arxiv.org
The representation landscape of few-shot learning and fine-tuning in large language models
In-context learning (ICL) and supervised fine-tuning (SFT) are two common strategies for improving the performance of modern large language models (LLMs) on specific tasks. Despite their different nat...
