Fine-tuning for Better Few Shot Prompting: An Empirical Comparison for Short Answer Grading

2025-08-05

AI Summary

This study investigates how fine-tuning enhances few-shot prompting performance in Automated Short Answer Grading (ASAG), particularly under real-world scarcity of labeled data. Methodologically, we compare closed-source OpenAI models against open-weight models—including Llama 3.1 8B-Instruct—employing QLoRA for efficient fine-tuning on consumer-grade GPUs, and integrate structured JSON few-shot prompts. A key innovation is the pre-filling of fine-tuning with low-cost synthetic data. Results show that fine-tuned OpenAI models substantially outperform few-shot prompting baselines; native fine-tuning of Llama yields limited gains, but incorporating synthetic data leads to significant performance improvements—modulated markedly by subject domain. This work is the first to systematically characterize the synergistic interplay among synthetic data quality, model openness (i.e., weight accessibility), and domain-specific characteristics in determining ASAG fine-tuning efficacy.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Prompt Engineering / Prompting
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Research to improve Automated Short Answer Grading has recently focused on Large Language Models (LLMs) with prompt engineering and no- or few-shot prompting to achieve best results. This is in contrast to the fine-tuning approach, which has historically required large-scale compute clusters inaccessible to most users. New closed-model approaches such as OpenAI's fine-tuning service promise results with as few as 100 examples, while methods using open weights such as quantized low-rank adaptive (QLORA) can be used to fine-tune models on consumer GPUs. We evaluate both of these fine-tuning methods, measuring their interaction with few-shot prompting for automated short answer grading (ASAG) with structured (JSON) outputs. Our results show that finetuning with small amounts of data has limited utility for Llama open-weight models, but that fine-tuning methods can outperform few-shot baseline instruction-tuned LLMs for OpenAI's closed models. While our evaluation set is limited, we find some evidence that the observed benefits of finetuning may be impacted by the domain subject matter. Lastly, we observed dramatic improvement with the LLama 3.1 8B-Instruct open-weight model by seeding the initial training examples with a significant amount of cheaply generated synthetic training data.

Problem

• Evaluating fine-tuning vs few-shot prompting for short answer grading
• Comparing closed and open-weight models' performance in ASAG
• Assessing impact of synthetic data on open-weight model fine-tuning

Innovation

• Fine-tuning LLMs with minimal data examples
• Using QLORA for consumer GPU fine-tuning
• Enhancing models with synthetic training data