Multimodal Reasoning on SEED
32.52ECEMergeMix
Evaluation Results
| Method | Links | |
|---|---|---|
| MergeMixVision Token Count=Full, Training=ranking loss2025.10 | 32.52 | |
| SFT VisionVision Token Count=Full2025.10 | 32.67 | |
| CutMixVision Token Count=Full, Training=ranking loss2025.10 | 32.77 | |
| MixUpVision Token Count=Full, Training=ranking loss2025.10 | 33.26 | |
| MergeMixVision Token Count=50%, Training=ranking loss2025.10 | 33.37 | |
| MixUpVision Token Count=50%, Training=ranking loss2025.10 | 33.6 | |
| Baseline2025.10 | 33.79 | |
| CutMixVision Token Count=50%, Training=ranking loss2025.10 | 33.83 | |
| SFT VisionVision Token Count=50%2025.10 | 34.41 | |
| ResizeMixVision Token Count=Full, Training=ranking loss2025.10 | 36.08 | |
| ResizeMixVision Token Count=50%, Training=ranking loss2025.10 | 37.41 |