Inference-Time Language Model Alignment via Integrated Value Guidance
This addresses the computational burden of fine-tuning for researchers and practitioners by enabling efficient alignment at inference time, though it is incremental as it builds on existing value guidance methods.
The paper tackles the problem of aligning large language models with human preferences without fine-tuning by introducing Integrated Value Guidance (IVG), which uses value functions to guide decoding at inference time, resulting in improved alignment metrics such as increasing win rates from 19.51% to 26.51% for Mistral-7B-Instruct-v0.2 on AlpacaEval 2.0.
Large language models are typically fine-tuned to align with human preferences, but tuning large models is computationally intensive and complex. In this work, we introduce $\textit{Integrated Value Guidance}$ (IVG), a method that uses implicit and explicit value functions to guide language model decoding at token and chunk-level respectively, efficiently aligning large language models purely at inference time. This approach circumvents the complexities of direct fine-tuning and outperforms traditional methods. Empirically, we demonstrate the versatility of IVG across various tasks. In controlled sentiment generation and summarization tasks, our method significantly improves the alignment of large models using inference-time guidance from $\texttt{gpt2}$-based value functions. Moreover, in a more challenging instruction-following benchmark AlpacaEval 2.0, we show that both specifically tuned and off-the-shelf value functions greatly improve the length-controlled win rates of large models against $\texttt{gpt-4-turbo}$ (e.g., $19.51\% \rightarrow 26.51\%$ for $\texttt{Mistral-7B-Instruct-v0.2}$ and $25.58\% \rightarrow 33.75\%$ for $\texttt{Mixtral-8x7B-Instruct-v0.1}$ with Tulu guidance).