TriAttention: Efficient Long Reasoning with Trigonometric KV Compression
This addresses efficiency issues for deploying long-context LLMs on resource-limited hardware, such as enabling OpenClaw on a single consumer GPU, though it is incremental as it builds on existing KV compression methods.
The paper tackles the KV cache memory bottleneck in large language models during extended reasoning by proposing TriAttention, a method that estimates key importance using trigonometric series based on stable pre-RoPE Q/K concentration, achieving 2.5x higher throughput or 10.7x KV memory reduction while matching full attention accuracy on AIME25 with 32K-token generation.
Extended reasoning in large language models (LLMs) creates severe KV cache memory bottlenecks. Leading KV cache compression methods estimate KV importance using attention scores from recent post-RoPE queries. However, queries rotate with position during RoPE, making representative queries very few, leading to poor top-key selection and unstable reasoning. To avoid this issue, we turn to the pre-RoPE space, where we observe that Q and K vectors are highly concentrated around fixed non-zero centers and remain stable across positions -- Q/K concentration. We show that this concentration causes queries to preferentially attend to keys at specific distances (e.g., nearest keys), with the centers determining which distances are preferred via a trigonometric series. Based on this, we propose TriAttention to estimate key importance by leveraging these centers. Via the trigonometric series, we use the distance preference characterized by these centers to score keys according to their positions, and also leverage Q/K norms as an additional signal for importance estimation. On AIME25 with 32K-token generation, TriAttention matches Full Attention reasoning accuracy while achieving 2.5x higher throughput or 10.7x KV memory reduction, whereas leading baselines achieve only about half the accuracy at the same efficiency. TriAttention enables OpenClaw deployment on a single consumer GPU, where long context would otherwise cause out-of-memory with Full Attention.