GPT-1 was OpenAI’s first publicly released Generative Pre-trained Transformer language model, introduced in 2018.
OpenAI presented the model in its paper “Improving Language Understanding by Generative Pre-Training.” GPT-1 used a Transformer architecture and was first pretrained on a large collection of unlabeled text before being fine-tuned for individual language tasks.
The central idea was to learn broad language patterns during pretraining and then adapt the same model to tasks such as classification, entailment, and question answering. This approach helped establish the modern pretraining-and-fine-tuning paradigm for language models.
GPT-1 is often overshadowed by GPT-2 and GPT-3, which were much larger and more widely discussed. It had 117 million parameters, a small figure by later standards, but its method was influential. The name GPT means Generative Pre-trained Transformer, not “general-purpose translator.”