Simple Transformers
A Python library, built on Hugging Face Transformers, that trains and evaluates a range of NLP models in as little as three lines of code, covering classification, question answering, retrieval and generation tasks.
Simple Transformers wraps the Hugging Face Transformers library so a model can be initialized, trained and evaluated in three lines of code. Supported tasks include dense information retrieval, large-language-model training and fine-tuning, encoder model training, sequence and token classification such as named-entity recognition, question answering, language generation, T5 and other sequence-to-sequence tasks, multimodal classification, and conversational AI.
Every task-specific model follows the same pattern: initialize it, train it with a training call, evaluate it, and generate predictions on new data, with task-specific configuration options documented separately for each model type. It integrates with Weights & Biases for optional experiment tracking and visualization during training.