- Dataset
- Token Count
- mC4 (ja)
- 159B
- Wikipedia (ja)
- 2B
- Pile (en)
- 25B
Specs
- DatasetToken Count
- mC4 (ja)159B
- Wikipedia (ja)2B
- Pile (en)25B
- Wikipedia (en)6B
- Stack (code)10B
- DatasetToken Count
- mC4 (ja)135.9B
- Wikipedia (ja)1.5B
- Pile (en)136.0B
- Wikipedia (en)5.1B
- Stack (code)136.0B
README
LLM-jp Corpus
This repository contains scripts to reproduce the LLM-jp corpus.
Number of Tokens
- v1.0.0
| Dataset | Token Count |
|---|---|
| mC4 (ja) | 159B |
| Wikipedia (ja) | 2B |
| Pile (en) | 25B |
| Wikipedia (en) | 6B |
| Stack (code) | 10B |
- v1.0.1
| Dataset | Token Count |
|---|---|
| mC4 (ja) | 135.9B |
| Wikipedia (ja) | 1.5B |
| Pile (en) | 136.0B |
| Wikipedia (en) | 5.1B |
| Stack (code) | 136.0B |
Data Preparation
In scripts, we provide scripts to download, filter, and tokenize the data.
License
The code in this repository is licensed under the Apache 2.0 license.
As for the dataset itself, refer to the licenses of the data subsets: