Hardware-Efficient Attention for Fast Decoding

• URL: http://arxiv.org/abs/2505.21487v1
• Date: Tue, 27 May 2025 17:54:07 GMT
• Title: Hardware-Efficient Attention for Fast Decoding
• Authors: Ted Zadouri, Hubert Strauss, Tri Dao,
• Abstract summary: Grouped Latent Attention (GLA) is a parallel-friendly latent attention paired with low-level optimizations for fast decoding.<n>Our optimized GLA kernel is up to 2$times$ faster than FlashMLA, for example, in a speculative decoding setting.
• Score: 13.958883001629644
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: LLM decoding is bottlenecked for large batches and long contexts by loading the key-value (KV) cache from high-bandwidth memory, which inflates per-token latency, while the sequential nature of decoding limits parallelism. We analyze the interplay among arithmetic intensity, parallelization, and model quality and question whether current architectures fully exploit modern hardware. This work redesigns attention to perform more computation per byte loaded from memory to maximize hardware efficiency without trading off parallel scalability. We first propose Grouped-Tied Attention (GTA), a simple variant that combines and reuses key and value states, reducing memory transfers without compromising model quality. We then introduce Grouped Latent Attention (GLA), a parallel-friendly latent attention paired with low-level optimizations for fast decoding while maintaining high model quality. Experiments show that GTA matches Grouped-Query Attention (GQA) quality while using roughly half the KV cache and that GLA matches Multi-head Latent Attention (MLA) and is easier to shard. Our optimized GLA kernel is up to 2$\times$ faster than FlashMLA, for example, in a speculative decoding setting when the query length exceeds one. Furthermore, by fetching a smaller KV cache per device, GLA reduces end-to-end latency and increases throughput in online serving benchmarks by up to 2$\times$.

