Bongard Problem Solving on Bongard HOI
91.4Generative AccuracyHuman
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Human2025.07 | 91.4 | — | — | |
| LoRA (LNT)Model=Gemma3 4B¹2025.07 | 84.2 | 74.1 | 50 | |
| LoRA (Lcombined)Model=Gemma3 4B¹2025.07 | 84.2 | 74.1 | 83.2 | |
| LoRA (Lcombined)Model=Pixtral2025.07 | 79.6 | 63.1 | 77.8 | |
| LoRA (Lcombined)Model=Phi2025.07 | 79.2 | 71.9 | 82 | |
| LoRA (LNT)Model=Phi2025.07 | 78.6 | 71.9 | 63.6 | |
| LoRA (LNT)Model=Pixtral2025.07 | 78 | 61.6 | 74.9 | |
| Direct baselineModel=Pixtral2025.07 | 57.8 | 62.7 | 70.2 | |
| Direct baselineModel=Gemma3 4B¹2025.07 | 56.5 | 74.1 | 50 | |
| Direct baselineModel=Phi2025.07 | 52.1 | 71.9 | 60.5 |