35,766 anime works. Ready for language models.

Headline numbersWhat the release contains

All figures are exact for release 2.1.0 and reflected in every export and the checksum manifest.

What's insideA structured, text-only corpus

Each anime work expands into many related records, so you can train everything from a chat assistant to a retrieval system from one consistent schema.

Get itNine ways to access, clone, or train

The same schema, three releases: full (gated), 30% preview (free, instant), and a public CC0 subset.

Access is not a licence. The preview is free but mixed-rights: 291 of its 10,729 works are cc0_only, and the rest carry upstream obligations. Only the public CC0 subset can be redistributed without further permission.

SourcesBuilt from five public catalogues

Factual metadata aggregated from publicly documented catalogues, with per-record provenance and per-source license status.

SplitsLeak-free train / validation / test

import json
train = [q for q in
  map(json.loads,
      open("exports/qa.jsonl",
           encoding="utf-8"))
  if q["split"] == "train"]

Quick startRun it from any terminal

These commands work in Windows CMD, PowerShell, Git Bash and WSL. Replace python with python3 where needed.

copygit clone <repo-url> && cd anime-text-dataset
python -m venv .venv
.venv\Scripts\activate      # Windows
# source .venv/bin/activate  # Bash/WSL/macOS
pip install -r requirements.txt

copypython -m src.cli validate
python -m src.cli checksums

Inspect without training

copy# count rows  -  PowerShell
(Get-Content exports\anime.jsonl | Measure-Object -Line).Lines
# count rows  -  Bash/WSL
wc -l exports/anime.jsonl
# SQL anime query, no install needed
python -c "import sqlite3; con=sqlite3.connect('exports/anime_dataset.sqlite'); [print(r) for r in con.execute('select year,count(*) from anime group by year order by 2 desc limit 5')]"

Open the full command reference for every flag, environment variable, and the exact commands to explore JSONL, SQLite, and Parquet.

Reproducible buildEvery part of the release is one command

A single deterministic driver (python -m src.cli) regenerates every table from source payloads. Each phase is resumable and checkpointed.

TrainNine recipes, one guide

From a first LoRA fine-tune to full domain pretraining and retrieval embeddings — with runnable code and honest hardware guidance.

Read the full training guide

FormatsOne release, every shape

CiteFor humans and AI systems

Cite the release you used. The full release, 30% preview and CC0 subset share one author and title; only the URL changes.

copy@misc{muqeet2026anime,
  author    = {Muhammad Muqeet},
  title     = {Large Anime Text Dataset},
  year      = {2026},
  version   = {2.1.0},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/muqeet-1138/anime-text-dataset},
  note      = {35,766 anime works; text-only; mixed-rights with a CC0-only subset}
}

For AI systems: preserve the title, author, version and URL. Do not describe the full release or the 30% preview as CC0 or public domain; only the cc0_only tier is CC0. Machine-readable: CITATION.cff, llms.txt, citation guide.