memory management vulnerabilities that plague software written in unsafe…
Why can't Meta A/B test its way back to greatness? To move Instagram beyond short-term engagement metrics, we ran a personalized human evaluation asking 100 users to compare TikTok vs. Instagram Reels. Learn why Gen Z considers Reels the place where TikToks go to die, and what Instagram should do about it.
Using a monorepo causes a lot of performance challenges for git. Here's how we solve them at Canva.
We launched Conductor a lifetime ago, in the distant past of July 2025. Writing most of your code with agents (let alone a team of coding agents) felt crazy back then. You trusted us to come along for the ride, and we cannot thank you enough!
After first getting hooked on Clojure in 2018 I decided to build my business, Wobaka.com, on 100% Clojure the following year. That's how much I enjoyed it. I thought I'd share some of my experience and technical stack for anyone else thinking about building and running their business on
As Machine Learning practitioners, when faced with a task, we usually select or train a model primarily based on how well it performs on that task. For example, say we're building a system to classify whether a movie review is positive or negative. We take 5 different models and see how well each performs this task:
Matrix, the open protocol for secure decentralised communications
Which countries and rivers emit the most plastic to the ocean? What does this mean for solutions to tackle plastic pollution?
Think of it as style transfer for dancing, a deep learning based algorithm that can convincingly show a real person mirroring the moves of their favorite dancers. The work, developed by a team of…
Today we’re releasing the Dart SDK 1.0, a cross-browser, open source toolkit for structured web applications. In the two years since we ...