STA5007
Fine-tuning Gemma2 on Chinese GPT-4 QA Dataset
Overview
- Final Project – Advanced Natural Language Processing (grade: 97)
- Fine-tuned the Gemma2 model for Chinese QA tasks using the Evol-Instruct-Chinese-GPT4 dataset.
- Leveraged LoRA for parameter-efficient adaptation and RAG for enhancing contextual understanding.
- Delivered a comprehensive analysis of the model’s performance, highlighting key improvement areas and proposing future optimizations for better generalization in unseen QA tasks.
- Report URL: Fine-tuning Gemma2 on Chinese GPT-4 QA Dataset (Kaggle)