Data and Model Checkpoints
Abstract
These data include training, test, and validation splits used for models in Atanas et al, 2025. Also included are model checkpoints from training. For all files, decompress using zstd first, then extract from .tar to access internal files. Full decompressed size of dataset ~980GB.
Publication details
- Journal / source
- Harvard Dataverse
- Publication date
- 22 January 2026
- Document type
- Dataset