QuantCache: Adaptive Importance-Guided Quantization with Hierarchical Latent and Layer Caching for Video Generation
This addresses deployment challenges for video generation models on resource-constrained devices, representing a strong incremental improvement over existing acceleration techniques.
The paper tackles the high computational and memory costs of Diffusion Transformers (DiTs) for video generation by proposing QuantCache, a training-free inference acceleration framework that combines hierarchical latent caching, adaptive quantization, and pruning, achieving a 6.72× latency speedup on Open-Sora with minimal quality loss.
Recently, Diffusion Transformers (DiTs) have emerged as a dominant architecture in video generation, surpassing U-Net-based models in terms of performance. However, the enhanced capabilities of DiTs come with significant drawbacks, including increased computational and memory costs, which hinder their deployment on resource-constrained devices. Current acceleration techniques, such as quantization and cache mechanism, offer limited speedup and are often applied in isolation, failing to fully address the complexities of DiT architectures. In this paper, we propose QuantCache, a novel training-free inference acceleration framework that jointly optimizes hierarchical latent caching, adaptive importance-guided quantization, and structural redundancy-aware pruning. QuantCache achieves an end-to-end latency speedup of 6.72$\times$ on Open-Sora with minimal loss in generation quality. Extensive experiments across multiple video generation benchmarks demonstrate the effectiveness of our method, setting a new standard for efficient DiT inference. The code and models will be available at https://github.com/JunyiWuCode/QuantCache.