MoBA: Mixture of Block Attention for Long-Context LLMs

TL;DR

MoBA introduces a flexible attention mechanism for long-context LLMs that balances efficiency and performance by dynamically switching between full and sparse attention, advancing capabilities for complex reasoning tasks.

Contribution

The paper presents MoBA, a novel mixture of block attention architecture that enables autonomous, adaptive attention in LLMs, improving efficiency without sacrificing performance.

Findings

MoBA outperforms existing methods on long-context tasks.

It enables seamless transition between full and sparse attention.

Supports efficient long-context processing in real-world applications.

Abstract

Scaling the effective context length is essential for advancing large language models (LLMs) toward artificial general intelligence (AGI). However, the quadratic increase in computational complexity inherent in traditional attention mechanisms presents a prohibitive overhead. Existing approaches either impose strongly biased structures, such as sink or window attention which are task-specific, or radically modify the attention mechanism into linear approximations, whose performance in complex reasoning tasks remains inadequately explored. In this work, we propose a solution that adheres to the ``less structure'' principle, allowing the model to determine where to attend autonomously, rather than introducing predefined biases. We introduce Mixture of Block Attention (MoBA), an innovative approach that applies the principles of Mixture of Experts (MoE) to the attention mechanism. This novel architecture demonstrates superior performance on long-context tasks while offering a key advantage: the ability to seamlessly transition between full and sparse attention, enhancing efficiency without the risk of compromising performance. MoBA has already been deployed to support Kimi's long-context requests and demonstrates significant advancements in efficient attention computation for LLMs. Our code is available at https://github.com/MoonshotAI/MoBA.