|
--- |
|
language: ja |
|
tags: |
|
- t5 |
|
- text2text-generation |
|
- seq2seq |
|
license: apache-2.0 |
|
datasets: |
|
- mc4 |
|
- wiki40b |
|
--- |
|
|
|
# t5-base-japanese-web (with Byte-fallback) |
|
|
|
## Description |
|
|
|
[megagonlabs/t5-base-japanese-web](https://huggingface.co/megagonlabs/t5-base-japanese-web) is a T5 (Text-to-Text Transfer Transformer) model pre-trained on Japanese web texts. |
|
Training codes are [available on GitHub](https://github.com/megagonlabs/t5-japanese). |
|
|
|
### Corpus |
|
|
|
- Japanese in [mC4/3.0.1](https://github.com/allenai/allennlp/discussions/5056) |
|
- [Japanese](https://www.tensorflow.org/datasets/catalog/wiki40b#wiki40bja) in [wiki40b/1.3.0](https://www.tensorflow.org/datasets/catalog/wiki40b) |
|
|
|
|
|
### Tokenizer |
|
|
|
SentencePiece trained on Japanese Wikipedia |
|
|
|
- Vocabulary size: 32,000 |
|
- [Byte-fallback](https://github.com/google/sentencepiece/releases/tag/v0.1.9): Enabled |
|
|
|
### Parameters |
|
|
|
- T5 model: [models/t5.1.1.base.gin](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/t5/models/gin/models/t5.1.1.base.gin) |
|
- Training steps: 1,000,000 |
|
|
|
It took about 126 hours with TPU v3-8 |
|
|
|
## Related models |
|
|
|
- [日本語T5事前学習済みモデル (sonoisa/t5-base-japanese)](https://huggingface.co/sonoisa/t5-base-japanese) |
|
- [日本語T5事前学習済みモデル (sonoisa/t5-base-japanese-mC4-Wikipedia)](https://huggingface.co/sonoisa/t5-base-japanese-mC4-Wikipedia) |
|
|
|
## License |
|
|
|
Apache License 2.0 |
|
|
