Selective Prediction on A-OKVQA 60 (test)
0.972AUROCRidge LTSA Chart
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ridge LTSA ChartModel=Idefics3-8B-Llama3, Variant=PR-dim/Ridge Proximity, Baseline=true2026.06 | 0.972 | 0.132 | 0.864 | 0.108 | |
| Ridge LTSA ChartModel=SmolVLM-Instruct, Variant=PR-dim/Ridge Proximity2026.06 | 0.951 | 0.245 | 0.75 | 0.182 | |
| logP(x)Model=Idefics3-8B-Llama3, Variant=Sequence logP(x)2026.06 | 0.795 | 0.244 | 0.752 | 0.152 | |
| logP(x)Model=SmolVLM-Instruct, Variant=Sequence logP(x)2026.06 | 0.791 | 0.344 | 0.651 | 0.221 | |
| Topology LID-MLEModel=Idefics3-8B-Llama3, Variant=Neg LID-MLE2026.06 | 0.725 | 0.252 | 0.744 | 0.169 | |
| Topology LID-MLEModel=SmolVLM-Instruct, Variant=Neg LID-MLE2026.06 | 0.47 | 0.593 | 0.402 | 0.299 |