The code is modified from this repo. Our adahessian optimizer code is here.

Please first install TensorFlow 2.0 and other dependencies:

conda env create -f environment.yml

Then, please activate your conda environment:

conda activate adahessian_tf

After all the previous steps, let's train the model!

export CUDA_VISIBLE_DEVICES=0; python run_experiments.py

Note that you can train the model with SGD/Adam by modifying the experiments.yaml file.

We also include the training log file in logs for resnet20 for your reference.