Multi-objective molecular optimization (BDP) on MuMOInstruct (unseen instruction)
31Success Rate (SR)SFT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SFT2026.03 | 31 | 26.1 | 8.1 | |
| RePO2026.03 | 19.8 | 57.2 | 11.3 | |
| GRPOinitialization=None2026.03 | 14.8 | 72.7 | 10.8 | |
| GRPOinitialization=SFT init2026.03 | 9.2 | 14.7 | 1.4 | |
| Base Model2026.03 | 5.2 | 14.3 | 0.7 | |
| Distill-SFT2026.03 | 1.6 | 9.9 | 0.2 |