Multi-objective molecular optimization (BDP) on MuMOInstruct (seen instruction)
39.8Success Rate (SR)SFT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SFT2026.03 | 39.8 | 25.4 | 10.1 | |
| RePO2026.03 | 20.6 | 56.9 | 11.7 | |
| GRPOinitialization=None2026.03 | 15.6 | 75.9 | 11.8 | |
| GRPOinitialization=SFT init2026.03 | 8.8 | 14.1 | 1.2 | |
| Distill-SFT2026.03 | 7.8 | 20.7 | 1.6 | |
| Base Model2026.03 | 5.2 | 14.9 | 0.8 |