Skip to content

Dataset · datamol-io

1.17 billion molecules in SMILES and SAFE formats

datamol-io/safe-gpt

A massive corpus of chemical molecules in text representation for pre-training generative models.

What's inside

Contains about 1.17 billion unique molecules, each recorded in two text representations simultaneously: canonical SMILES and SAFE. This is the second version of the dataset, designed for large-scale pre-training and evaluation of chemistry language models.

What you can do with it

  • Pre-training chemical language models (LLMs) from scratch
  • Generating new molecular structures with target properties
  • Evaluating models on molecular structural analysis tasks
Tasks
Text generation
Languages
English
Size
1B–10B rows
Downloads this month
2.1K
Likes
6

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("datamol-io/safe-gpt")