Knowledge Retention on Knowledge Retention Evaluation Suite
82.54MMEVanilla
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| VanillaBackbone=Qwen2.5-VL (7B)2025.10 | 82.54 | 79.81 | 69.61 | 71.03 | 72.1 | 58.6 | 78.46 | 61.25 | 69.7 | 25.69 | 86.51 | 47.42 | 66.89 | |
| ReplayBackbone=Qwen2.5-VL (7B)2025.10 | 75.38 | 81.7 | 69.16 | 69.17 | 85.12 | 45.4 | 70.2 | 50.72 | 63.9 | 21.58 | 87.49 | 47.48 | 63.94 | |
| LoRABackbone=Qwen2.5-VL (7B)2025.10 | 67.88 | 37.2 | 59.29 | 69.79 | 42.3 | 2.4 | 21.39 | 23.25 | 39.4 | 13.52 | 73.73 | 9.02 | 38.16 | |
| VanillaBackbone=LLaVA-v1.5 (13B)2025.10 | 65.33 | 68.38 | 42.25 | 59.99 | 73.9 | 31.5 | 66.04 | 33.93 | 27.4 | 11.88 | 87.07 | 26.46 | 49.51 | |
| ReplayBackbone=LLaVA-v1.5 (13B)2025.10 | 57.49 | 65.81 | 40.27 | 54.75 | 70.94 | 25.9 | 61.04 | 24.62 | 27 | 12.11 | 87.09 | 21.23 | 45.69 | |
| KOREBackbone=LLaVA-v1.5 (13B)2025.10 | 55.99 | 62.71 | 40.32 | 51.6 | 71.97 | 30.8 | 65.1 | 26.84 | 27.3 | 13.32 | 79.29 | 18.91 | 45.35 | |
| KOREBackbone=Qwen2.5-VL (7B)2025.10 | 36.23 | 76.98 | 66.8 | 68.69 | 85.55 | 45.4 | 70.51 | 45.02 | 63.1 | 24.34 | 75.24 | 41.89 | 58.31 | |
| LoRABackbone=LLaVA-v1.5 (13B)2025.10 | 30 | 60.57 | 36.93 | 28.22 | 69.13 | 18.3 | 23.26 | 17.43 | 23.9 | 7.73 | 71.64 | 4.52 | 32.64 |