| 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 | |