Log-Linear Attention
This work addresses the scalability limitations of attention mechanisms in sequence modeling for AI applications, offering a novel approach that improves upon linear attention variants while maintaining efficiency.
The paper tackled the quadratic compute and linear memory bottlenecks of Transformer attention by introducing log-linear attention, which balances efficiency and expressiveness with a compute cost that grows log-linearly in sequence length, and demonstrated its effectiveness by applying it to Mamba-2 and Gated DeltaNet architectures.
The attention mechanism in Transformers is an important primitive for accurate and scalable sequence modeling. Its quadratic-compute and linear-memory complexity however remain significant bottlenecks. Linear attention and state-space models enable linear-time, constant-memory sequence modeling and can moreover be trained efficiently through matmul-rich parallelization across sequence length. However, at their core these models are still RNNs, and thus their use of a fixed-size hidden state to model the context is a fundamental limitation. This paper develops log-linear attention, an attention mechanism that balances linear attention's efficiency and the expressiveness of softmax attention. Log-linear attention replaces the fixed-size hidden state with a logarithmically growing set of hidden states. We show that with a particular growth function, log-linear attention admits a similarly matmul-rich parallel form whose compute cost is log-linear in sequence length. Log-linear attention is a general framework and can be applied on top of existing linear attention variants. As case studies, we instantiate log-linear variants of two recent architectures -- Mamba-2 and Gated DeltaNet -- and find they perform well compared to their linear-time variants.