List your company

Autoregressive Image Generation on VAR Models Efficiency Benchmark

12.88Memory (GB)HACK++
11.035223.487635.9448.3924Jun 6, 2026

Evaluation results

MethodLinks
HACK++Model=Infinity-2B, Attention Implementation=Standard Attention, Hardware=NVIDIA RTX PRO 6000 GPU (96 GB), Batch Size=12026.0612.882.20.081.540.651.51
HACKModel=Infinity-2B, Attention Implementation=Standard Attention, Hardware=NVIDIA RTX PRO 6000 GPU (96 GB), Batch Size=12026.0614.51.950.171.710.591.37
HACK++Model=HART, Attention Implementation=Standard Attention, Hardware=NVIDIA RTX PRO 6000 GPU (96 GB), Batch Size=12026.0617.661.760.030.991.011.6
HACKModel=HART, Attention Implementation=Standard Attention, Hardware=NVIDIA RTX PRO 6000 GPU (96 GB), Batch Size=12026.0620.671.50.191.160.861.37
ScaleKVModel=Infinity-2B, Attention Implementation=Standard Attention, Hardware=NVIDIA RTX PRO 6000 GPU (96 GB), Batch Size=12026.0621.371.320.372.090.481.12
FullModel=Infinity-2B, Attention Implementation=Standard Attention, Hardware=NVIDIA RTX PRO 6000 GPU (96 GB), Batch Size=12026.0628.281—2.350.431
HACK++Model=Infinity-8B, Attention Implementation=Standard Attention, Hardware=NVIDIA RTX PRO 6000 GPU (96 GB), Batch Size=12026.0628.892.040.143.110.321.52
FullModel=HART, Attention Implementation=Standard Attention, Hardware=NVIDIA RTX PRO 6000 GPU (96 GB), Batch Size=12026.0631.021—1.60.631
HACKModel=Infinity-8B, Attention Implementation=Standard Attention, Hardware=NVIDIA RTX PRO 6000 GPU (96 GB), Batch Size=12026.0632.341.820.333.450.291.38
ScaleKVModel=Infinity-8B, Attention Implementation=Standard Attention, Hardware=NVIDIA RTX PRO 6000 GPU (96 GB), Batch Size=12026.0643.881.340.824.290.231.1
FullModel=Infinity-8B, Attention Implementation=Standard Attention, Hardware=NVIDIA RTX PRO 6000 GPU (96 GB), Batch Size=12026.06591—4.880.211