In October 2019, the team at Hugging Face documented their open-source Transformers library in the paper “HuggingFace’s Transformers: State-of-the-art Natural Language Processing,” led by Thomas Wolf with a large group of co-authors including Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, and Alexander M. Rush. The paper describes a library that provides “state-of-the-art Transformer architectures under a unified API” together with “a curated collection of pretrained models made by and available for the community.”
The practical breakthrough was access. Before Transformers, using a published model meant finding the authors’ research code, matching its dependencies, and porting it to your setup - work measured in days. The library collapsed that to a single install and a few lines of code: load a pretrained model by name and run it. Per the official documentation, there are now over one million model checkpoints usable through the library on the Hugging Face Hub.
Transformers became the distribution layer of the large language model era. It standardized how models are defined and shared, so a model published in the format spreads across the ecosystem of training frameworks and inference engines that build on it. The library and the company behind it have an existing AI Library entry (see hugging-face), which this milestone links to rather than duplicates.