Real-world Multi-modal Question Answering on RWQA
74.77AccuracyFlash Linear Attention
Evaluation Results
| Method | Links | |
|---|---|---|
| Flash Linear AttentionModel scale=4B2026.06 | 74.77 | |
| Multiplication-Only Matrix Inversion ApproximationModel scale=4B2026.06 | 74.64 | |
| Flash Linear AttentionModel scale=9B2026.06 | 74.38 | |
| Multiplication-Only Matrix Inversion ApproximationModel scale=9B2026.06 | 74.12 | |
| MergeMixbaseline=SFT Vision2025.10 | 70.46 | |
| VisionThink-7B2025.10 | 69.28 | |
| Qwen2.5-VL-Ins-7B2025.10 | 68.63 | |
| SFT Vision2025.10 | 68.63 | |
| Multiplication-Only Matrix Inversion ApproximationModel scale=2B2026.06 | 65.62 | |
| Flash Linear AttentionModel scale=2B2026.06 | 65.23 | |
| Flash Linear AttentionModel scale=0.8B2026.06 | 62.35 | |
| Multiplication-Only Matrix Inversion ApproximationModel scale=0.8B2026.06 | 61.7 |