Skip to content

Dataset · nvidia

Massive NVIDIA biomolecular structure corpus

nvidia/RFD4-Proteina

Tens of millions of molecular structure records for training protein design models.

What's inside

The repository aggregates 10M to 100M tabular and text records with indexed molecular structures, including experimental PDB data. This is the complete training dataset from NVIDIA for the RFD4-Proteína models (flow models and biomolecular autoencoder) from October 2026.

What you can do with it

  • Pre-training generative models for protein and biomolecule design
  • Training molecular autoencoders on tabular and structural data
Size
10M–100M rows
Downloads this month
88.3K
Likes
49

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("nvidia/RFD4-Proteina")