Multimodal Understanding on MLLM Benchmark Suite
63.16GQA ScoreLLaVA (LoRA)
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| LLaVA (LoRA)Model Type=LLaVA, Tuning Method=LoRA2024.06 | 63.16 | 64.26 | 85.32 | 66.78 | 66.41 | 46.26 | 52.36 | 77.71 | 65.28 | |
| TG-SFT (Balance-Tree)Data Augmentation Strategy=Balance-Tree2024.06 | 62.79 | 64.35 | 85.13 | 68.02 | 65.11 | 45.44 | 52.45 | 77.02 | 65.04 | |
| Human (ShareGPT)Data Augmentation Strategy=Human (ShareGPT)2024.06 | 62.64 | 63.57 | 84.63 | 67.28 | 65.33 | 44.89 | 53.61 | 77.02 | 64.87 | |
| LLaVA (Full-Param)Model Type=LLaVA, Tuning Method=Full-Param2024.06 | 62.55 | 63.66 | 85.71 | 69.31 | 66.08 | 45.28 | 54.79 | 77.19 | 65.57 | |
| TG-SFT (Wide-Tree)Data Augmentation Strategy=Wide-Tree2024.06 | 62.26 | 64.6 | 84.69 | 68.47 | 65.53 | 45.35 | 52.41 | 77.04 | 65.04 |