rinnaJapanese GPT
Japanese GPT-2 Small
A Japanese GPT-2 completion model with a larger decoder than the XSmall release.
Research release
How this checkpoint was trained
The card records roughly 15 days of training on eight V100 GPUs using Japanese CC-100 and Wikipedia. It documents a tokenizer setting needed by the example loader.Japanese GPT-2 Small official model card
Keep preprocessing fixed when comparing sizes
Practical guidance. Use the same text normalization and tokenizer configuration for XSmall and Small. Otherwise a change in segmentation can look like a change in model quality.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | japanese-gpt2-small |
| Architecture | 12 layers with hidden width 768Japanese GPT-2 Small official model card |
| License | MIT.Japanese GPT-2 Small official model card |
Related models
- Japanese GPT-2 Medium
rinna · Japanese GPT
A Japanese GPT-2 language model trained to predict continuations of text.
Launch date not confirmedResearch releaseDownloadable - Japanese GPT-2 XSmall
rinna · Japanese GPT
A compact Japanese text-completion model trained on Japanese CC-100 and Wikipedia.
Launch date not confirmedResearch releaseDownloadable