rohanphanse / cpsc477-final Goto Github PK

By: Rohan Phanse and Joonhee Park

First, install Jupyter Lab by running pip install jupyterlab in the terminal and start the server with the jupyter-lab command. Next, run the cells in data.ipynb in order to generate the augmented dataset. Then, run the train.ipynb notebook on a GPU - we used the A100 through Google Colab. Load the CSV files for each split of the augmented dataset into the environment: train.csv, val.csv, test.csv in the dataset directory. The train notebook will finetune the Mistral model, perform inference on the test dataset, and store the generated summaries in a directory. Finally, add that directory of summaries to the inference directory and run eval.ipynb to obtain the ROUGE scores and evaluation metrics of those summaries.

See report.pdf for details about project motivations, related work, approach, and results.