CS2916: Large Language Model Course Repository

This repository contains the code for my course assignments related to the CS2916: Large Language Model course in Shanghai Jiao Tong University.

Contents

Note: HW2, HW3 & HW4, and the Additional Project were all completed in collaboration with my group members.

• HW1: A simple assignment, located in the ./hw1 directory.

• HW2:

  • Repository: mini-lima
  • Description: This project is a reproduction of the paper "Self-Instruct: Aligning Language Models with Self-Generated Instructions" using smaller parameter models Qwen1.5-7B, Qwen1.5-1.8B, and Qwen1.5-0.5B. We generated instruction data using Qwen1.5-7B and Qwen1.5-1.8B and performed instruction fine-tuning on Qwen1.5-0.5B, resulting in the model Qwen1.5-0.5B-SFT.

• HW3 & HW4:

  • Repository: Multimodal Learning
  • Description: This project explores the relationship between the performance of the multimodal model LLaVA on the Science QA dataset and the quality of alignment data, providing conjectures on how alignment affects model performance.

• Additional Project:

  • Repository: Self-Instruct CAAFE
  • Description: Self-Instruct CAAFE is an enhanced approach of CAAFE, combining its foundational principles with the innovative Self-Instruct method to achieve unprecedented automatic feature engineering performance. This project is related to a machine learning course focusing on large models.
  • Note: This project is my final project for a machine learning course, but it is also closely related to large language models.