Table 1.

Validation accuracy for pre-training runs with WordPiece tokenizer.

Nodes	Molecules	Batch size	Accuracy
1	2.4 · 106	1.4 · 103	0.760
10	2.4 · 107	1.4 · 104	0.783
100	2.4 · 108	1.4 · 105	0.797
1000	2.4 · 109	1.4 · 106	0.798
1008	9.6 · 109	1.5 · 106	0.808
4032	9.6 · 109	5.8 · 106	0.801