Multi-objective molecular optimization (BPQ) on MuMOInstruct unseen instruction
41.9Success Rate (SR)SFT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SFT2026.03 | 41.9 | 24.8 | 10.4 | |
| RePO2026.03 | 24.2 | 59.6 | 14.4 | |
| GRPOinitialization=None2026.03 | 18.6 | 57.3 | 10.7 | |
| Base Model2026.03 | 5 | 13 | 0.7 | |
| Distill-SFT2026.03 | 5 | 14.3 | 0.7 | |
| GRPOinitialization=SFT init2026.03 | 4.2 | 6.3 | 0.3 |