In-Context Learning Distillation for Efficient Few-Shot Fine-Tuning
This work addresses the challenge of high computational costs for few-shot fine-tuning in natural language processing, offering a more efficient method for deploying large models in resource-constrained settings.
The paper tackled the problem of efficient few-shot fine-tuning by applying in-context learning distillation to reduce model size from 1.3B to 125M and memory from 2.5GB to 0.25GB, achieving a 50% improvement in out-of-domain accuracy over in-context learning alone and a 20% improvement with 60% memory reduction compared to conventional fine-tuning.
We applied few-shot in-context learning on the OPT-1.3B model for the natural language inference task and employed knowledge distillation to internalize the context information, reducing model parameter from 1.3B to 125M and achieving a size reduction from 2.5GB to 0.25GB. Compared to using in-context learning alone on similarly sized models, this context distillation approach achieved a nearly 50% improvement in out-of-domain accuracy, demonstrating superior knowledge transfer capabilities over prompt-based methods. Furthermore, this approach reduced memory consumption by up to 60% while delivering a 20% improvement in out-of-domain accuracy compared to conventional pattern-based fine-tuning.