Paper

PaLM 2 Technical Report

Clark, Jonathan H, Robinson, Kevin, Tokumine, Simon, Shachi Dave, Lin, Hanzhao, Catasta, Michele, Dehghani, Mostafa, Yu, Jiahui, Botha, Jan, Zheng, Ce, Dai, Andrew M, Wu, Yonghui, Gustavo Hernandez Abrego, Chowdhery, Aakanksha, Hui, Jeffrey, Liu, Zhongtao, Johnson, Melvin, Nham, John, Slone, Ambrose, Zhang, Qiao, Meier-Hellstern, Kathy, Brahma, Siddhartha, Crepy, Clément, Rajkumar, Samuel, Laurent El Shafey, Qiao, Siyuan, Chu, Eric, Vodrahalli, Kiran, Xue, Linting, Zhang, Yujing, Devlin, Jacob, Nado, Zachary, Austin, Jacob, Richter, Bryan, Cheng, Yong, Xiao, Kefan, Zhou, Weikang, Wieting, John, Garcia, Xavier, Li, Jian, Hashemi, Hadi, Zheng, Steven, Ruder, Sebastian, Moussalem, Maysam, Gur-Ari, Guy, Krikun, Maxim, Li, YaGuang, Lee, Benjamin, Jia, Wenhao, Ahn, Junwhan, Bradbury, James, Freitag, Markus, Choquette-Choo, Christopher A, Moreira, Erica, Chen, Zhifeng, Maynez, Joshua, Reif, Emily, Wang, Zirui, Isard, Michael, Lee, Katherine, Xu, Kelvin, Sunipa Dev, Hand, Steven, Hu, Andrea, Shelby, Renee, Brennan Saeta, Wang, Tao, Aurko Roy, Parrish, Alicia, Barham, Paul, Brooks, Kevin, Pellat, Marie, Polozov, Alex, Hou, Le, Feinberg, Vl

2023

Overview

We introduce PaLM 2, a new state-of-the-art language model that has better multilingual and reasoning capabilities and is more compute-efficient than its predecessor PaLM. PaLM 2 is a Transformer-based model trained using a mixture of objectives. Through extensive evaluations on English and multilingual language, and reasoning tasks, we demonstrate that PaLM 2 has significantly improved quality on downstream tasks across different model sizes, while simultaneously exhibiting faster and more efficient inference compared to PaLM. This improved efficiency enables broader deployment while also allowing the model to respond faster, for a more natural pace of interaction. PaLM 2 demonstrates robust reasoning capabilities exemplified by large improvements over PaLM on BIG-Bench and other reasoning tasks. PaLM 2 exhibits stable performance on a suite of responsible AI evaluations, and enables inference-time control over toxicity without additional overhead or impact on other capabilities. Overall, PaLM 2 achieves state-of-the-art performance across a diverse set of tasks and capabilities. When discussing the PaLM 2 family, it is important to distinguish between pre-trained models (of various sizes), fine-tuned variants of these models, and the user-facing products that use these models. In particular, user-facing products typically include additional pre- and post-processing steps. Additionally, the underlying models may evolve over time. Therefore, one should not expect the performance of user-facing products to exactly match the results reported in this report.

• Publisher: Cornell University Library, arXiv.org
• Subject: Inference / Multilingualism / Reasoning / State of the art / Toxicity