Long-context Visual Question Answering on Visual HayStack Long Window (test)
68.17Accuracy (50 images)Base
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| BaseModel Parameters=32B2025.12 | 68.17 | 62.56 | 60.65 | — | |
| LAidModel Parameters=7B2025.12 | 67.04 | 63.37 | 60.17 | 24.5 | |
| SFT(LoRA)Model Parameters=7B2025.12 | 63.1 | 52.28 | 43.08 | 3.6 | |
| LAidModel Parameters=3B2025.12 | 58.2 | 53.91 | 50.23 | 16.4 | |
| YaRNModel Parameters=7B2025.12 | 56.26 | 47.96 | 42.36 | -4.7 | |
| BaseModel Parameters=7B2025.12 | 54.73 | 51.08 | 47.43 | — | |
| SFT(LoRA)Model Parameters=3B2025.12 | 52.34 | 48.01 | 42.18 | 1.96 | |
| YaRNModel Parameters=3B2025.12 | 52.2 | 45.07 | 39.97 | -1.9 | |
| SelfExtendModel Parameters=7B2025.12 | 50.35 | 45.12 | 40.12 | -11.7 | |
| BaseModel Parameters=3B2025.12 | 50.22 | 47.8 | 41.67 | — | |
| SelfExtendModel Parameters=3B2025.12 | 41.13 | 35.05 | 31.25 | -23.26 |