Adaptive Guidance: Training-free Acceleration of Conditional Diffusion Models

TL;DR

This paper introduces Adaptive Guidance, a training-free method that reduces computation in text-conditioned diffusion models by adaptively skipping neural network evaluations during denoising, maintaining quality while significantly increasing speed.

Contribution

The paper proposes Adaptive Guidance, a novel, training-free approach that adaptively omits guidance network evaluations, improving inference efficiency in diffusion models without retraining.

Findings

AG reduces computation by 25% while preserving image quality.

AG achieves 50% of Guidance Distillation's speed-up without training.

LinearAG replaces neural evaluations with affine transforms for even cheaper inference.

Abstract

This paper presents a comprehensive study on the role of Classifier-Free Guidance (CFG) in text-conditioned diffusion models from the perspective of inference efficiency. In particular, we relax the default choice of applying CFG in all diffusion steps and instead search for efficient guidance policies. We formulate the discovery of such policies in the differentiable Neural Architecture Search framework. Our findings suggest that the denoising steps proposed by CFG become increasingly aligned with simple conditional steps, which renders the extra neural network evaluation of CFG redundant, especially in the second half of the denoising process. Building upon this insight, we propose "Adaptive Guidance" (AG), an efficient variant of CFG, that adaptively omits network evaluations when the denoising process displays convergence. Our experiments demonstrate that AG preserves CFG's image quality while reducing computation by 25%. Thus, AG constitutes a plug-and-play alternative to Guidance Distillation, achieving 50% of the speed-ups of the latter while being training-free and retaining the capacity to handle negative prompts. Finally, we uncover further redundancies of CFG in the first half of the diffusion process, showing that entire neural function evaluations can be replaced by simple affine transformations of past score estimates. This method, termed LinearAG, offers even cheaper inference at the cost of deviating from the baseline model. Our findings provide insights into the efficiency of the conditional denoising process that contribute to more practical and swift deployment of text-conditioned diffusion models.