Visual Mathematical Reasoning on MathVerse (Accuracy & Efficiency)
62.4AccuracyMt(Skywork-R1V-38B)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Mt(Skywork-R1V-38B)2026.06 | 62.4 | 101.4 | — | |
| JRL-trained M_d_joint (7B)Draft (1 - pt)%=20%2026.06 | 61.2 | 82.1 | 1.24 | |
| JRL-trained M_d_joint (7B)Draft (1 - pt)%=25%2026.06 | 60.77 | 76.6 | 1.32 | |
| JRL-trained M_d_joint (7B)Draft (1 - pt)%=30%2026.06 | 60.4 | 71.5 | 1.42 | |
| DRL-trained M_d_dft (7B)Draft (1 - pt)%=20%2026.06 | 59.8 | 86.4 | 1.17 | |
| DRL-trained M_d_dft (7B)Draft (1 - pt)%=30%2026.06 | 58.7 | 78.1 | 1.3 | |
| JRL-trained M_d_joint (7B)Draft (1 - pt)%=50%2026.06 | 58.5 | 55.4 | 1.83 | |
| OursDraft (1 - pt)%=50%2026.06 | 58.5 | 55.4 | 1.83 | |
| DRL-trained M_d_dft (7B)Draft (1 - pt)%=40%2026.06 | 57.2 | 69.1 | 1.47 | |
| Token-probability-based (Paradigm 1)Draft (1 - pt)%=50%2026.06 | 56.9 | 56.3 | 1.8 | |
| RandomDraft (1 - pt)%=50%2026.06 | 56.4 | 56 | 1.81 | |
| Special-token-based (Paradigm 2)Draft (1 - pt)%=50%2026.06 | 56.3 | 55.6 | 1.82 | |
| DRL-trained M_d_dft (7B)Draft (1 - pt)%=100%2026.06 | 50 | 11.8 | — | |
| JRL-trained M_d_joint (7B)Draft (1 - pt)%=100%2026.06 | 49.9 | 10.9 | — |