Multi-Property Molecular Optimization (QED+SA) on ZINC 250k
64.5Success Rate (SR)MolMem
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MolMemBackbone=Qwen2.5-1.5B, Optimization Paradigm=Expert2026.04 | 64.5 | 47 | 24 | |
| PEIT-LLMBackbone=Llama3.1-8B, Optimization Paradigm=Apprentice, Optimization Strategy=Task-Specific LLM2026.04 | 41.5 | 44 | 26 | |
| Direct RetrievalBackbone=N/A, Optimization Paradigm=Apprentice, Optimization Strategy=Retrieval-based2026.04 | 31 | 43 | 25 | |
| Direct PromptBackbone=Qwen2.5-1.5B, Optimization Paradigm=Apprentice, Optimization Strategy=Prompting & SFT2026.04 | 24 | 49 | 22 | |
| GeLLM3OBackbone=Llama3.1-8B, Optimization Paradigm=Apprentice, Optimization Strategy=Task-Specific LLM2026.04 | 22.5 | 57 | 14 | |
| Direct PromptBackbone=Qwen2.5-3B, Optimization Paradigm=Apprentice, Optimization Strategy=Prompting & SFT2026.04 | 21.5 | 55 | 16 | |
| SFT-onlyBackbone=Qwen2.5-3B, Optimization Paradigm=Apprentice, Optimization Strategy=Prompting & SFT2026.04 | 21 | 60 | 14 | |
| LlaSMolBackbone=Mistral-7B, Optimization Paradigm=Apprentice, Optimization Strategy=Task-Specific LLM2026.04 | 21 | 48 | 19 | |
| ChemLLMBackbone=ChemLLM-7B, Optimization Paradigm=Apprentice, Optimization Strategy=Task-Specific LLM2026.04 | 20.5 | 51 | 21 | |
| SFT-onlyBackbone=Qwen2.5-1.5B, Optimization Paradigm=Apprentice, Optimization Strategy=Prompting & SFT2026.04 | 20 | 59 | 13 | |
| Reinvent 4Backbone=Transformer, Optimization Paradigm=Novice2026.04 | 17 | 66 | 5 | |
| Graph-GABackbone=N/A, Optimization Paradigm=Novice2026.04 | 8 | 52 | 12 | |
| QMOBackbone=GRU, Optimization Paradigm=Novice2026.04 | 1.5 | 53 | 11 |