Inference-Time Policy Adapters (IPA): Tailoring Extreme-Scale LMs without Fine-tuning
This addresses the challenge of tailoring large language models for specific tasks without expensive fine-tuning, making it more accessible for the broader community, though it is incremental as it builds on existing reinforcement learning and adapter methods.
The paper tackles the problem of controlling extreme-scale language models without costly fine-tuning by proposing Inference-time Policy Adapters (IPA), which guide models during decoding using a lightweight policy trained with reinforcement learning, resulting in significant improvements on tasks like toxicity reduction and outperforming baselines including fine-tuning, with GPT-3 tailored by IPA sometimes beating GPT-4.
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.