Project Explainer: Summarization using Large Language Models

Large Language Models are picking pace very quickly and they are turning out to be extremely good in multiple tasks. With the help of zero-shot, few-shot, and fine tuning techniques we could effectively specialize a language model for the use case. Summarization is one such use case that has been widely researched for a couple of years now. Broadly there are techniques such as Abstractive and Extractive approaches. The motive of this project proposal is to handle the summarization task (mostly Abstractive + Extractive hybrid approach) through the language model’s (foundation model) lens. This project aims to cover everything from data collection, EDA, experimenting with different language models to developing production-scale system that can take GitHub repo as reference and provide summary. One of the challenges that is novel is to use smaller sized models to achieve great performance in summarization. SCoRe Lab has been into developing solutions in the space of making user life easier with products such as D4D, Bassa, Track Pal, and others. This project will add to that portfolio and would be a great reference for AI practitioners and system developers which aims to work right from data to production-grade end product using AI and Systems.