Implementation of the Eve optimizer and experiments from Adaptive Optimization with Examplewise Gradients.
If you find this code useful, please reference in your paper:
@article{kunze2021eve,
title={Adaptive Optimization with Examplewise Gradients},
author={Kunze, Julius and Townsend, Jamie and Barber, David},
journal={arXiv preprint arXiv:2112.00174},
year={2021}
}
Install dependencies for experiments of choice:
pip install -r <cifar10|seq2seq|wmt|ppo|pixelcnn>/requirements.txt
See README.mds in subfolders to run experiments.
Default configs are as used in the paper.
Results are logged to wandb.