aiqbench analysis
This paper targets a concrete bottleneck in long-context LLM inference: at one million tokens, the lightweight indexer that selects which keys each query attends to consumes more than half of prefill time. AttentionIndex replaces exhaustive scoring with an inverted-file index over indexer keys, reporting 1.3-1.8x prefill throughput over exhaustive selection on GLM-5.3 and DeepSeek-V4-Flash. The gains against HiSA are shown at matched throughput and candidate budget, so they are more meaningful than raw speedups alone, but they come from the authors' own benchmarks and have not been independently reproduced.
AttentionIndex: Accelerating Learned Sparse Attention with a Causal Inverted-File Index
arXiv:2611.02508v1 Announce Type: new Abstract: Learned sparse attention, used in production models such as DeepSeek-V4 and GLM-5, lets each query attend to only k keys chosen by a lightweight indexer, but the indexer still scores every key in the prefix. At one million tokens, this selection takes more than half of prefill time. Two-stage selection over contiguous blocks, as in HiSA, reduces this cost, but its recall declines as context grows at a fixed candidate budget. We present AttentionIndex, which routes each query over an inverted-file index that groups indexer keys by similarity and refines the candidates with the native indexer score. AttentionIndex builds the index during prefill, making new keys searchable immediately, reusing previous clustering to keep maintenance cheap, and exposing each query only to clusters built from its own prefix. On GLM-5.3 and DeepSeek-V4-Flash at one million tokens, AttentionIndex delivers 1.3-1.8 times the prefill throughput of exhaustive selection, and at throughput within 5% of HiSA, it recovers 17.0 and 10.0 percentage points more of the exhaustive top-k, respectively. AttentionIndex also improves accuracy on long-context benchmarks by up to 18.4 percentage points over HiSA at the same candidate budget.
Abstract:Learned sparse attention, used in production models such as DeepSeek-V4 and GLM-5, lets each query attend to only k keys chosen by a lightweight indexer, but the indexer still scores every key in the prefix. At one million tokens, this selection takes more than half of prefill time. Two-stage selection over contiguous blocks, as in HiSA, reduces this cost, but its recall declines as context grows at a fixed candidate budget. We present AttentionIndex, which routes each query over an inverted-file index that groups indexer keys by similarity and refines the candidates with the native indexer score. AttentionIndex builds the index during prefill, making new keys searchable immediately, reusing previous clustering to keep maintenance cheap, and exposing each query only to clusters built from its own prefix. On GLM-5.3 and DeepSeek-V4-Flash at one million tokens, AttentionIndex delivers 1.3-1.8 times the prefill throughput of exhaustive selection, and at throughput within 5% of HiSA, it recovers 17.0 and 10.0 percentage points more of the exhaustive top-k, respectively. AttentionIndex also improves accuracy on long-context benchmarks by up to 18.4 percentage points over HiSA at the same candidate budget.
Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2611.02508 [cs.AI]
(or arXiv:2611.02508v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2611.02508 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 282 words