Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer

Transfer learning, where a model is first pre-trained on a data-rich task being fine-tuned on a downstream task, has emerged as a powerful in natural language processing (NLP). The effectiveness of transfer has given rise to a diversity of approaches, methodology, and. In this paper, we explore the landscape of transfer learning for NLP by introducing a unified framework that converts all-based language problems into a text-to-text format. Our systematic study pre-training objectives, architectures, unlabeled data sets, transfer, and other factors on dozens of language understanding tasks. By the insights from our exploration with scale and our new ``Colossal Crawled Corpus'', we achieve state-of-the-art results on many benchmarks summarization, question answering, text classification, and more. To future work on transfer learning for NLP, we release our data set, -trained models, and code.