@inproceedings{wang2023magneto,
author = {Wang, Hongyu and Ma, Shuming and Huang, Shaohan and Dong, Li and Wang, Wenhui and Peng, Zhiliang and Wu, Yu and Bajaj, Payal and Singhal, Saksham and Benhaim, Alon and Patra, Barun and Liu, Zhun and Chaudhary, Vishrav and Song, Xia and Wei, Furu},
title = {Magneto: A Foundation Transformer},
booktitle = {ICML 2023},
year = {2023},
month = {June},
abstract = {A big convergence of model architectures across language, vision, speech, and multimodal is emerging. However, under the same name "Transformers", the above areas use different implementations for better performance, e.g., Post-LayerNorm for BERT, and Pre-LayerNorm for GPT and vision Transformers. We call for the development of Foundation Transformer for true general-purpose modeling, which serves as a go-to architecture for various tasks and modalities with guaranteed training stability. In this work, we introduce a Transformer variant, named Magneto, to fulfill the goal. Specifically, we propose Sub-LayerNorm for good expressivity, and the initialization strategy theoretically derived from DeepNet for stable scaling up. Extensive experiments demonstrate its superior performance and better stability than the de facto Transformer variants designed for various applications, including language modeling (i.e., BERT, and GPT), machine translation, vision pretraining (i.e., BEiT), speech recognition, and multimodal pretraining (i.e., BEiT-3).},
url = {http://approjects.co.za/?big=en-us/research/publication/magneto-a-foundation-transformer/},
}