مقاله و کد مدل زبانی GPT-3 منتشر شد!

GPT-3 is an autoregressive language model and trained with 175 billion parameters, 10x more than any previous non-sparse language model before.

Language Models are Few-Shot Learners: https://arxiv.org/abs/2005.14165

GitHub: https://github.com/openai/gpt-3

GPT-3 achieves strong performance on many NLP datasets, including translation, question-answering, and cloze tasks as well as several tasks that require on-the-fly reasoning or domain adaptation, such as unscrambling words, using a novel word in a sentence, or performing 3-digit arithmetics.

#nlp #gpt3