The paper introduces GAAC, a lossless key‑value (KV) compression method for large language model inference on personal devices. GAAC encodes BF16 exponents with per‑channel base‑plus‑offset codes, adapts code widths across small groups, and reconstructs KV tiles on the fly to avoid full‑cache decompression. Experiments show 1.4× compression for BF16 KV, speedups of up to 1.27× in decode attention, and a 26% reduction in resident KV memory while boosting throughput on NVIDIA and Apple hardware.
arXiv:2610.03027v1 Announce Type: cross Abstract: The key-value (KV) cache becomes a major memory bottleneck in long-context LLM inference, placing substantial pressure on memory capacity and bandwid...
arXiv:2607. 00760v1 Announce Type: new Abstract: Long-context LLM services now sustain prompts with hundreds of thousands to millions of tokens, making the key-value (KV) cache a first-order serving cost.
The paper introduces ReadKV, a query‑adaptive quantization method for KV caches that stores keys and values in progressive codes, allowing each decoding query to allocate different precision prefixes for keys and values. By optimizing a calibrated distortion objective under fixed budgets, ReadKV achieves near‑optimal allocation and demonstrates that query‑dependent access can outperform any query‑independent reader at the same read budget. Experiments on six base models show that reading only four bits on average from an eight‑bit cache increases perplexity by at most 0.66% while reducing logical reads and retained capacity compared to a 16‑bit cache, and it achieves lower latency than existing codecs on an NVIDIA A10G.
arXiv:2605. 01708v3 Announce Type: replace-cross Abstract: Contemporary systems serving large language models (LLMs) have adopted prefill-decode disaggregation to load-balance between the compute-bound prefill phase and the memory-bound decode phase.
arXiv:2606. 24033v1 Announce Type: new Abstract: Existing low-bit KV-cache quantizers often treat each cached key as a flat vector.
arXiv:2611.06377v1 Announce Type: new Abstract: Autoregressive video models rely on historical key-value (KV) caches to preserve scene consistency over extended rollouts, making KV storage and repeat...
arXiv:2605. 06675v2 Announce Type: replace Abstract: Large language models cache all previously computed key-value (KV) pairs during generation, and this KV cache grows linearly with sequence length, making it a primary memory bottleneck for serving.
arXiv:2609.21172v1 Announce Type: new Abstract: Large language models (LLMs) are moving onto mobile devices for increasingly diverse workloads over text, images, video, and audio. These applications...
arXiv:2609.05764v1 Announce Type: cross Abstract: The key-value (KV) cache is the dominant memory bottleneck in long-context large language model (LLM) decoding: every step reads it entirely, so deco...
arXiv:2610.15384v1 Announce Type: cross Abstract: Compressing the key-value (KV) cache is essential for serving long-context large language models (LLMs) under memory constraints. While current scala...
Minima-KV introduces a retention‑preserving hierarchy for mixed‑format paged attention that keeps recent and protected anchor pages in FP8 while older pages are compressed into packed TQ3, allowing every live‑request page to remain addressable. The approach uses format‑specific kernels and a globally normalized online‑softmax merge to compute partial attention states, enabling direct heterogeneous decoding without a dense shadow cache. Experiments on Qwen3.6‑27B on a 96‑GB NVIDIA RTX PRO 6000 Blackwell GPU show 3.50× compression over BF16 and 1.75× over FP8, with minimal impact on performance across long‑context benchmarks.
The paper introduces CORE, a method that jointly calibrates coverage and redistributes evicted mass for key–value (KV) cache in long‑context decoding. CORE combines query utility and log‑determinant coverage into a lightweight indexer, using a single calibrated distribution to both retain complementary KV states and parameterize latent‑memory writes. Experiments on three backbones show CORE outperforms the RULER baseline by up to 3.78 points at 90% compression and demonstrates strong decoding efficiency on LongBench and repeated‑eviction tests.