Packages

All the packages used are added into the requirements.txt file.

Task-1 (Fine Tuning LLM)

Task 1 is completed here.

Task-2 (Computing Influence Scores for all the Training Data Points)

Task 2 is completed in here

Task-3 (Implementing Parameter Efficient Fine-Tuning and Data Selection Methods)

To apply the ts-shapley data selection method - TS-DShapley

We have implemented three parameter efficient fine tuning methods - BitFit, LoRA, IA3. The following notebooks contain implementation and results for all the fine tuning methods for different data selection approaches -

• With No Data Selection
• Top 100 Influential Points Selected
• All Positive Influential Points Selected
• TS-DShapley Selected Points

Report

An overview of the implemented tasks and the results achieved can be found in - report and slide