README.md

Challenge:

https://pan.webis.de/semeval23/pan23-web/clickbait-challenge.html

Final Report

Presentation

preprocess.ipynb:

We preprocessed the train and validation datasets in this script.

openai-2shot.ipynb:

Requires an OpenAI API key. We prepared our gpt baseline here. We get predictions for validation data using 2-shot.

tf-idf.ipynb:

We prepeared our TF-IDF baseline here. Both the prediction and evaluation done in this notebook.

llama-lora.ipynb:

We finetuned a LLaMA-7B model using LoRA. We also save the predictions for validation dataset in a txt file

falcon.ipynb:

We finetuned a Falcon-7B model using QLoRA. We quantized into 4 bits. We also save the predictions for validation dataset.

roberta.ipynb:

We finetuned a RoBERTa and saved the validation outputs.

t5.py:

We finetuned a T5 model and save a checkpoint.

t5-eval.py:

We make predictions for the trained T5 model, which loads from the saved checkpoint, and save the results.

eval-scores.ipynb:

We calculate the Bleu and Bert scores in this script for all save validation outputs from all models.