Hugging Face Transformers is an open-source Python library developed and maintained by Hugging Face. It began in 2018 as pytorch-pretrained-bert, a PyTorch port of BERT, and was renamed Transformers in 2019. It lets developers download the many pretrained models hosted on the Hugging Face Hub with a few lines of code and use them for inference and fine-tuning.
Auto classes such as AutoModel and AutoTokenizer load the appropriate architecture and tokenizer from a model name alone, and the pipeline API runs tasks such as text generation or classification in a single call. The tokenizer's chat template feature converts conversation messages into the special-token format a model expects. The library is built primarily on PyTorch and supports vision and speech models as well as natural language processing.