Two papers propose token-adaptive KV cache compression for long-context LLMs

DepthWeave-KV (arXiv, 2026-07-07) proposes token-adaptive cross-layer residual factorization for KV cache compression. It shares low-rank channel bases across neighboring layers and retains lightweight token-specific residuals where attention behavior is sensitive. FreqDepthKV (arXiv, 2026-07-07) uses frequency-guided depth sharing, factorizing adjacent-layer KV states into shared low-frequency depth components and sparse high-frequency residuals. A lightweight online probe assigns attention heads to shared-depth, residual-depth, or exact cache modes based on their contribution to reconstruction-sensitive attention logits. Both methods target the memory bandwidth and capacity bottleneck in long-context LLM inference, aiming to preserve retrieval and multi-step reasoning quality under aggressive compression.