TriangleMix: Accelerating Prefilling via Decoding-time Contribution Sparsity
This addresses the time bottleneck in LLM prefilling for users needing fast inference, offering a novel acceleration method that can be combined with existing approaches for incremental improvements.
The paper tackles the quadratic attention complexity bottleneck in LLM prefilling by identifying decoding-time contribution sparsity and proposing TriangleMix, a training-free static attention pattern that uses dense and Triangle attention in different layers, achieving a 15.3x speedup in attention computation for 128K inputs while preserving nearly lossless performance.
Large Language Models (LLMs) incur quadratic attention complexity with input length, creating a major time bottleneck in the prefilling stage. Existing acceleration methods largely exploit attention score sparsity by estimating blocks with high attention scores and applying dynamic sparse attention. In this work, we identify another untapped form of sparsity in the prefilling stage, namely decoding-time contribution sparsity, where many attention blocks exhibit nontrivial attention scores during prefilling yet contribute negligibly to subsequent decoding, as indicated by gradient-based analysis. Building on this observation, we propose TriangleMix, a training-free static attention pattern that uses dense attention in a subset of layers and switches to Triangle attention in the others. Extensive experiments show that TriangleMix preserves nearly lossless performance relative to dense attention while substantially reducing attention overhead in Triangle layers. For 128K inputs, Triangle attention achieves a 15.3x speedup in attention computation, significantly exceeding the acceleration of typical dynamic sparse methods (1.9x to 3.4x). Furthermore, TriangleMix can be seamlessly combined with dynamic sparsity approaches, delivering an additional 6% to 19% reduction in TTFT over using dynamic sparsity alone.