Mark Saroufim is a software engineer and writer focused on machine learning systems, PyTorch, reinforcement learning, and the intersection of computation, mathematics, and product thinking. His writing blends hands-on engineering with theory, exploring everything from deep learning infrastructure to philosophy of computation.
6 articles from this blog
A developer shares quick, one-line reviews of popular Linux distributions for desktop use, covering Ubuntu, Pop!_OS, Arch, and more.
An in-depth look at torchdata's internal architecture, focusing on datapipes and how they optimize data loading for PyTorch to improve GPU memory bandwidth.
Explores the practical uses of reversible computing, from automatic differentiation in deep learning to distributed systems and database operations.
A personal review of the Zachtronics book, exploring the developer's creative process and the intersection of programming games, curiosity, and multidisciplinary work.
A developer argues that adopting product management skills like communication and big-picture thinking makes developers more effective and influential.
Explores efficient state representations for robots to accelerate Reinforcement Learning training, comparing pixel-based and model-based approaches.
