2D Vision Reasoning on CVBench 2D
58.5AccuracyLLaVA-1.5-7B
Evaluation results
| Method | Links | |
|---|---|---|
| LLaVA-1.5-7BImage Token=576, TFLOPS=10.12024.12 | 58.5 | |
| Dynamic-LLaVA-7B†Training-Free=false, Image Token=115 (-80%)2024.12 | 58.3 | |
| LLaVA-FastVTraining-Free=true, Image Token=144 (-75%)2024.12 | 56.7 | |
| LLaVA-1.5-7B+H2OTFLOPS=10.12024.12 | 49.7 |