FlashPrefill V2: Block-Sparse Prefill Attention for Long-Context LLM Serving

Hugging Face Daily Papers Papers

Summary

FlashPrefill V2 improves long-context LLM serving through mean-corrected sparse attention and optimized GPU operators, delivering substantial speedups over FlashAttention-2 and dense baselines.

Long-context modeling is a pivotal capability for Large Language Models, yet the quadratic complexity of attention remains a critical bottleneck, particularly during the compute-intensive prefilling phase. Our previous work, FlashPrefill, mitigates this cost through instantaneous pattern discovery and max-based dynamic thresholding; however, it remains an algorithmic prototype that is still distant from production deployment. In this paper, we present FlashPrefill V2, which evolves FlashPrefill from a prototype toward practical long-context serving along three dimensions. First, we introduce a mean correction term that effectively suppresses the approximation error, keeping performance degradation manageable even at extreme sparsity levels. Second, we redesign the sparse attention operator with PackGQA memory access, warp specialization, and pingpong pipelining, fully aligning with the latest FlashAttention-3/4 implementations and supporting FP8 inference to meet practical quantization requirements. Third, FlashPrefill V2 natively supports paged KV cache and continuous batching, allowing integration as an attention backend in modern inference frameworks such as SGLang. Extensive evaluations on NVIDIA H20 GPUs---among the most widely deployed inference accelerators---demonstrate that FlashPrefill V2 delivers up to 47.26x and 27.19x speedups over FlashAttention-2 at 128K context length under FP8 and BF16 precision, respectively, and, in FP8, still achieves a 30.49x speedup against an FA3/4-aligned dense baseline.
Original Article
View Cached Full Text

Cached at: 08/21/26, 08:09 AM

Paper page - FlashPrefill V2: Block-Sparse Prefill Attention for Long-Context LLM Serving

Source: https://huggingface.co/papers/2608.19758

Abstract

FlashPrefill V2 improves long-context serving via mean-corrected sparse attention, optimized GPU operators, and framework integration, achieving large speedups over dense baselines.

Long-context modelingis a pivotal capability for Large Language Models, yet the quadratic complexity of attention remains a critical bottleneck, particularly during the compute-intensive prefilling phase. Our previous work, FlashPrefill, mitigates this cost through instantaneous pattern discovery and max-based dynamic thresholding; however, it remains an algorithmic prototype that is still distant from production deployment. In this paper, we presentFlashPrefill V2, which evolves FlashPrefill from a prototype toward practical long-context serving along three dimensions. First, we introduce amean correctionterm that effectively suppresses the approximation error, keeping performance degradation manageable even at extreme sparsity levels. Second, we redesign thesparse attentionoperator withPackGQAmemory access,warp specialization, andpingpong pipelining, fully aligning with the latestFlashAttention-3/4implementations and supportingFP8 inferenceto meet practical quantization requirements. Third,FlashPrefill V2natively supportspaged KV cacheandcontinuous batching, allowing integration as an attention backend in modern inference frameworks such as SGLang. Extensive evaluations on NVIDIA H20 GPUs---among the most widely deployed inference accelerators---demonstrate thatFlashPrefill V2delivers up to 47.26x and 27.19x speedups over FlashAttention-2 at 128K context length under FP8 and BF16 precision, respectively, and, in FP8, still achieves a 30.49x speedup against an FA3/4-aligned dense baseline.

View arXiv pageView PDFGitHub1Add to collection

Get this paper in your agent:

hf papers read 2608\.19758

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2608.19758 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2608.19758 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2608.19758 in a Space README.md to link it from this page.

Collections including this paper1

Similar Articles

SparDA: Sparse Decoupled Attention for Efficient Long-Context LLM Inference

arXiv cs.CL

SparDA proposes a decoupled sparse attention architecture that adds a lightweight 'Forecast' projection to predict future KV cache needs, enabling lookahead prefetching from CPU to GPU and reducing selection overhead. On 8B sparse-pretrained models, it achieves up to 1.25× prefill and 1.7× decode speedup, with up to 5.3× higher decode throughput over non-offload baselines.