Adaptive Classifier-Free Guidance via Dynamic Low-Confidence Masking

TL;DR

This paper introduces Adaptive Classifier-Free Guidance (A-CFG), a dynamic method that improves language generation by adjusting guidance based on the model's real-time confidence, leading to better control and quality.

Contribution

A-CFG dynamically adjusts guidance during iterative generation by masking low-confidence tokens, enhancing controllability and performance over standard static guidance methods.

Findings

Achieves a 3.9 point improvement on GPQA benchmark.

Effectively targets ambiguous tokens for guidance.

Demonstrates significant gains across diverse benchmarks.

Abstract

Classifier-Free Guidance (CFG) significantly enhances controllability in generative models by interpolating conditional and unconditional predictions. However, standard CFG often employs a static unconditional input, which can be suboptimal for iterative generation processes where model uncertainty varies dynamically. We introduce Adaptive Classifier-Free Guidance (A-CFG), a novel method that tailors the unconditional input by leveraging the model's instantaneous predictive confidence. At each step of an iterative (masked) diffusion language model, A-CFG identifies tokens in the currently generated sequence for which the model exhibits low confidence. These tokens are temporarily re-masked to create a dynamic, localized unconditional input. This focuses CFG's corrective influence precisely on areas of ambiguity, leading to more effective guidance. We integrate A-CFG into a state-of-the-art masked diffusion language model and demonstrate its efficacy. Experiments on diverse language generation benchmarks show that A-CFG yields substantial improvements over standard CFG, achieving, for instance, a 3.9 point gain on GPQA. Our work highlights the benefit of dynamically adapting guidance mechanisms to model uncertainty in iterative generation.