Logits are All We Need to Adapt Closed Models
This addresses the limitation for developers who rely on prompt tuning with closed-source LLMs, offering a more powerful adaptation method if logit access is provided.
The paper tackles the problem of adapting closed-source large language models (LLMs) for task-specific content generation by proposing a token-level probability reweighting framework that uses logits and small task-specific data, achieving effective steering without model modifications.
Many commercial Large Language Models (LLMs) are often closed-source, limiting developers to prompt tuning for aligning content generation with specific applications. While these models currently do not provide access to token logits, we argue that if such access were available, it would enable more powerful adaptation techniques beyond prompt engineering. In this paper, we propose a token-level probability reweighting framework that, given access to logits and a small amount of task-specific data, can effectively steer black-box LLMs toward application-specific content generation. Our approach views next-token prediction through the lens of supervised classification. We show that aligning black-box LLMs with task-specific data can be formulated as a label noise correction problem, leading to Plugin model -- an autoregressive probability reweighting model that operates solely on logits. We provide theoretical justification for why reweighting logits alone is sufficient for task adaptation. Extensive experiments with multiple datasets, LLMs, and reweighting models demonstrate the effectiveness of our method, advocating for broader access to token logits in closed-source models.