UniPrefill is a prefill‑acceleration framework that works with virtually any model architecture, directly speeding up token‑level computation. It is implemented as a continuous‑batching operator and extends vLLM’s scheduling to support prefill‑decode co‑processing and tensor parallelism. The method delivers up to a 2.1× speedup in Time‑to‑First‑Token, with gains growing as concurrent requests increase.
By Qihang Fan, Huaibo Huang, Zhiying Wu, Bingning Wang, Ran He
arXiv:2506. 01969v3 Announce Type: replace-cross Abstract: Efficient inference of Multi-Head Latent Attention (MLA) is challenged by deploying the DeepSeek-R1 671B model on a single Multi-GPU server.
By Pengcuo Dege, Qiuming Luo, Rui Mao, Chang Kong
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×.
By Daniel Ohayon, Itay Lamprecht, Itay Hubara, Israel Cohen, Daniel Soudry, Noam Elata
arXiv:2508. 18224v3 Announce Type: replace-cross Abstract: Recent advances in sparse attention mechanisms have demonstrated strong potential for reducing the computational cost of long-context training and inference in large language models (LLMs).
By Ran Yan, Youhe Jiang, Zhuoming Chen, Haohui Mai, Beidi Chen, Binhang Yuan
The paper introduces Faster Flash Decoding (FFD), a hardware‑algorithm co‑design that fuses the selector and computation into a single kernel to eliminate memory‑bandwidth bottlenecks in long‑context decoding. By replacing metadata indices with low‑bit quantized, content‑aware scanning and employing a top‑delta strategy for dynamic block filtering, FFD achieves up to 11.6× kernel‑level speedup and scales to 256K context length while preserving model accuracy. The approach is training‑free, plug‑and‑play, and demonstrates significant throughput gains on benchmarks such as RULER and LongBench.
By Zhigeng Liu, Zhiyuan Ning, Ruixiao Li, Xiaoran Liu, Yuerong Song, Min Zhang, Ziwei He, Xipeng Qiu
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.
By Yan Wang, Qifan Zhang, Jiachen Yu, Tian Liang, Dongyang Ma, Xiang Hu, Zibo Lin, Chunyang Li, Zhichao Wang, Jia Li, Yujiu Yang, Haitao Mi, Dong Yu
FlashAttention-V is a blocked FlashAttention implementation optimized for scalable vector architectures, designed to reduce the memory bandwidth bottleneck of transformer attention on CPUs. By fusing operations, exploiting parallelism across attention heads, and inter‑head packing, it improves vector register utilization and memory locality, enabling efficient scaling from short to very long vectors. Benchmarks on TinyLlama, Llama 3.2, Qwen2.5, and Pythia‑410M show 22×–42× speedups over scalar FlashAttention in prefill and 8×–11× in decode on a Banana Pi BPI‑F3, while also revealing quantization‑related bottlenecks that limit long‑vector scalability.
By Sonia Rani Gupta, Nikela Papadopoulou, Miquel Peric\`as
arXiv:2507.19608v2 Announce Type: replace
Abstract: On-device LLM inference is increasingly important for edge platforms but remains challenging due to high computational and memory demands and tight...
By Jiawen Qi, Jiayu Liu, Chang Gao, Zhaochun Ren, Qinyu Chen
Elastic Threshold Attention (ETA) is a trainable attention mechanism that dynamically predicts contextual thresholds from query representations, enabling selective pruning of KV cache tokens during long‑context decoding. By multiplicatively suppressing sub‑threshold logits during training, ETA avoids representation collapse and eliminates localized attention sinks, allowing a 1.45B model to match dense attention performance at roughly 85% training sparsity and 38% active decode density. At inference, a custom Triton kernel achieves up to 2.5× faster decoding on sequences up to 512K tokens, and an offline calibration step can further reduce compute by 27% by freezing per‑head thresholds.
By Themistoklis Haris, Henry Li, Maryam Karimzadehgan
The paper "LLM Inference in a Flash!" proposes an integer‑only quantization scheme and a dictionary‑based KV cache compression technique to enable large language model inference on compute‑in‑flash (CIF) devices. By eliminating floating‑point operations and reducing KV cache traffic through sparse dictionary coding, the authors achieve minimal accuracy loss while cutting dynamic KV cache traffic by 15× on Llama‑3.1‑8B and Qwen‑2.5‑7B models.
By Sebastian Zhao, Minseo Kim, Coleman Hooper, Luca Manolache, Michael W. Mahoney, Yakun Sophia Shao, Kurt Keutzer, Amir Gholami
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.
By Matthias Seeger, Zeyu Zhang, Vihang Patil, Konstantinos Benidis, Sebastian Schelter
arXiv:2603. 29002v3 Announce Type: replace-cross Abstract: Modern large language models (LLMs) increasingly depends on efficient long-context processing and generation mechanisms, including sparse attention, retrieval-augmented generation (RAG), and compressed contextual memory, to support complex reasoning.
By Zifan He, Rui Ma, Yizhou Sun, Jason Cong