{"ID":3083555,"CreatedAt":"2026-06-05T06:46:15.197025399Z","UpdatedAt":"2026-06-07T06:54:00.442624098Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2606.06467","arxiv_id":"2606.06467","title":"You Only Index Once: Cross-Layer Sparse Attention with Shared Routing","abstract":"Long-context inference in modern LLMs is increasingly constrained by decoding efficiency, especially in reasoning-heavy settings where models generate long intermediate chains of thought. Existing sparse attention methods often face a practical efficiency-quality trade-off. Structured block sparse methods typically provide stronger acceleration but incur noticeable quality loss, while token sparse methods are usually more accurate yet deliver limited end-to-end speedup because top-k routing over the full cache remains expensive. In this work, we propose cross-layer sparse attention (CLSA), which is built on top of KV-sharing architectures such as YOCO. The core idea is to share not only the KV cache across cross-decoder layers, but also the routing index. A single indexer computes token-level top-k selection once and reuses the resulting index across layers, thereby preserving the fine-grained selectivity of token sparse attention while amortizing the routing overhead. The resulting architecture improves all major inference bottlenecks jointly, including pre-filling, KV-cache storage, and long-context decoding. Experiments across short-context and long-context benchmarks show that CLSA is both accurate and efficient, achieving up to 7.6x decoding speedup and 17.1x overall throughput improvement at 128K context. These results suggest a more complete architectural solution for long-context LLMs that jointly advances model quality and inference efficiency.","short_abstract":"Long-context inference in modern LLMs is increasingly constrained by decoding efficiency, especially in reasoning-heavy settings where models generate long intermediate chains of thought. Existing sparse attention methods often face a practical efficiency-quality trade-off. Structured block sparse methods typically pro...","url_abs":"https://arxiv.org/abs/2606.06467","url_pdf":"https://arxiv.org/pdf/2606.06467v1","authors":"[\"Yutao Sun\",\"Yanqi Zhang\",\"Li Dong\",\"Jianyong Wang\",\"Furu Wei\"]","published":"2026-06-04T17:54:04Z","proceeding":"cs.CL","tasks":"[\"cs.CL\",\"cs.AI\",\"cs.LG\"]","methods":"[\"Large Language Model\"]","has_code":false}