Fast.ai

fast.ai exists to make deep learning usable by people who are not machine learning specialists, and it does that through software, courses and writing rather than a single application. The best known output is the fastai library, a deep learning library built on PyTorch that wraps common training patterns so that useful models can be trained in a small amount of code. Alongside it sit nbdev, a toolkit for developing Python libraries in notebooks, and fasttransform, a library that makes data transformations reversible and extensible through multiple dispatch. The teaching side is the other half. Practical Deep Learning for Coders is the long running course, joined more recently by How to Solve it With Code, and the material is also published as a book written around the library and PyTorch. The site itself runs as a technical blog covering model training behavior, tokenization, foundation models in areas such as computational pathology, and the social consequences of AI in education and work. The group is explicit that it wants people from unlikely backgrounds, uncommon languages, ordinary operating systems and small datasets involved, rather than only well resourced research teams. The courses are published free and the libraries are open source, and the site carries no pricing page, plans or subscription of any kind. fast.ai has since joined Answer.AI, the research outfit that also runs Solveit, a separate interactive development platform where a user builds solutions one or two lines at a time on a personal virtual server while the assistant sees the full working context; that platform is run and priced at its own site.