The repository houses GPT-3, a large autoregressive language model trained on diverse web text. It demonstrates that scaling model size dramatically improves few-shot performance across translation, question-answering, cloze tasks, and reasoning challenges without fine-tuning. The release includes model samples and documentation for the seminal 2020 paper.