Multimodal Math Reasoning on MathVista In-domain
65.3Overall AccuracyADPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ADPOSample size=122026.01 | 65.3 | 71.7 | 59.8 | |
| ADPOSample size=82026.01 | 65 | 72.2 | 58.9 | |
| ADPOSample size=42026.01 | 64.8 | 71.3 | 59.3 | |
| MM-VerifierSample size=122026.01 | 64.1 | 70.4 | 58.7 | |
| R1-VL-7BSample size=12026.01 | 63.5 | — | — | |
| GRPOSample size=42026.01 | 63.4 | 69.8 | 58 | |
| GRPOSample size=122026.01 | 63.4 | 70.7 | 57.2 | |
| GRPOSample size=82026.01 | 62.9 | 70.4 | 56.5 | |
| MM-VerifierSample size=82026.01 | 62.5 | 68.5 | 57.4 | |
| ADPOSample size=12026.01 | 62.4 | 68.7 | 57 | |
| GRPOSample size=12026.01 | 62.2 | 69.8 | 55.7 | |
| Base (Qwen2-VL-7B)Sample size=122026.01 | 60.7 | 67.4 | 55 | |
| Base (Qwen2-VL-7B)Sample size=82026.01 | 60.1 | 68 | 53.3 | |
| MM-VerifierSample size=42026.01 | 59.8 | 67 | 53.7 | |
| Base (Qwen2-VL-7B)Sample size=42026.01 | 58.2 | 65.7 | 51.9 | |
| Base (Qwen2-VL-7B)Sample size=12026.01 | 57.9 | 68.9 | 48.5 |