XAttention: Block Sparse Attention with Antidiagonal Scoring
This addresses efficiency bottlenecks for deploying long-context Transformers in real-world applications like language, video understanding, and generation, representing a novel method for a known bottleneck.
The paper tackles the high computational cost of long-context Transformer models by introducing XAttention, a block-sparse attention method that uses antidiagonal scoring to identify and prune non-essential blocks, achieving up to 13.5x acceleration in attention computation while maintaining accuracy comparable to full attention.
Long-Context Transformer Models (LCTMs) are vital for real-world applications but suffer high computational costs due to attention's quadratic complexity. Block-sparse attention mitigates this by focusing computation on critical regions, yet existing methods struggle with balancing accuracy and efficiency due to costly block importance measurements. In this paper, we introduce XAttention, a plug-and-play framework that dramatically accelerates long-context inference in Transformers models using sparse attention. XAttention's key innovation is the insight that the sum of antidiagonal values (i.e., from the lower-left to upper-right) in the attention matrix provides a powerful proxy for block importance. This allows for precise identification and pruning of non-essential blocks, resulting in high sparsity and dramatically accelerated inference. Across comprehensive evaluations on demanding long-context benchmarks-including RULER and LongBench for language, VideoMME for video understanding, and VBench for video generation. XAttention achieves accuracy comparable to full attention while delivering substantial computational gains. We demonstrate up to 13.5x acceleration in attention computation. These results underscore XAttention's ability to unlock the practical potential of block sparse attention, paving the way for scalable and efficient deployment of LCTMs in real-world applications. Code is available at https://github.com/mit-han-lab/x-attention.