Multimodal Reasoning on MMMU (Acc, AvgLen, Ratio)
55.7AccuracyPrune-on-Logic
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Prune-on-LogicBackbone=Qwen3-VL, Evaluation Protocol=SFT2026.02 | 55.7 | 382.4 | 6.9 | |
| No CompressionBackbone=Qwen3-VL, Evaluation Protocol=SFT2026.02 | 55.2 | 558.9 | 10.1 | |
| XMCCBackbone=Qwen3-VL, Evaluation Protocol=SFT2026.02 | 54.8 | 107.3 | 2 | |
| StepEntropyBackbone=Qwen3-VL, Evaluation Protocol=SFT2026.02 | 54.4 | 270.5 | 5 | |
| No CompressionBackbone=Qwen2.5-VL, Evaluation Protocol=SFT2026.02 | 50.4 | 565.7 | 11.2 | |
| XMCCBackbone=Qwen2.5-VL, Evaluation Protocol=SFT2026.02 | 49.2 | 105.8 | 2.2 | |
| StepEntropyBackbone=Qwen2.5-VL, Evaluation Protocol=SFT2026.02 | 48.5 | 281.3 | 5.8 | |
| Prune-on-LogicBackbone=Qwen2.5-VL, Evaluation Protocol=SFT2026.02 | 48.4 | 414.9 | 8.6 |