Related papers

• Nexus:Proactive Intra-GPU Disaggregation of Prefill and Decode in LLM Serving [4.309392302169281]
  Engine-level prefill-decode (PD) disaggregation avoids interference but incurs higher hardware and coordination overhead.<n>PD achieves up to 2.2x higher throughput, 20x lower TTFT, and 2.5x lower TBT than vLLM; outperforms SG by up to 2x; and matches or exceeds disaggregated vLLM.
  arXiv  Detail & Related papers  (2025-07-09T07:27:18Z)
• Decoder-Hybrid-Decoder Architecture for Efficient Reasoning with Long Generation [129.45368843861917]
  We introduce the Gated Memory Unit (GMU), a simple yet effective mechanism for efficient memory sharing across layers.<n>We apply it to create SambaY, a decoder-hybrid-decoder architecture that incorporates GMUs to share memory readout states from a Samba-based self-decoder.
  arXiv  Detail & Related papers  (2025-07-09T07:27:00Z)
• HGCA: Hybrid GPU-CPU Attention for Long Context LLM Inference [8.826966369389893]
  We present HGCA, a hybrid CPU- GPU attention mechanism for large language models.<n>We show that HGCA achieves superior scalability, supports longer sequences and larger batch sizes, and outperforms existing sparse attention baselines in both performance and accuracy.<n> Experiments across diverse models and workloads show that HGCA achieves superior scalability, supports longer sequences and larger batch sizes, and outperforms existing sparse attention baselines in both performance and accuracy.
  arXiv  Detail & Related papers  (2025-07-03T20:20:33Z)
• Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding [51.711605076319216]
  Diffusion-based large language models (Diffusion LLMs) have shown promise for non-autoregressive text generation with parallel decoding capabilities.<n>We introduce a novel block-wise approximate KV Cache mechanism tailored for bidirectional diffusion models, enabling cache reuse with negligible performance drop.<n>We propose a confidence-aware parallel decoding strategy that selectively decodes tokens exceeding a confidence threshold, mitigating dependency violations and maintaining generation quality.
  arXiv  Detail & Related papers  (2025-05-28T17:39:15Z)
• FastCar: Cache Attentive Replay for Fast Auto-Regressive Video Generation on the Edge [60.000984252907195]
  Auto-regressive (AR) models have recently shown promise in visual generation tasks due to their superior sampling efficiency.<n>Video generation requires a substantially larger number of tokens to produce coherent temporal frames, resulting in significant overhead during the decoding phase.<n>We propose the textbfFastCar framework to accelerate the decode phase for the AR video generation by exploring the temporal redundancy.
  arXiv  Detail & Related papers  (2025-05-17T05:00:39Z)
• Oaken: Fast and Efficient LLM Serving with Online-Offline Hybrid KV Cache Quantization [17.202495171443932]
  We propose Oaken, an acceleration solution that achieves high accuracy and high performance simultaneously.<n>Oaken employs an online-offline hybrid approach, setting offline thresholds, which are then used to determine the quantization scale online.<n>Our experiments show that for a batch size of 256, Oaken achieves up to 1.58x throughput improvement over A100 GPU, incurring a minimal accuracy loss of only 0.54% on average.
  arXiv  Detail & Related papers  (2025-03-24T11:56:50Z)
• LServe: Efficient Long-sequence LLM Serving with Unified Sparse Attention [26.54297116028556]
  Large language models (LLMs) have shown remarkable potential in processing long sequences and complex reasoning tasks.<n>We introduce LServe, an efficient system that accelerates long-sequence LLM serving via hybrid sparse attention.<n>On average, LServe accelerates LLM prefilling by up to 2.9x and decoding by 1.3-2.1x over vLLM.
  arXiv  Detail & Related papers  (2025-02-20T18:59:52Z)
• QuantSpec: Self-Speculative Decoding with Hierarchical Quantized KV Cache [67.84112700032007]
  Large Language Models (LLMs) are increasingly being deployed on edge devices for long-context settings.<n>In these scenarios, the Key-Value ( KV) cache is the primary bottleneck in terms of both GPU memory and latency.<n>We propose a novel self-speculative decoding framework, QuantSpec, where the draft model shares the architecture of the target model but employs a hierarchical 4-bit quantized KV cache and 4-bit quantized weights for acceleration.
  arXiv  Detail & Related papers  (2025-02-05T20:43:48Z)
• ThinK: Thinner Key Cache by Query-Driven Pruning [63.13363917871414]
  Large Language Models (LLMs) have revolutionized the field of natural language processing, achieving unprecedented performance across a variety of applications.<n>This paper focuses on the long-context scenario, addressing the inefficiencies in KV cache memory consumption during inference.<n>We propose ThinK, a novel query-dependent KV cache pruning method designed to minimize attention weight loss while selectively pruning the least significant channels.
  arXiv  Detail & Related papers  (2024-07-30T17:59:08Z)
• Hardware-Aware Parallel Prompt Decoding for Memory-Efficient Acceleration of LLM Inference [19.167604927651073]
  Auto-regressive decoding of Large Language Models (LLMs) results in significant overheads in their hardware performance. We propose a novel parallel prompt decoding that requires only $0.0002$% trainable parameters, enabling efficient training on a single A100-40GB GPU in just 16 hours. Our approach demonstrates up to 2.49$times$ speedup and maintains a minimal memory overhead of just $0.0004$%.
  arXiv  Detail & Related papers  (2024-05-28T22:19:30Z)
• Break the Sequential Dependency of LLM Inference Using Lookahead Decoding [27.87483106859749]
  Lookahead decoding is an exact, parallel decoding algorithm for large language models (LLMs) Our implementation can speed up autoregressive decoding by up to 1.8x on MT-bench and 4x with strong scaling on multiple GPUs in code completion tasks.
  arXiv  Detail & Related papers  (2024-02-03T06:37:50Z)
• Fast Chain-of-Thought: A Glance of Future from Parallel Decoding Leads to Answers Faster [61.83949316226113]
  FastCoT is a model-agnostic framework based on parallel decoding. We show that FastCoT saves inference time by nearly 20% with only a negligible performance drop compared to the regular approach.
  arXiv  Detail & Related papers  (2023-11-14T15:56:18Z)
• FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness [80.3586155104237]
  FlashAttention is an IO-aware exact attention algorithm for Transformers. It reduces the number of memory reads/writes between GPU high bandwidth memory (HBM) and GPU on-chip. FlashAttention and block-sparse FlashAttention enable longer context in Transformers.
  arXiv  Detail & Related papers  (2022-05-27T17:53:09Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.