Secokd: Aligning Large Language Models For In-context Learning With Fewer Shots

Wang Weixing, Yang Haojin, Meinel Christoph. Arxiv 2024

[Paper]    
Distillation Efficiency And Optimization Fine Tuning In Context Learning Pretraining Methods Prompting Tools Training Techniques

Previous studies have shown that demonstrations can significantly help Large Language Models (LLMs ) perform better on the given tasks. However, this so-called In-Context Learning ( ICL ) ability is very sensitive to the presenting context, and often dozens of demonstrations are needed. In this work, we investigate if we can reduce the shot number while still maintaining a competitive performance. We present SeCoKD, a self-Knowledge Distillation ( KD ) training framework that aligns the student model with a heavily prompted variation, thereby increasing the utilization of a single demonstration. We experiment with the SeCoKD across three LLMs and six benchmarks focusing mainly on reasoning tasks. Results show that our method outperforms the base model and Supervised Fine-tuning ( SFT ), especially in zero-shot and one-shot settings by 30% and 10%, respectively. Moreover, SeCoKD brings little negative artifacts when evaluated on new tasks, which is more robust than Supervised Fine-tuning.

Similar Work