SparsePR Benchmarks: Density, PSNR, SSIM, and Speedup

Table 1

Quality and efficiency

Reference fidelity, task quality, executed-pair density, attention PFLOPs, and end-to-end speedup.

Model	Method	PSNR ↑	SSIM ↑	LPIPS ↓	ImgQual ↑	SubCons ↑	PBench ↑	Density ↓	PFLOPs ↓	E2E ↑
HunyuanVideo-13B	Dense	–	–	–	0.850	0.976	–	100.0%	612.38	1.00×
HunyuanVideo-13B	SpargeAttn†	25.31	0.832	0.217	0.763	0.928	–	40.19%	389.76	1.38×
HunyuanVideo-13B	SVG2†	29.87	0.907	0.121	0.850	0.927	–	25.45%	299.02	2.30×
HunyuanVideo-13B	SVOO†	24.87	0.843	0.224	0.679	0.976	–	33.26%	345.81	2.17×
HunyuanVideo-13B	SVG-EAR†	30.54	0.918	0.098	0.845	0.903	–	22.17%	281.86	1.93×
HunyuanVideo-13B	SparsePR	31.84	0.932	0.087	0.850	0.976	–	21.92%	255.95	2.61×
Wan2.2-I2V-A14B	Dense	–	–	–	0.689	0.974	–	100.0%	658.46	1.00×
Wan2.2-I2V-A14B	SpargeAttn†	26.74	0.871	0.116	0.680	0.953	–	30.15%	396.83	1.58×
Wan2.2-I2V-A14B	SVG2†	28.18	0.878	0.105	0.668	0.970	–	31.28%	393.95	1.59×
Wan2.2-I2V-A14B	SVOO†	29.67	0.913	0.095	0.689	0.973	–	31.67%	389.62	1.61×
Wan2.2-I2V-A14B	SVG-EAR†	29.75	0.921	0.086	0.680	0.970	–	23.64%	378.88	1.61×
Wan2.2-I2V-A14B	SparsePR	30.66	0.907	0.044	0.687	0.973	–	21.97%	336.55	1.80×
Cosmos-Predict2.5-14B	Dense	–	–	–	0.714	0.976	77.76	100.0%	526.87	1.00×
Cosmos-Predict2.5-14B	SVG2	20.07	0.624	0.330	0.678	0.896	76.14	28.81%	286.51	1.24×
Cosmos-Predict2.5-14B	SVOO	22.06	0.685	0.289	0.701	0.909	76.03	37.63%	315.38	1.03×
Cosmos-Predict2.5-14B	SVG-EAR	25.54	0.908	0.062	0.710	0.976	77.76	29.75%	289.69	1.10×
Cosmos-Predict2.5-14B	SparsePR	26.33	0.942	0.068	0.714	0.976	77.75	22.14%	253.61	1.51×
Cosmos3-Nano-16B	Dense	–	–	–	0.700	0.950	77.31	100.0%	127.61	1.00×
Cosmos3-Nano-16B	SVG2	22.45	0.735	0.216	0.677	0.915	75.03	37.29%	96.19	1.16×
Cosmos3-Nano-16B	SVOO	16.64	0.573	0.381	0.707	0.962	77.59	67.32%	108.01	1.02×
Cosmos3-Nano-16B	SVG-EAR	21.16	0.709	0.261	0.658	0.872	72.85	37.18%	96.14	1.10×
Cosmos3-Nano-16B	SparsePR	24.42	0.801	0.176	0.699	0.949	77.30	25.96%	81.79	1.48×

Table 2

Partitioning and reconstruction ablations

Mean and p99 normalized attention output error at 22% total executed-pair density. Lower is better.

Configuration	HunyuanVideo	Wan2.2	Cosmos-Predict2.5	Cosmos3-Nano
Semantic partition	0.089 / 0.714	0.164 / 1.734	0.791 / 7.532	0.359 / 3.356
Key-response K/V partition	0.085 / 0.812	0.156 / 1.659	0.769 / 7.270	0.341 / 3.174
Response-coupled partition	0.073 / 0.696	0.108 / 0.647	0.761 / 7.227	0.331 / 3.150
Semantic + probe repair	0.053 / 0.356	0.105 / 0.819	0.262 / 0.826	0.124 / 0.668
SparsePR	0.030 / 0.207	0.063 / 0.323	0.075 / 0.489	0.076 / 0.447