Visual Question Answering on COCO-Caption
70.8TextVQAZeroshot
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| ZeroshotBackbone=NVILA-Lite 2B, Fine-tuned Layers=None, Update Ratio=N/A, Evaluation Protocol=Zero-shot2026.02 | 70.8 | 51.2 | 69.4 | 78.1 | 42.2 | 62.3 | 26.1 | 44.2 | 36.8 | |
| Model-DowserBackbone=NVILA-Lite 2B, Fine-tuned Layers=Last 20, Update Ratio=0.1, Evaluation Protocol=Data-Free Importance Probing2026.02 | 69.3 | 48.6 | 68.8 | 77.7 | 50.8 | 60.7 | 135.5 | 99.1 | 85.7 | |
| SPIDERBackbone=NVILA-Lite 2B, Fine-tuned Layers=Last 20, Update Ratio=0.1, Evaluation Protocol=Selective Parameter Update2026.02 | 65.3 | 42.3 | 67.8 | 72 | 48.4 | 59.6 | 115.4 | 87.3 | 78.3 | |
| GraftingBackbone=NVILA-Lite 2B, Fine-tuned Layers=Last 20, Update Ratio=0.1, Evaluation Protocol=Model Grafting2026.02 | 19.8 | 19.1 | 66.1 | 26.8 | 17 | 38.7 | 115.7 | 73.5 | 49.2 | |
| Full-FTBackbone=NVILA-Lite 2B, Fine-tuned Layers=Last 20, Update Ratio=0.1, Evaluation Protocol=Full Fine-tuning2026.02 | 8.1 | 11.5 | 65.1 | 25.1 | 15.5 | 24 | 98.5 | 61.7 | 39.7 | |
| DAREBackbone=NVILA-Lite 2B, Fine-tuned Layers=Last 20, Update Ratio=0.1, Evaluation Protocol=Drop & Rescale2026.02 | 8 | 11.3 | 64.1 | 24.7 | 14.2 | 24.9 | 96.8 | 60.6 | 39.1 | |
| TailorBackbone=NVILA-Lite 2B, Fine-tuned Layers=Last 20, Update Ratio=0.1, Evaluation Protocol=ModelTailor2026.02 | 7.7 | 12.5 | 64.2 | 40.7 | 25.9 | 18.9 | 105.6 | 67 | 44.7 |