%0 Generic %1 raffel2019exploring %A Raffel, Colin %A Shazeer, Noam %A Roberts, Adam %A Lee, Katherine %A Narang, Sharan %A Matena, Michael %A Zhou, Yanqi %A Li, Wei %A Liu, Peter J. %D 2019 %K large_language_models machine_learning natural_language_processing transformer %R 10.48550/arXiv.1910.10683 %T Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer %U http://arxiv.org/abs/1910.10683 %X Transfer learning, where a model is first pre-trained on a data-rich task before being fine-tuned on a downstream task, has emerged as a powerful technique in natural language processing (NLP). The effectiveness of transfer learning has given rise to a diversity of approaches, methodology, and practice. In this paper, we explore the landscape of transfer learning techniques for NLP by introducing a unified framework that converts all text-based language problems into a text-to-text format. Our systematic study compares pre-training objectives, architectures, unlabeled data sets, transfer approaches, and other factors on dozens of language understanding tasks. By combining the insights from our exploration with scale and our new ``Colossal Clean Crawled Corpus'', we achieve state-of-the-art results on many benchmarks covering summarization, question answering, text classification, and more. To facilitate future work on transfer learning for NLP, we release our data set, pre-trained models, and code.

@misc{raffel2019exploring, abstract = {Transfer learning, where a model is first pre-trained on a data-rich task before being fine-tuned on a downstream task, has emerged as a powerful technique in natural language processing (NLP). The effectiveness of transfer learning has given rise to a diversity of approaches, methodology, and practice. In this paper, we explore the landscape of transfer learning techniques for NLP by introducing a unified framework that converts all text-based language problems into a text-to-text format. Our systematic study compares pre-training objectives, architectures, unlabeled data sets, transfer approaches, and other factors on dozens of language understanding tasks. By combining the insights from our exploration with scale and our new ``Colossal Clean Crawled Corpus'', we achieve state-of-the-art results on many benchmarks covering summarization, question answering, text classification, and more. To facilitate future work on transfer learning for NLP, we release our data set, pre-trained models, and code.}, added-at = {2023-09-23T10:40:33.000+0200}, author = {Raffel, Colin and Shazeer, Noam and Roberts, Adam and Lee, Katherine and Narang, Sharan and Matena, Michael and Zhou, Yanqi and Li, Wei and Liu, Peter J.}, biburl = {https://www.bibsonomy.org/bibtex/2701ad0bb796e58c22d41791ea1979f57/meneteqel}, doi = {10.48550/arXiv.1910.10683}, interhash = {6b0b5e706bd467d86a80348cebfb1208}, intrahash = {701ad0bb796e58c22d41791ea1979f57}, keywords = {large_language_models machine_learning natural_language_processing transformer}, language = {en-US}, month = oct, note = {cite arxiv:1910.10683}, timestamp = {2023-09-23T10:40:33.000+0200}, title = {Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer}, url = {http://arxiv.org/abs/1910.10683}, year = 2019 }