File size: 416 Bytes
f60a695 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 |
---
license: odc-by
---
# snimu/finemath-4plus-tiktokenized
This dataset contains GPT-2 tokenized shards of the FineWeb-4plus dataset.
Each shard is stored in a binary format with the following structure:
- First comes a header with 256 int32s
- The tokens follow, each as uint16 (GPT-2 format)
The first shard is the validation set, subsequent shards are training data.
Original dataset: HuggingFaceTB/finemath
|