Molecule Generation on QM9 (Validity and Uniqueness Checks)
95.26ValidityJointly trained ZATOM-1-L
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Jointly trained ZATOM-1-L# Params=160M, Training stages=One-stage training, Training data configuration=Jointly trained2026.02 | 95.26 | 96.84 | |
| QM9-only PLATOM-1# Params=23M, Training stages=One-stage training, Training data configuration=QM9-only2026.02 | 95.2 | 98.13 | |
| Jointly trained ZATOM-1-XL# Params=300M, Training stages=One-stage training, Training data configuration=Jointly trained2026.02 | 95.19 | 97.1 | |
| Jointly trained ZATOM-1# Params=80M, Training stages=One-stage training, Training data configuration=Jointly trained2026.02 | 94.94 | 97.16 | |
| Jointly trained ADiT# Params=180M, Training stages=Two-stage training, Training data configuration=Jointly trained2026.02 | 94.45 | 97.82 | |
| GeoLDM# Params=20M, Training stages=Two-stage training2026.02 | 93.8 | 98.82 | |
| QM9-only ZATOM-1# Params=80M, Training stages=One-stage training, Training data configuration=QM9-only2026.02 | 92.88 | 97.71 | |
| QM9-only ADiT# Params=180M, Training stages=Two-stage training, Training data configuration=QM9-only2026.02 | 92.19 | 97.9 | |
| Equivariant Diffusion# Params=20M, Training stages=One-stage training2026.02 | 91.9 | 98.69 | |
| Symphony# Params=2M, Training stages=One-stage training2026.02 | 83.5 | 97.98 |