Visual Question Answering on ImageNet
60.5Normalized AccuracyTARA-Variant A
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TARA-Variant ABackbone=LLaVA, Adapter Rank=16, Optimization Strategy=gradient-based2026.03 | 60.5 | — | |
| TARA-Variant BBackbone=LLaVA, Adapter Rank=16, Optimization Strategy=gradient-based2026.03 | 59.6 | — | |
| TIESBackbone=LLaVA, Adapter Rank=16, Optimization Strategy=gradient-free2026.03 | 54.8 | — | |
| AdaMergingBackbone=LLaVA, Adapter Rank=16, Optimization Strategy=gradient-based2026.03 | 47 | — | |
| KnOTS-TIESBackbone=LLaVA, Adapter Rank=16, Optimization Strategy=gradient-free2026.03 | 44.8 | — | |
| TABackbone=LLaVA, Adapter Rank=16, Optimization Strategy=gradient-free2026.03 | 43.6 | — | |
| FinetunedBackbone=LLaVA, Adapter Rank=162026.03 | — | 96 |