Stem: Rethinking Causal Information Flow in Sparse Attention

The quadratic computational complexity of self-attention remains a fundamental bottleneck for scaling Large Language Models (LLMs) to long contexts, particularly during the pre-filling phase. In this paper, we rethink the causal attention mechanism from the perspective of information flow. Due to causal constraints, tokens at initial positions participate in the aggregation of every subsequent token. However, existing sparse methods typically apply a uniform top-$k$ selection across all token positions within a layer, ignoring the cumulative dependency of token information inherent in causal architectures. To address this, we propose Stem, a novel, plug-and-play sparsity module aligned with information flow. First, Stem employs the Token Position-Decay strategy, applying position-dependent top-$k$ within each layer to retain initial tokens for recursive dependencies. Second, to preserve information-rich tokens, Stem utilizes the Output-Aware Metric. It prioritizes high-impact tokens based on approximate output magnitude. Extensive evaluations demonstrate that Stem achieves superior accuracy with reduced computation and pre-filling latency. Code is available at https://anonymous.4open.science/r/stem-254ENLXL.

Large language models are increasingly used to process long documents, but this can be slow and costly because the model must compare many parts of the input with one another. We propose Stem, a simple method for making this initial processing more efficient. Stem is based on the observation that early parts of a long input often influence many later computations, so they should be preserved more carefully when reducing computation. The method also keeps pieces of text that contribute strongly to the model’s internal representation. Experiments show that Stem can speed up long-context processing and reduce computation while maintaining performance close to the original model. It does not require retraining, making it easy to add to existing language models.