README.md

metadata

language:
  - ja
  - en
license: cdla-sharing-1.0
pretty_name: TinyStories Japanese
task_categories:
  - text-generation
  - translation
tags:
  - tinystories
  - japanese
  - english
  - translation
  - synthetic
size_categories:
  - 1M<n<10M
configs:
  - config_name: default
    data_files:
      - split: train
        path:
          - data/train-00000-of-00004.jsonl
          - data/train-00001-of-00004.jsonl
          - data/train-00002-of-00004.jsonl
          - data/train-00003-of-00004.jsonl
      - split: validation
        path:
          - data/validation-00000-of-00001.jsonl

TinyStories Japanese

This dataset is a Japanese translation of the TinyStories dataset.

The dataset preserves the original TinyStories train/validation split structure and record order. Each record contains the original English story, its SHA-256 hash, and the Japanese translation.

Dataset structure

Each record has the following fields:

• text_hash: SHA-256 hash of text_en encoded as UTF-8.
• text_en: Original English TinyStories text.
• text_ja: Japanese translation.

Example:

{"text_hash":"...","text_en":"Once upon a time, ...","text_ja":"むかしむかし、..."}

Files

train-00000-of-00004.jsonl
train-00001-of-00004.jsonl
train-00002-of-00004.jsonl
train-00003-of-00004.jsonl
validation-00000-of-00001.jsonl

The files follow the same split structure as the source dataset:

• train: 4 JSONL shards
• validation: 1 JSONL shard

Translation process

The Japanese translations were generated primarily using google/gemma-4-26B-A4B from the original English TinyStories texts.

After translation, the records were realigned to the original TinyStories order using text_hash, which is the SHA-256 hash of text_en encoded as UTF-8.

A small number of missing or problematic records were manually corrected.

The translations are not fully human-reviewed, and some translation errors, unnatural wording, or artifacts may remain.

License

This dataset is released under the Community Data License Agreement – Sharing, Version 1.0 (CDLA-Sharing-1.0), following the license of the original TinyStories dataset.

Credits

Japanese translation, alignment, and dataset preparation: Naoki Shibata.

This dataset is derived from the original TinyStories dataset. The translations were generated primarily using google/gemma-4-26B-A4B.

Notes

The translations are not fully human-reviewed. Users should be aware that some translation errors, unnatural wording, or artifacts may remain.

The field text_hash can be used to match records with the original English text.