SlideSparse: Fast and Flexible (2N-2):2N Structured Sparsity

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Shao, Hanyong, Hao, Yingbo, Song, Ting, Xia, Yan, Zhang, Di, Huang, Shaohan, Wu, Xun, Xu, Songchen, Xu, Le, Dong, Li, Chi, Zewen, Zou, Yi, Wei, Furu
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910042451607552
author Shao, Hanyong
Hao, Yingbo
Song, Ting
Xia, Yan
Zhang, Di
Huang, Shaohan
Wu, Xun
Xu, Songchen
Xu, Le
Dong, Li
Chi, Zewen
Zou, Yi
Wei, Furu
author_facet Shao, Hanyong
Hao, Yingbo
Song, Ting
Xia, Yan
Zhang, Di
Huang, Shaohan
Wu, Xun
Xu, Songchen
Xu, Le
Dong, Li
Chi, Zewen
Zou, Yi
Wei, Furu
contents NVIDIA's 2:4 Sparse Tensor Cores deliver 2x throughput but demand strict 50% pruning -- a ratio that collapses LLM reasoning accuracy (Qwen3: 54% to 15%). Milder $(2N-2):2N$ patterns (e.g., 6:8, 25% pruning) preserve accuracy yet receive no hardware support, falling back to dense execution without any benefit from sparsity. We present SlideSparse, the first system to unlock Sparse Tensor Core acceleration for the $(2N-2):2N$ model family on commodity GPUs. Our Sliding Window Decomposition reconstructs any $(2N-2):2N$ weight block into $N-1$ overlapping 2:4-compliant windows without any accuracy loss; Activation Lifting fuses the corresponding activation rearrangement into per-token quantization at near-zero cost. Integrated into vLLM, SlideSparse is evaluated across various GPUs (A100, H100, B200, RTX 4090, RTX 5080, DGX-spark), precisions (FP4, INT8, FP8, BF16, FP16), and model families (Llama, Qwen, BitNet). On compute-bound workloads, the measured speedup ratio (1.33x) approaches the theoretical upper-bound $N/(N-1)=4/3$ at 6:8 weight sparsity in Qwen2.5-7B, establishing $(2N-2):2N$ as a practical path to accuracy-preserving LLM acceleration. Code available at https://github.com/bcacdwk/vllmbench.
format Preprint
id arxiv_https___arxiv_org_abs_2603_05232
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle SlideSparse: Fast and Flexible (2N-2):2N Structured Sparsity
Shao, Hanyong
Hao, Yingbo
Song, Ting
Xia, Yan
Zhang, Di
Huang, Shaohan
Wu, Xun
Xu, Songchen
Xu, Le
Dong, Li
Chi, Zewen
Zou, Yi
Wei, Furu
Machine Learning
NVIDIA's 2:4 Sparse Tensor Cores deliver 2x throughput but demand strict 50% pruning -- a ratio that collapses LLM reasoning accuracy (Qwen3: 54% to 15%). Milder $(2N-2):2N$ patterns (e.g., 6:8, 25% pruning) preserve accuracy yet receive no hardware support, falling back to dense execution without any benefit from sparsity. We present SlideSparse, the first system to unlock Sparse Tensor Core acceleration for the $(2N-2):2N$ model family on commodity GPUs. Our Sliding Window Decomposition reconstructs any $(2N-2):2N$ weight block into $N-1$ overlapping 2:4-compliant windows without any accuracy loss; Activation Lifting fuses the corresponding activation rearrangement into per-token quantization at near-zero cost. Integrated into vLLM, SlideSparse is evaluated across various GPUs (A100, H100, B200, RTX 4090, RTX 5080, DGX-spark), precisions (FP4, INT8, FP8, BF16, FP16), and model families (Llama, Qwen, BitNet). On compute-bound workloads, the measured speedup ratio (1.33x) approaches the theoretical upper-bound $N/(N-1)=4/3$ at 6:8 weight sparsity in Qwen2.5-7B, establishing $(2N-2):2N$ as a practical path to accuracy-preserving LLM acceleration. Code available at https://github.com/bcacdwk/vllmbench.
title SlideSparse: Fast and Flexible (2N-2):2N Structured Sparsity
topic Machine Learning
url https://arxiv.org/abs/2603.05232