|
--- |
|
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 | |
|
|