Context Tuning for In-Context Optimization
This addresses the need for efficient and effective lightweight adaptation methods for large language models, though it is incremental over existing prompt-based techniques.
The paper tackles the problem of improving few-shot adaptation of language models without fine-tuning by introducing Context Tuning, which initializes trainable prompts with task-specific examples to leverage in-context learning, resulting in competitive accuracy to Test-Time Training with higher efficiency across benchmarks like CrossFit and MMLU.
We introduce Context Tuning, a simple and effective method to significantly enhance few-shot adaptation of language models (LLMs) without fine-tuning model parameters. While prompt-based adaptation techniques have demonstrated the effectiveness of lightweight adaptation methods for LLMs, they typically initialize a trainable prompt or prefix with irrelevant tokens for the task at hand. In contrast, Context Tuning initializes the trainable prompt or prefix with task-specific demonstration examples, leveraging the model's inherent In-Context Learning (ICL) ability to extract relevant information for improved few-shot learning performance. Extensive evaluations on benchmarks such as CrossFit, UnifiedQA, MMLU, BIG-Bench Hard, and ARC demonstrate that Context Tuning outperforms traditional prompt-based adaptation methods and achieves competitive accuracy to Test-Time Training with significantly higher training efficiency.