Molecule Captioning on PubChem324k (test)
38.7BLEU-2MolCA, Galac1.3B
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| MolCA, Galac1.3BInput Modality=1D SMILES + 2D Graph, #Trainable params=100M, Fine-tuning protocol=LoRA ft*2023.10 | 38.7 | 30.3 | 50.2 | 35.9 | 44.5 | 45.6 | |
| MolCA, MolT5-LargeInput Modality=1D SMILES + 2D Graph, #Trainable params=877M, Fine-tuning protocol=full ft2023.10 | 32.9 | 26.3 | 49.8 | 35.7 | 44.2 | 42.4 | |
| MolCA, Galac125MInput Modality=1D SMILES + 2D Graph, #Trainable params=222M, Fine-tuning protocol=full ft2023.10 | 31.9 | 24.3 | 47.3 | 33.9 | 43.2 | 41.6 | |
| MoMu-LargeInput Modality=1D SMILES + 2D Graph, #Trainable params=782M, Fine-tuning protocol=full ft2023.10 | 31.1 | 22.8 | 41.8 | 25.7 | 36.7 | 36.2 | |
| MolT5-LargeInput Modality=1D SMILES, #Trainable params=780M, Fine-tuning protocol=full ft2023.10 | 30.2 | 22.2 | 41.5 | 25.9 | 34.8 | 36.6 | |
| MoMu-BaseInput Modality=1D SMILES + 2D Graph, #Trainable params=252M, Fine-tuning protocol=full ft2023.10 | 30.2 | 21.5 | 40.5 | 25.1 | 34.4 | 34.2 | |
| MolT5-BaseInput Modality=1D SMILES, #Trainable params=250M, Fine-tuning protocol=full ft2023.10 | 30.1 | 20.9 | 40.3 | 25.1 | 33.8 | 35.6 | |
| MoMu-SmallInput Modality=1D SMILES + 2D Graph, #Trainable params=82M, Fine-tuning protocol=full ft2023.10 | 19.1 | 12 | 29.7 | 16.3 | 26.7 | 21.8 | |
| MolT5-SmallInput Modality=1D SMILES, #Trainable params=80M, Fine-tuning protocol=full ft2023.10 | 14.8 | 8.5 | 26.5 | 13.5 | 23.6 | 18.5 |