arXiv:2607. 07724v1 Announce Type: new Abstract: Block-sparse attention scales long-context language models by replacing the O(N^2) softmax with a per-query top-k selection over key blocks.
arXiv:2606. 07703v1 Announce Type: cross Abstract: Long-context prefill remains expensive because full/GQA layers still score the historical sequence, even in hybrid models with local, sparse, linear, or recurrent components.
arXiv:2606. 13392v1 Announce Type: new Abstract: Ultra-long-context capability is becoming indispensable for frontier LLMs: agentic workflows, repository-scale code reasoning, and persistent memory all require the model to jointly attend over hundreds of thousands to millions of tokens, yet the quadratic cost of softmax attention makes this untenable at deployment scale.
RBS-Attention introduces a training‑free, radius‑bounded sparse prefill strategy for long‑context large language models, addressing the mean dilution problem where a block centroid can miss highly relevant tokens. The method employs two complementary selection branches: a centroid base branch that captures average relevance and a rescue branch that uses the maximum key‑block radius to flag under‑estimated blocks. Experiments on Qwen3 models demonstrate significant speedups—over 20× in standalone prefill‑attention and nearly 6× in end‑to‑first‑token time—while maintaining competitive accuracy compared to dense attention.
arXiv:2609.26086v1 Announce Type: new Abstract: An agentic retrieval system issues a sequence of search queries and must decide, at each step, whether the evidence collected so far is enough to stop....
arXiv:2608. 02947v1 Announce Type: new Abstract: The attention score with rotary position embeddings (RoPE) decomposes exactly into a sum over its 2D-rotation frequency pairs, and each pair's wavelength limits how far it can discriminate position.
Block Sparse Flash Attention (BSFA) is a drop‑in replacement for FlashAttention that speeds up long‑context inference by pruning about 50% of computation and memory transfers. It selects the top‑k most important value blocks for each query using exact query‑key similarities and calibrated per‑layer, per‑head thresholds, requiring only a one‑time training‑free calibration. On Llama‑3.1‑8B, BSFA delivers up to 1.13× speedup on LongBench with a 1.1% accuracy drop and up to 1.24× on Needle‑in‑a‑Haystack retrieval with a 1% drop, while the attention kernel itself accelerates by up to 1.38×.
arXiv:2609.05637v2 Announce Type: replace Abstract: A popular way to improve Retrieval-Augmented Generation (RAG) is to rewrite the user's question into several variants and search with all of them....
ValueDiff introduces a value‑geometric KV cache eviction strategy for large language models that suppress attention sinks. It ranks tokens by the L2 deviation of their value vectors from the cache mean, a score that aligns with minimal‑disturbance eviction under a max‑entropy assumption. Across several benchmarks—RULER, LongBench, and MATH‑500—ValueDiff consistently retains a higher proportion of useful tokens than prior methods, especially under tight cache budgets.
arXiv:2608. 08853v1 Announce Type: new Abstract: Sparse Mixture-of-Experts (MoE) routers commonly use the same scores both to select experts and to weight their already-computed outputs.
arXiv:2607. 09052v1 Announce Type: new Abstract: Block sparse attention is a hardware friendly way to alleviate the key-value (KV) cache read bottleneck in large language models (LLMs).
arXiv:2609.03949v2 Announce Type: replace-cross Abstract: A long-lived KV cache must be compressed before the queries that will read it exist. Selection by observed attention collapses there: on a No...
The paper investigates whether training Mixture-of-Experts (MoE) routers can improve memory‑bandwidth locality on consumer GPUs. Using a new zero‑surgery telemetry tool, the authors measure that a large Qwen3‑235B model is bottlenecked by disk‑based expert access, and that an LRU cache can serve a majority of requests. They pre‑register experiments training 137 M‑parameter MoE models with locality‑aware losses, finding that while cache misses can drop up to 60 % (99 % static‑pin hit rate), every configuration fails to meet a strict 1 % perplexity threshold, indicating a tight coupling between cache efficiency and model quality.