ALTER: All-in-One Layer Pruning and Temporal Expert Routing for Efficient Diffusion Generation
This addresses the problem of deploying diffusion models in resource-constrained environments, representing a strong specific gain rather than an incremental improvement.
The paper tackles the computational inefficiency of diffusion models during inference by introducing ALTER, a unified framework that achieves same-level visual fidelity as the original 50-step Stable Diffusion v2.1 model while using only 25.9% of total MACs with 20 inference steps and delivering a 3.64x speedup through 35% sparsity.
Diffusion models have demonstrated exceptional capabilities in generating high-fidelity images. However, their iterative denoising process results in significant computational overhead during inference, limiting their practical deployment in resource-constrained environments. Existing acceleration methods often adopt uniform strategies that fail to capture the temporal variations during diffusion generation, while the commonly adopted sequential pruning-then-fine-tuning strategy suffers from sub-optimality due to the misalignment between pruning decisions made on pretrained weights and the model's final parameters. To address these limitations, we introduce ALTER: All-in-One Layer Pruning and Temporal Expert Routing, a unified framework that transforms diffusion models into a mixture of efficient temporal experts. ALTER achieves a single-stage optimization that unifies layer pruning, expert routing, and model fine-tuning by employing a trainable hypernetwork, which dynamically generates layer pruning decisions and manages timestep routing to specialized, pruned expert sub-networks throughout the ongoing fine-tuning of the UNet. This unified co-optimization strategy enables significant efficiency gains while preserving high generative quality. Specifically, ALTER achieves same-level visual fidelity to the original 50-step Stable Diffusion v2.1 model while utilizing only 25.9% of its total MACs with just 20 inference steps and delivering a 3.64x speedup through 35% sparsity.