Music Preference Prediction on CMI-RewardBench (Music Arena)
73.88Pairwise AccuracySongEval-RM
Evaluation Results
| Method | Links | |
|---|---|---|
| SongEval-RMTrain Inputs=A, Train Data=5-axis MOS, Encoder=MuQ-large2026.06 | 73.88 | |
| CMI-RMTrain Inputs=TLRA, Train Data=+110K pseudo, Encoder=MuQ-MuLan2026.06 | 73.43 | |
| TuneJury (T+A)Train Inputs=TA, Train Data=17.5K, Encoder=CLAP+MERT2026.06 | 71.94 | |
| TuneJury, -SE (T+A)Train Inputs=TA, Train Data=15K, Encoder=CLAP+MERT2026.06 | 71.49 | |
| TuneJury (A only)Train Inputs=TA, Train Data=17.5K, Encoder=CLAP+MERT, Inference Prompt=empty2026.06 | 70.07 | |
| TuneJury, MuQ (T+A)Train Inputs=TA, Train Data=17K, Encoder=MuQ-MuLan2026.06 | 70.04 | |
| TuneJury, -MP (T+A)Train Inputs=TA, Train Data=15.5K, Encoder=CLAP+MERT2026.06 | 70 | |
| TuneJury, -MA (T+A)Train Inputs=TA, Train Data=17K, Encoder=CLAP+MERT2026.06 | 69.1 | |
| MuQ-Eval-A1Train Inputs=A, Train Data=2-axis MOS, Encoder=MuQ-large2026.06 | 67.61 | |
| Audiobox-AestheticsTrain Inputs=A, Train Data=4-axis MOS2026.06 | 67.39 | |
| PAM scoreTrain Inputs=A, Mode=zero-shot, Encoder=MS-CLAP2026.06 | 67.09 |