The paper introduces Memory Attention (MA), a new attention mechanism for language models that replaces the traditional value projection with token-indexed memory combined with contextual keys. MA generates values by merging layer‑specific token memory with contextual information, allowing normalization to be folded into memory tables and reducing value construction to a simple lookup and addition. Experiments show that, with matched training token budgets and additional memory parameters, MA improves language modeling performance and average downstream task results across various attention configurations.
A lot of prior work addressed key-value (KV) cache selection and compression by sparse attention to enable long-context inference for transformer language models without excessive hardware budgets. We provide a new method for fine-tuning models with sparse attention.
arXiv:2608. 19920v1 Announce Type: new Abstract: A lot of prior work addressed key-value (KV) cache selection and compression by sparse attention to enable long-context inference for transformer language models without excessive hardware budgets.
The paper introduces On‑Demand Attention (ODA), a local‑first decoding strategy that predicts when a pretrained language model would benefit from global attention. By training only a lightweight recall head, ODA selectively triggers global attention during generation, keeping pretrained weights unchanged and preserving the full key‑value cache for future recall. Experiments on Qwen, Gemma, and hybrid‑attention models show that ODA largely recovers performance lost with local attention while significantly cutting global reads, enabling faster long‑context inference.
The paper introduces On‑Demand Attention (ODA), a decoding strategy that lets pretrained language models decide when to use global attention based on a lightweight recall head. ODA keeps the original model weights unchanged, only training the recall head, and can be implemented with GPU‑side conditional execution to reduce global reads. Experiments on Qwen, Gemma, and hybrid‑attention models show that ODA largely recovers performance lost by local attention while cutting the number of global attention operations.
The paper introduces Declarative Attention (DA), a protocol that lets language models explicitly declare which parts of their context to focus on during generation. By partitioning decoding into full-context, region-specific, and recent-output-only modes, the inference engine can skip large portions of the KV cache, dramatically reducing attended tokens. Experiments on 15 long-context tasks with off-the-shelf models show significant savings (52.0% and 31.1% reductions) with only modest accuracy drops that diminish as model size increases.
arXiv:2606. 18587v1 Announce Type: cross Abstract: Decoder-only Transformers compute attention over the KV cache of preceding tokens.
The paper investigates whether deep transformer layers require context from the residual stream to compute value vectors. It finds that allowing deeper layers to use a context‑free value vector—preserving original token information—significantly improves performance, and adding context afterward yields little extra benefit. The authors introduce Bank of Values (BoV), a lookup table of token‑specific value vectors for the last third of layers, which reduces compute and memory while matching or surpassing prior methods on large models.
arXiv:2507. 07247v2 Announce Type: replace-cross Abstract: As large language models (LLMs) and visual language models (VLMs) grow in scale and application, attention mechanisms have become a central computational bottleneck due to their high memory and time complexity.
arXiv:2606. 31519v1 Announce Type: new Abstract: Long-context Large Language Model inference is severely bottlenecked by the massive Key-Value (KV) cache, yet existing sparse attention methods often suffer from static fixed-budget (Top-k) retrieval or rely on proxy scores that are computationally expensive and biased.
arXiv:2606. 04302v1 Announce Type: cross Abstract: Key-value (KV) caching accelerates inference of large language models (LLMs) by reusing past computations for generated tokens.
arXiv:2609.25537v1 Announce Type: new Abstract: Large language model (LLM) inference is constrained by the quadratic scaling of self-attention and the linear scaling of the KV cache, increasing laten...
CompKV introduces a compensation‑aware sparse attention framework for long‑context LLM inference. It partitions tokens into blocks and optimizes token selection to minimize the error introduced by block‑level mean compensation, using compact block‑level statistics. Experiments on RULER and LongBench‑Pro show CompKV outperforms other sparse baselines and achieves up to a 6.85× speedup over full attention.