Mark Saroufim

Over the past few weeks I’ve spent a lot of time reviving some of my older machines, installing many different distros and tweaking them and I wanted to share my 1-line reviews in case they help someone else looking to take the plunge into Linux for deskt…

If you’re excited about optimizing code that runs equally well on a single or thousands of GPUs and if you have the ability to submit a single substantial PR to a major OSS library, we want you on the PyTorch team - especially if you’re early in your care…

Why won’t Llama 13B fit on my 4090 I recently gave a talk that a bunch of people liked, maybe I’ll write a proper blog post at some point but in the meantime, if you care to learn about the different kinds of memory overhead in Neural Networks and how to…

A batch’s journey: torchdata internals The core code is in torchdata.

Reinforcement Learning is a Game Design Problem 2 pager describing how I’d like the field to move towards environments that are differentiable, multi-agent, compositional, multi-modal , continuous and self supervised RL is Game Design

Applications of Reversible Computing I first came across reversible computing back when I was helping organize a quantum computing seminar back in grad school.

A personal review of Zach-like the book This was the first book I ever bought on Twitter.

Why developers should also be product managers This post is summarizing some observations I’ve made working at Microsoft as a Program Manager, Applied Scientist and Developer.

Work in Progress Representation Theory for Robotics This goal of this post is to teach you how to efficiently represent the state of a robot with as little memory as possible.

Controlling a 2D robot arm with deep reinforcement learning and Floydhub My new post is on the Floydhub blog.