Multi-objective molecular optimization (BDQ) on MuMOInstruct seen instruction
31.9Success Rate (SR)SFT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SFT2026.03 | 31.9 | 27.9 | 8.9 | |
| RePO2026.03 | 16 | 36.5 | 5.8 | |
| GRPOinitialization=None2026.03 | 8.2 | 47.9 | 3.9 | |
| Base Model2026.03 | 3.4 | 11.7 | 0.4 | |
| Distill-SFT2026.03 | 2.2 | 10.6 | 0.2 | |
| GRPOinitialization=SFT init2026.03 | 2.2 | 4.5 | 0.1 |