To setup the training and testing environment, first clone the repository:
git clone https://github.com/AlexWan0/cs182_final_project.git
Then, create the conda environment:
cd cs182_final_project/
conda env create -f environment.yml
Note that CUDA is required to run these experiments.
To run with the full hyperparameter sweep (i.e., epochs, LR, and datasets), simply run:
python model.py
Due to resource limitations, a "debug" version is enabled by default.
To run the entire set of psychometric tests, simply run:
python eval.py
The full set of 160 checkpoints is unfortunately too cost-prohibitive to distribute (and impractical for reviewers/graders to run). We instead distribute a subset of these checkpoints. We additionally provide a Colab notebook to run these checkpoints.
The checkpoints can be found here. The Colab notebook can be found here (make sure the GPU runtime is enabled).
The plots can be reproduced using the plots.ipynb
and lossplots.ipynb
notebooks.