PulseAugur
LIVE 12:26:25
research · [1 source] ·
0
research

AdaCluster framework speeds up video generation with adaptive query-key clustering

Researchers have developed AdaCluster, a new framework designed to significantly speed up video diffusion transformers (DiTs). This method addresses the slow inference times caused by the quadratic complexity of attention mechanisms in these models. AdaCluster employs adaptive clustering techniques for both query and key vectors to compress attention, achieving up to a 4.31x speedup on various video generation models with minimal loss in quality. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

RANK_REASON Academic paper detailing a new method for improving AI model efficiency.

Read on Hugging Face Daily Papers →

AdaCluster framework speeds up video generation with adaptive query-key clustering

COVERAGE [1]

  1. Hugging Face Daily Papers TIER_1 ·

    AdaCluster: Adaptive Query-Key Clustering for Sparse Attention in Video Generation

    Video diffusion transformers (DiTs) suffer from prohibitive inference latency due to quadratic attention complexity. Existing sparse attention methods either overlook semantic similarity or fail to adapt to heterogeneous token distributions across layers, leading to model perform…