Molecular Captioning on MolReasoner original (test)
0.461BLEU-2ReGuLaR
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| ReGuLaRBackbone=LLaMA-8B2026.01 | 0.461 | 0.3691 | 0.5479 | 0.6094 | 0.4428 | 0.5439 | 1 | |
| ReGuLaRBackbone=LLaMA-3B2026.01 | 0.4329 | 0.3394 | 0.5158 | 0.586 | 0.4164 | 0.5208 | 1 | |
| ReGuLaRBackbone=LLaMA-1B2026.01 | 0.3673 | 0.2692 | 0.4593 | 0.5319 | 0.3502 | 0.4635 | 1 | |
| ReGuLaRBackbone=DS-1.5B2026.01 | 0.3536 | 0.2545 | 0.4397 | 0.5187 | 0.3347 | 0.4514 | 1 | |
| CoTBackbone=LLaMA-8B2026.01 | 0.341 | 0.2378 | 0.4377 | 0.5138 | 0.3212 | 0.4403 | 295.736 | |
| ReGuLaRBackbone=LLaMA-8B, variant=w/o 2D2026.01 | 0.3403 | 0.2364 | 0.4394 | 0.5132 | 0.3187 | 0.4367 | 1 | |
| ReGuLaRBackbone=LLaMA-3B, variant=w/o 2D2026.01 | 0.3182 | 0.2119 | 0.4172 | 0.494 | 0.2961 | 0.4184 | 1 | |
| CoTBackbone=LLaMA-3B2026.01 | 0.3167 | 0.2146 | 0.4117 | 0.4948 | 0.3011 | 0.423 | 316.829 | |
| ReGuLaRBackbone=LLaMA-1B, variant=w/o 2D2026.01 | 0.2872 | 0.1845 | 0.3777 | 0.4678 | 0.2716 | 0.3962 | 1 | |
| CoTBackbone=LLaMA-1B2026.01 | 0.2828 | 0.1804 | 0.3778 | 0.4632 | 0.2675 | 0.3914 | 314.825 | |
| CoTBackbone=DS-1.5B2026.01 | 0.2682 | 0.1659 | 0.3637 | 0.4469 | 0.2523 | 0.3774 | 344.067 | |
| ReGuLaRBackbone=DS-1.5B, variant=w/o 2D2026.01 | 0.2672 | 0.164 | 0.3764 | 0.4645 | 0.2499 | 0.3476 | 1 | |
| CoLaRBackbone=LLaMA-8B2026.01 | 0.2031 | 0.1385 | 0.2803 | 0.3883 | 0.234 | 0.3449 | 163.655 | |
| CoLaRBackbone=LLaMA-3B2026.01 | 0.1734 | 0.0861 | 0.2487 | 0.3381 | 0.1576 | 0.284 | 200.267 | |
| GPT-4o2026.01 | 0.1093 | 0.0313 | 0.1682 | 0.2491 | 0.0778 | 0.1875 | — | |
| DeepSeek-R12026.01 | 0.1011 | 0.0297 | 0.2178 | 0.2506 | 0.0709 | 0.1722 | — | |
| CoLaRBackbone=LLaMA-1B2026.01 | 0.0995 | 0.04 | 0.1774 | 0.2306 | 0.0911 | 0.19 | 212.347 | |
| CoLaRBackbone=DS-1.5B2026.01 | 0.0968 | 0.0516 | 0.1644 | 0.2149 | 0.0916 | 0.1801 | 580.558 |