CLAA: Cross-Layer Attention Aggregation for Accelerating LLM Prefill
This work addresses a key performance issue in LLM inference for applications requiring fast response times, though it is incremental as it builds on existing token-ranking heuristics.
The paper tackled the computational bottleneck in the prefill stage of long-context LLM inference by introducing Cross-Layer Attention Aggregation (CLAA), which reduced Time-to-First-Token by up to 39% compared to the Full KV Cache baseline.
The prefill stage in long-context LLM inference remains a computational bottleneck. Recent token-ranking heuristics accelerate inference by selectively processing a subset of semantically relevant tokens. However, existing methods suffer from unstable token importance estimation, often varying between layers. Evaluating token-ranking quality independently from heuristic-specific architectures is challenging. To address this, we introduce an Answer-Informed Oracle, which defines ground-truth token importance by measuring attention from generated answers back to the prompt. This oracle reveals that existing heuristics exhibit high variance across layers: rankings can degrade sharply at specific layers, a failure mode invisible to end-to-end benchmarks. The diagnosis suggests a simple fix: aggregate scores across layers rather than relying on any single one. We implement this as Cross-Layer Attention Aggregation (CLAA), which closes the gap to the oracle upper bound and reduces Time-to-First-Token (TTFT) by up to 39\% compared to the Full KV Cache baseline.