We have used a pre-trained language model (PLM) to train a system that produces a layman’s summary given a research publication from the biomedical domain.
We have also participated in a related competition BioLaysumm, hosted by BioNLP Workshop, ACL 2024 (https://www.codabench.org/competitions/1920/). Our submission (userid- lkm1ml) was in the top 10 at the time of this submission to the competition.
-flant5-finetune.py - Training code to finetune a FLAN-T5 model
-flant5-inference.py - A code used to do inference with the model
-run_model.sh - Entry point for model training and inference.
-writeup.txt - A text file listing your approach, results and settings of experiments
-plos.txt and elife.txt - Outputs generated on the test data.
-link.txt - A link to the Google Drive folder containing the trained model
-screenshot.jpg - a screenshot of our submission to CodaBench
-prediction_result.zip - predicted results on the test data that is submitted to CodaBench
-requirements.txt - installation requirements to create a required environment. Other required things are assumed to be pre-installed or cached in the environment
This work was done collaboratively by Tanishq and Lalit during COL772, Natural Language Processing course (Spring 2024,Prof. Mausam, IIT Delhi).