Datasets:

muhammadhasyim
/

equiv-dens-qm9

index int64 1 134k
1
2
3
5
6
7
8
9
10
11
12
14
15
16
17
18
20
21
22
24
25
29
30
31
33
37
39
40
41
42
43
44
45
46
47
48
49
51
52
53
54
55
56
57
58
59
61
63
64
66
67
68
69
70
72
73
74
75
76
77
79
80
81
83
84
86
88
90
91
92
93
94
95
96
98
99
100
101
102
103
104
105
107
108
109
113
114
115
116
117
119
120
121
122
123
124
125
127
130
132

End of preview. Expand in Data Studio

equiv-dens QM9

QM9 geometries, manifest, train/valid/test splits, and processed training files for the equiv_dens_ml electron-density ML project.

Path	Format	Description
`raw/dsgdb9nsd.xyz.tar.bz2`	bz2 archive	gdb9 QM9 xyz geometries (~134k molecules)
`qm9_manifest.npz`	NPZ	`positions`, `atom_numbers`, `mol_ids`
`manifest.parquet`	Parquet	Viewer-friendly manifest copy
`splits/{train,valid,test}_idx.npy`	int64	80/10/10 split indices
`processed/{split}_npy.npy`	pickled dict	Geometry dicts per split
`processed/{split}_dft_list_b3lyp_631gdp.npy`	pickled list	B3LYP/6-31G(2df,p) DFT labels (optional)

Usage with equiv_dens_ml

./scripts/download_hf_datasets.sh --group qm9
./scripts/qm9/setup_training_data.sh   # extracts xyz if needed
python run.py train --mode joint @config/training/qm9_joint_b3lyp_lazy_local.txt

Pickled .npy files require PySCF and are loaded via np.load(..., allow_pickle=True). Do not use load_dataset() for training labels.

Citation

@misc{equiv-dens-qm9,
  author = {Bogojeski, Mihail and Hasyim, Muhammad},
  title = {equiv-dens QM9 Training Data},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/muhammadhasyim/equiv-dens-qm9}},
  note = {Revision v1.0.0}
}

Downloads last month: 42

equiv-dens QM9

Contents

Usage with equiv_dens_ml

Citation