Skip to content

Latest commit

 

History

History
15 lines (12 loc) · 795 Bytes

README.md

File metadata and controls

15 lines (12 loc) · 795 Bytes

Adahessian in Fastai

This is an implementation of the AdaHessian optimizer in fastai. Adahessian is the first, practically useful, second-order optimizer to have been developed and has shown impressive performance, especially within NLP where it consistently out-performs AdamW, the field's current default optimizer.

You can find the author's implementation along with a guides about second-order methods here

Author's citation:

@article{yao2020adahessian,
  title={ADAHESSIAN: An Adaptive Second Order Optimizer for Machine Learning},
  author={Yao, Zhewei and Gholami, Amir and Shen, Sheng and Keutzer, Kurt and Mahoney, Michael W},
  journal={arXiv preprint arXiv:2006.00719},
  year={2020}
}