arXiv:2605. 25085v2 Announce Type: replace-cross Abstract: We study the rate-distortion limits of online KV cache compression in autoregressive language models, formulating it as sequential Wyner-Ziv source coding on the filtration induced by the model, with the next-step query as decoder side information.
arXiv:2606. 01563v1 Announce Type: new Abstract: Autoregressive decoding in Transformer-based language models relies on the KV cache, whose memory footprint grows linearly with sequence length and becomes the primary bottleneck for long-context inference.
arXiv:2607. 27539v1 Announce Type: new Abstract: Exact deletion from persistent language-model memory depends on how that memory represents a record.
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:2607. 07144v1 Announce Type: new Abstract: The key-value (KV) cache dominates the memory cost of long-context autoregressive inference, and a growing body of work compresses it through quantization, eviction, or offloading.
arXiv:2608. 26515v1 Announce Type: cross Abstract: We study online prediction for a specific finite-alphabet, exogenously driven source with infinite input memory.
The paper introduces Window-Diffusion, a method that accelerates diffusion language model inference by pruning and caching tokens within a sliding window. It categorizes undecoded tokens into active, buffer, and far-field groups, computing only the first two while discarding the rest. Experiments on LLaDA and Dream demonstrate up to 99× speedup with minimal loss in generation quality.
arXiv:2606. 17872v1 Announce Type: cross Abstract: Large language models (LLMs) outperform earlier architectures on generative inference and long-context tasks, but their large size introduces significant challenges in memory usage, energy cost, and on-device deployment.
arXiv:2609.37887v1 Announce Type: new Abstract: Activation and key-value cache precision change what a quantized language model computes without altering its stored weights. Direct weight-code bounds...
arXiv:2604. 25975v2 Announce Type: replace-cross Abstract: Key-Value (KV) caching is essential for large language model inference, yet its memory overhead poses a critical bottleneck for long-context generation.
arXiv:2607. 10441v1 Announce Type: cross Abstract: Context engineering decides what information a model carries forward, and current designs meter it in tokens: compressing the past into a bounded recurrent state, keeping a key-value entry for every token, or imposing a fixed budget through a window or eviction rule.
PAGE is a partition‑aware gated KV‑cache eviction method that reframes eviction as a per‑input admission decision. It uses a single label‑free scalar— the early‑to‑late drop in pairwise top‑k head agreement—to classify inputs into a capacity‑bound class (where eviction is catastrophic) and a dilution‑prone class (where eviction is safe or beneficial). By thresholding this drop, PAGE applies a base evictor only when necessary, reducing the harm rate in the capacity‑bound regime from 0.75 to 0.026 and achieving a 29× improvement across four models and benchmarks without retraining the evictor.
The paper investigates how block‑diffusion language models can use a constant‑size cache to enable efficient parallel decoding. By employing sequence mixers that summarize completed blocks into a reusable state and a block‑causal training objective, the authors pretrain three 3B block‑diffusion denoisers (attention, Mamba, and hybrid) on 300 B tokens. The resulting state‑space cache remains O(1) in memory and latency regardless of context length, yielding significant speed‑up and memory savings compared to traditional attention‑based caches, especially at very long sequences.