Multi-property Molecular Optimization on BPQ
97.13Success Rate (SR)DPO-LLM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DPO-LLMBackbone=LLaMA, Protocol=DPO2026.04 | 97.13 | 45 | 1.98 | |
| DPO-LLMBackbone=Mistral, Protocol=DPO2026.04 | 97.07 | 47 | 1.74 | |
| SFT-LLMBackbone=Mistral, Protocol=SFT2026.04 | 97.05 | 54 | 1.45 | |
| SFT-LLMBackbone=LLaMA, Protocol=SFT2026.04 | 94.55 | 50 | 1.55 | |
| LlaSmolBackbone=Mistral2026.04 | 86 | 58 | 0.84 | |
| Claude-3.5shot_count=5-shot2026.04 | 76.8 | 53 | 1.24 | |
| Mistralshot_count=5-shot2026.04 | 68.6 | 63 | 0.79 | |
| Claude-3.5shot_count=0-shot2026.04 | 56 | 62 | 0.86 | |
| GPT-4oshot_count=1-shot2026.04 | 40 | 75 | 0.41 | |
| GPT-4oshot_count=0-shot2026.04 | 36.4 | 73 | 0.42 | |
| LLaMAshot_count=5-shot2026.04 | 34.6 | 70 | 0.64 | |
| LLaMAshot_count=0-shot2026.04 | 28.4 | 64 | 0.72 | |
| Mistralshot_count=0-shot2026.04 | 15.8 | 73 | 0.51 | |
| ChemLLM2026.04 | 4.8 | 29 | 0.96 |