Inference-Time Policy Adapters (IPA): Tailoring Extreme-Scale LMs without Fine-tuning
Abstract
While extreme-scale language models have demonstrated exceptional performance on a variety of language tasks, the degree of control over these language models through pure prompting can often be limited. Directly fine-tuning such language models can be effective for tailoring them, but it can be either extremely costly (e.g., GPT-3) or not even feasible for the broader community (e.g., GPT-4). We propose Inference-time Policy Adapters (IPA), which efficiently tailors a language model such as GPT-3 without fine-tuning it. IPA guides a large base model during decoding time through a lightweight policy adapter trained to optimize an arbitrary user objective with reinforcement learning. On five challenging text generation tasks, such as toxicity reduction and lexically constrained generation, IPA consistently brings significant improvements over off-the-shelf language models. It outperforms competitive baseline methods, sometimes even including expensive fine-tuning. In particular, tailoring GPT-2 with IPA can outperform GPT-3, while tailoring GPT-3 with IPA brings a major performance boost over GPT-3 (and sometimes even over GPT-4). Our promising results highlight the potential of IPA as a lightweight alternative to tailoring extreme-scale language models.
Cite
@article{arxiv.2305.15065,
title = {Inference-Time Policy Adapters (IPA): Tailoring Extreme-Scale LMs without Fine-tuning},
author = {Ximing Lu and Faeze Brahman and Peter West and Jaehun Jang and Khyathi Chandu and Abhilasha Ravichander and Lianhui Qin and Prithviraj Ammanabrolu and Liwei Jiang and Sahana Ramnath and Nouha Dziri and Jillian Fisher and Bill Yuchen Lin and Skyler Hallinan and Xiang Ren and Sean Welleck and Yejin Choi},
journal= {arXiv preprint arXiv:2305.15065},
year = {2023}
}
Comments
EMNLP 2023