List your company

Multi-modal Inference on Qwen3-VL-2B (inference)

2,514.9Total LatencyCIVIC
2,447.1962,904.1983,361.23,818.202May 27, 2026

Evaluation results

MethodLinks
CIVIC2026.052,514.932.530.1618.942,229.42,248.30.49407.944.61
Baseline2026.053,54347.820.3935.23,128.43,274.301,122.2122.7
ZOO-Prune2026.053,564.347.80.3827.23,141.23,168.418.45848.592.81
DyMU2026.053,688.296.740.434.043,224.23,258.29.281,122.2122.7
DynamicViTToken reduction strategy=hard shrink2026.053,785.352.980.3934.433,346.63,381.13.451,122.2122.7
DiffRate2026.053,804.9115.90.4233.953,311.23,345.11.261,122.2122.7
DynamicViTToken reduction strategy=soft gumbel2026.054,132.6186.50.3934.513,543.53,5782.621,122.2122.7
VisionTrim2026.054,207.547.710.3826.43,734.93,761.33.96848.592.81