Sylvain Gugger

Sylvain Gugger is a Machine Learning Engineer at Jane Street on the ML-infra team, where he helps traders and researchers accelerate their model training and inference. Before Jane Street, he spent three years at Hugging Face (2020-2023) as a core maintainer of the Transformers library (1,250+ commits) and creator of the Accelerate library (290+ commits, 9.5k+ GitHub stars), which simplifies distributed PyTorch training across GPUs, TPUs, and mixed-precision setups with minimal code changes. Prior to Hugging Face, he worked at fast.ai (2017-2019) alongside Jeremy Howard, co-authoring "Deep Learning for Coders with fastai and PyTorch" (O'Reilly, 2020) and helping build the fastai library and courses. Gugger began his career as a mathematics and computer science teacher in France, spending seven years at CPGE (Classes Preparatoires aux Grandes Ecoles), where he authored 10 math textbooks published by Dunod. He discovered machine learning through Jeremy Howard's fast.ai MOOC after relocating to New York City around 2015. A key moment in his early ML career was the Stanford DAWNBench competition, where his implementation of Leslie Smith's super-convergence method helped the fast.ai team achieve first place on CIFAR-10 (fastest and cheapest training). His widely read blog posts on the 1-cycle learning rate policy and learning rate finders have become foundational references in the deep learning community. He holds a Masters in Mathematics and Computer Science from Ecole normale superieure (2003-2007).