snimu's picture
Upload README.md with huggingface_hub
f60a695 verified
|
raw
history blame
416 Bytes
metadata
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