Training-free Context-adaptive Attention for Efficient Long Context Modeling

TL;DR

This paper introduces TCA-Attention, a training-free, adaptive sparse attention mechanism that significantly speeds up long-context modeling in large language models while maintaining performance, without additional training or architectural changes.

Contribution

The paper proposes a novel training-free sparse attention method that adaptively selects informative tokens, reducing computational cost and memory footprint for long-context inference.

Findings

Achieves 2.8× speedup in inference

Reduces KV cache by 61% at 128K context length

Maintains comparable performance to full attention

Abstract

Large Language Models (LLMs) have demonstrated remarkable capabilities across a wide range of natural language processing tasks. These capabilities stem primarily from the self-attention mechanism, which enables modeling of long-range dependencies. However, the quadratic complexity of self-attention with respect to sequence length poses significant computational and memory challenges, especially as sequence length extends to extremes. While various sparse attention and KV cache compression methods have been proposed to improve efficiency, they often suffer from limitations such as reliance on fixed patterns, inability to handle both prefilling and decoding stages, or the requirement for additional training. In this paper, we propose Training-free Context-adaptive Attention (TCA-Attention), a training-free sparse attention mechanism that selectively attends to only the informative tokens for efficient long-context inference. Our method consists of two lightweight phases: i) an offline calibration phase that determines head-specific sparsity budgets via a single forward pass, and ii) an online token selection phase that adaptively retains core context tokens using a lightweight redundancy metric. TCA-Attention provides a unified solution that accelerates both prefilling and decoding while reducing KV cache memory footprint, without requiring parameter updates or architectural changes. Theoretical analysis shows that our approach maintains bounded approximation error. Extensive experiments demonstrate that TCA-Attention achieves a 2.8$\times$ speedup and reduces KV cache by 61% at 128K context length while maintaining performance comparable to full attention across various benchmarks, offering a practical plug-and-play solution for efficient long-context inference.