The Order Matters: Sequential Fine-Tuning of LLaMA for Coherent Automated Essay Scoring

The paper presents a study on fine-tuning the LLaMA-3.1-8B model for Automated Essay Scoring (AES) using parameter-efficient LoRA with 4-bit quantization. It compares three training curricula—Sequential, Independent, and Randomized—finding that Sequential fine-tuning significantly outperforms the others, achieving F1 scores of 65% and 87% for evidence and conclusion, respectively, and surpassing the LLaMA-70B baseline despite its smaller size. This research highlights the importance of curriculum design aligned with discourse structure in enhancing AES performance and suggests that smaller, optimized models can effectively compete with larger LLMs, providing a scalable approach for educational applications.