The code revolves around utilizing the conventional libraries from HuggingFace to build a customized chatbot that is only inferior to GPT 3 in terms of limited computational resources. Further improvement requires adapting the optimal rank of the decomposed matrices and targeting modules in the Attention layers of the pre-trained model to be augmented with LoRA. The intent is to focus more on fine-tuning approaches and less on prompt engineering techniques that could also boost performance.
Scatter-plot of family relationships in GOT https://www.kaggle.com/datasets/khulasasndh/game-of-thrones-books