| | --- |
| | language: |
| | - en |
| | pretty_name: SlimPajama_300B |
| | --- |
| | |
| | The SlimPajama_300B is a 300B token sample of de-duplicated Slim Pajama dataset tokenized using the [EleutherAI/gpt-neox-20b](https://huggingface.co/EleutherAI/gpt-neox-20b) tokenizer |
| | |
| | Due to file size constraints, C4 and CommonCrawl has been uploaded in multiple chunks, you can use the following commands to merge them back into a single file: |
| | ```bash |
| | cat C4_part_* > C4.bin |
| | cat CommonCrawl_part_* > CommonCrawl.bin |
| | ``` |
| | |
| | #### Data Distribution |
| | |
| | | Data source | Composition | |
| | | ------------- | ------------------------------- | |
| | | Commoncrawl | 0.5208 | |
| | | C4 | 0.2668 | |
| | | GitHub | 0.0522 | |
| | | Books | 0.0420 | |
| | | ArXiv | 0.0442 | |
| | | Wikpedia | 0.0399 | |
| | | StackExchange | 0.0337 | |
| | |
