When Do Prompting and Prefix-Tuning Work? A Theory of Capabilities and Limitations

arXiv:2310.19698v2 Announce Type: replace
Abstract: Context-based fine-tuning methods, including prompting, in-context learning, soft prompting (also known as prompt tuning), and prefix-tuning, have gained popularity due to their ability to often match the performance of full fine-tuning with a fraction of the parameters. Despite their empirical successes, there is little theoretical understanding of how these techniques influence the internal computation of the model and their expressiveness limitations. We show that despite the continuous embedding space being more expressive than the discrete …

abstract arxiv capabilities context cs.cl cs.lg fine-tuning in-context learning limitations match parameters performance prompt prompting prompt tuning theory type work