Hugging Face Transformers is an open-source deep learning library developed by Hugging Face. It provides implementations of pretrained transformer and other neural network models for training and inference, including models for text, images, audio, video, and multimodal learning.
History
Transformers was first released in November 2018 as pytorch-pretrained-bert, a PyTorch implementation of BERT. It was later renamed pytorch-transformers as support expanded to additional transformer models, before becoming Transformers.
By September 2019, its PyTorch version had been pip installed more than 500,000 times that year and was being used by more than 1,000 companies, according to VentureBeat. Version 5, released in 2025, made PyTorch the library's sole machine learning backend, ending direct support for TensorFlow and JAX.
Features
Transformers provides standardized model definitions and tools for downloading, running, and training pretrained machine learning models. It is used for tasks including text generation, machine translation, text classification, question answering, image classification, object detection, speech synthesis, and speech recognition. Models can be downloaded from and shared through the Hugging Face Hub.
