arXiv:2608. 08569v1 Announce Type: new Abstract: Recent advancements in Speech Large Language Models have demonstrated remarkable capabilities in understanding complex audio tasks.
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 acoustic-to-text KV compression for full‑duplex speech models, converting acoustic key‑value states into compact textual memory during listening‑time slack. When the KV cache exceeds a target budget, older acoustic states are evicted while transcripts and recent acoustic context are retained. Experiments on ten‑minute LongSpeech sessions show a 64.6% reduction in peak streaming KV‑cache size and improved transcription, temporal question answering, and summarization, with comparable pause‑handling, turn‑taking, and interruption performance in Full‑Duplex‑Bench.
arXiv:2607. 22389v1 Announce Type: cross Abstract: With the rapid adoption of long-context large language models (LLMs), the continuously growing KV cache during decoding has become the critical memory bottleneck.
OmniKVQuant introduces a training‑free framework for quantizing the key‑value (KV) cache of omni‑modal large language models (Omni‑LLMs) that process audio, video, and text simultaneously. The method addresses two identified problems—temporal key drift and heterogeneous value geometry—by setting key quantization ranges over short input windows and rotating values separately for each modality. Applied to Qwen2.5‑Omni and Qwen3‑Omni, OmniKVQuant achieves 2‑bit KV caches while largely preserving performance across seven audio‑visual benchmarks, and includes a fused Triton decode kernel that eliminates the need for a dense FP16 cache.
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:2606. 02964v1 Announce Type: cross Abstract: Large Language Model (LLM) inference relies on key-value (KV) caches to avoid redundant attention computation.
arXiv:2606. 09079v1 Announce Type: cross Abstract: Conventional LLMs keep the full KV cache loaded during decoding, causing a severe GPU memory bottleneck for ultra-long context serving.
arXiv:2608.21362v1 Announce Type: new Abstract: Transformer-based large language models (LLMs) incur high prefill latency because key-value (KV) tensors must be recomputed for each request. Existing...
arXiv:2608. 14191v1 Announce Type: new Abstract: The key-value (KV) cache stores information from past tokens and is a major memory bottleneck in long-context inference.
arXiv:2609.14507v1 Announce Type: cross Abstract: Single-GPU long-context inference with Mixture-of-Experts (MoE) models requires spilling the key-value cache (KVCache) to CPU memory. The spilled KV...
HeadWiseKV is a training‑free framework that compresses the residual global key–value caches of hybrid long‑context language models by assigning each physical KV head a static, multilevel history window. It formulates cache allocation as a restricted operational rate–distortion problem and uses the SeqCalib algorithm to generate per‑head residency policies that account for interactions across layers. In evaluations on four hybrid models, HeadWiseKV preserves near‑full‑KV quality while reducing peak device memory usage by 8.59% at a 112K context length and extending the largest verified context from 114K to 161K.