Literotica

Content warning: This dataset contains adult/NSFW fiction and may include explicit sexual content. You must be 18+ (or the age of majority in your jurisdiction) to use it. Please apply appropriate content filters and safeguards when training or serving models built on this data.

Dataset Summary

A large collection of user-submitted erotic short stories scraped from the public website Literotica.
The dataset (single default configuration) contains approximately ~460k records with story text and metadata (title, author, URL, rating, votes, category, tags). Files are stored as Parquet shards and exposed via the datasets library.

• Rows: ~460,000 (train only)
• Shards: data/train-00000-of-00030.parquet … data/train-00029-of-00030.parquet

This is a community upload intended for research on content moderation, NSFW detection, stylistic modeling, and general LM pretraining with safety controls. Respect the original authors and site policies.

Supported Tasks and Benchmarks

• Language Modeling / Next-Token Prediction — train causal LMs or adapters on long-form narrative text.
• Content/Topic Classification — use category and tags to build multi-label classifiers (e.g., safety taxonomies, topic tags).
• Rating/Vote Modeling — predict continuous ratings or engagement proxies (rating, votes).
• Retrieval / RAG — index text + metadata for similarity search and content filtering.

Languages

Primarily English (may contain other languages in small amounts; no language field is provided).

Dataset Structure

Data Instances

Each row represents one story.

{
  "title": "Example title",
  "author": "Example author or null",
  "url": "https://www.literotica.com/s/example-slug",
  "rating": 4.6,
  "votes": 1234.0,
  "category": "Romance",
  "text": "<string containing adult fiction>",
  "tags": ["tag1", "tag2"]
}

Note: Actual texts are explicit; sample above is redacted.

Data Fields

Field	Type	Description
title	string	Story title.
author	string	Author display name (may be null).
url	string	Source URL on literotica.com.
rating	float64	Site rating (<=4.0).
votes	float64	Number of votes (stored as float in source schema).
category	string	High-level category chosen on the site.
text	string	Full story text (NSFW).
tags	list[str]	User/site tags, if available.

Data Splits

Single split:

Split	Rows
train	~460,000

Usage

Load with datasets

from datasets import load_dataset

ds = load_dataset("timpal0l/literotica", split="train", trust_remote_code=False)
print(ds)
print(ds.features)

Basic filtering examples

# Highly rated stories
top = ds.filter(lambda ex: ex["rating"] is not None and ex["rating"] >= 4.5)

# Specific category
romance = ds.filter(lambda ex: ex["category"] == "Romance")

# Tag contains (case-insensitive)
def has_tag(ex, q="lesbian"):
    tags = ex["tags"] or []
    return any((t or "").lower() == q for t in tags)

filtered = ds.filter(has_tag)

Tokenization for LM training (streaming recommended)

from datasets import load_dataset
from transformers import AutoTokenizer

stream = load_dataset("timpal0l/literotica", split="train", streaming=True)
tok = AutoTokenizer.from_pretrained("gpt2", use_fast=True)

def batch_tokenize(batch):
    return tok(batch["text"], truncation=True, max_length=2048)

Consider deduplication, length filtering, and safety red-teaming before training.

Source Data

• Original source: User-submitted stories on literotica.com (community site).
• Provenance: This repo hosts a compiled dump in Parquet format; scraping/collection details may not be fully documented.

Annotations

No additional human annotations beyond site metadata (category, tags, ratings).

Licensing Information

License is set to other because the underlying works are authored by third parties on literotica.com.
Rights likely remain with the original authors and/or the platform. Before redistributing, training commercial models, or hosting outputs, review and comply with Literotica’s Terms of Use and copyright policies and seek permission where required. This repository does not claim ownership of the texts and shares them for research; downstream users are responsible for legal compliance.

Ethical Considerations & Risks

• NSFW & explicit content: Contains graphic sexual material; do not expose to minors or general audiences. Apply strict content moderation when using in demos or products.
• Consent & harm: Stories may depict sensitive themes. Avoid deploying models that generate unsafe or non-consensual content; implement safety classifiers, prompts, and guardrails.
• Bias & stereotypes: Tags/categories can encode social biases. Evaluate and mitigate representational harm.
• Attribution: Respect original authors; avoid implying endorsement.
• Data quality: Some fields may be missing (author, tags), and votes is stored as float in the schema.

Known Limitations

• Single split (train) only; no official dev/test.
• Text length varies widely; some stories are very long.
• Metadata is noisy/inconsistent; tags are user-generated.
• Language is primarily English but not guaranteed.

Maintenance

Owner: @timpal0l (uploader on the Hub). Use the Community tab on the dataset page for issues and discussions.

Citation

If you use this dataset, please cite both the dataset and the original platform:

@misc{timpal0l_literotica_2025,
  title        = {Literotica (community dataset)},
  author       = {timpal0l},
  howpublished = {Hugging Face Datasets},
  year         = {2025},
  url          = {https://huggingface.co/datasets/timpal0l/literotica}
}

Also acknowledge the original authors and literotica.com in your work.

Changelog

• Initial community upload with 30 Parquet shards (train split only).