In this work, we demonstrate that multilingual large-scale sequence-to-sequence (seq2seq) models, pre-trained on a mixture of denoising and Causal Language Modeling (CLM) tasks, are more efficient few-shot learners than decoder-only models on various tasks. In particular, we train a 20 billion parameter multilingual seq2seq model called Alexa Teacher Model (AlexaTM 20B) and show that it achieves state-of-the-art (SOTA) performance on 1-shot summarization tasks, outperforming a much larger 540B PaLM decoder model. AlexaTM 20B also achieves SOTA in 1-shot machine translation, especially for low-resource languages, across almost all language pairs supported by the model (Arabic, English, French, German, Hindi, Italian, Japanese, Marathi, Portuguese, Spanish, Tamil, and Telugu) on Flores-101 dataset. We also show in zero-shot setting, AlexaTM 20B outperforms GPT3 (175B) on SuperGLUE and SQuADv2 datasets and provides SOTA performance on multilingual tasks such as XNLI, XCOPA, Paws-X, and XWinograd. Overall, our results present a compelling case for seq2seq models as a powerful alternative to decoder-only models for Large-scale Language Model (LLM) training.
@article{arxiv.2208.01448,
title = {AlexaTM 20B: Few-Shot Learning Using a Large-Scale Multilingual Seq2Seq Model},
author = {Saleh Soltan and Shankar Ananthakrishnan and Jack FitzGerald and Rahul Gupta and Wael Hamza and Haidar Khan and Charith Peris and Stephen Rawls and Andy Rosenbaum and Anna Rumshisky and Chandana Satya Prakash and Mukund Sridhar and Fabian Triefenbach and Apurv Verma and Gokhan Tur and Prem Natarajan},
journal= {arXiv preprint arXiv:2208.01448},
year = {2022}
}