Multimodal Reasoning on MMBench EN V1.1 (test)
80.68AccuracyLASER-noQ
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LASER-noQModel Backbone=Qwen2-VL 7B, Compression Ratio (rho1)=90%2026.05 | 80.68 | — | |
| LASER-noQModel Backbone=LLaVA-Next 13B, Compression Ratio (rho1)=90%2026.05 | 67.58 | — | |
| LASER-noQModel Backbone=LLaVA-Next 7B, Compression Ratio (rho1)=90%2026.05 | 64.99 | — | |
| FP16Model Backbone=SmolVLM 2B2026.05 | 64.47 | 64.47 | |
| LASER-noQModel Backbone=SmolVLM 2B, Compression Ratio (rho1)=90%2026.05 | 64.31 | — | |
| LASER-noQModel Backbone=LLaVA-v1.5 7B, Compression Ratio (rho1)=90%2026.05 | 61.51 | — | |
| LASER-noQModel Backbone=SmolVLM 2B, Compression Ratio (rho1)=50%2026.05 | 61.09 | — | |
| LASER-noQModel Backbone=SmolVLM 2B2026.05 | — | 62.74 |