metadata

annotations_creators:
  - no-annotation
language_creators:
  - crowdsourced
license:
  - cc-by-sa-4.0
task_categories:
  - text-generation
  - fill-mask
task_ids:
  - language-modeling
  - masked-language-modeling
source_datasets:
  - original
language:
  - en
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/Fandom-v0.5.jsonl
  - config_name: raw-pre-roblox
    data_files:
      - split: train
        path: v2.5-chunks/*.jsonl
  - config_name: raw-post-roblox
    data_files:
      - split: train
        path: v2.5-chunks-roblox-filter/*.jsonl
pretty_name: Fanatic Fandom

Dataset Card for Fanatic Fandom

Waifu to catch your attention.

Dataset Details

Dataset Description

Fanatic Fandom is a cleaned dataset of a raw scrape of fandom wikis. We crawled all the publicly available wikis and crawled each page.
Filtering to a total amount of tokens of ~7.43B (llama-2-7b-chat-tokenizer) / ~6.27B (RWKV Tokenizer) from primarily English language.

Curated by: KaraKaraWitch
Funded by: Recursal.ai (I work there lol)
Shared by: KaraKaraWitch
Language(s) (NLP): Primarily English
License: cc-by-sa-4.0

Dataset Sources

Source Data: https://fandom.com/ (Bot Crawled.)

Processing and Filtering

We detail the following steps involved in scraping, indexing and cleaning fandom wikis to the html content files. Here's a breakdown of the process:

Wiki Identification:
- WikisIndexer.py script retrieves a list of wikis from https://community.fandom.com/Special:NewWikis.
Page Indexing:
- IndexFandomPages.py script utilizes the MediaWiki API (api.php) to gather a list of pages per each wiki.
Page Fetching:
- WikiPageFetcher.py script utilizes the MediaWiki API (api.php) to render the render the wiki page and save it to a large JSONL file.
- Additionally, any wikis with less than 5 pages are not scrapped due to assumed low-quality.
Data Chunking:
- A single large JSONL file containing all fetched pages is split into smaller, more manageable chunks.
- This is in preparation from the 4th step.
Roblox Wiki Removal:
- The RobloxWikiFilter.py script identifies and removes Roblox wikis due to the high volume of low-quality content they often generate. This filtering step simplifies the subsequent stub article removal process.
- From quick napkin math: around 15.2% (Comparing Step 3 and Step 4 results) of fandom wikis are Roblox data.
Content Transformation:
- HTML content is converted to Markdown format. The conversion process removes unnecessary elements like figures, stub article notices, and other irrelevant data.

Note: Due to the passage of time (approximately 3 months as of May 6, 2024), the specific details of the crawling process may be a little hazy. The primary challenge encountered was the significant time required to complete the crawling operation.

Data Splits

There are 3 splits for this dataset:

final
- Contains the final 25GB jsonl file.
- You probably want this for training.
raw-pre-roblox
- Raw files, before Roblox filtering.
- Use this if you want to start from scratch and don't want to crawl fandom again.
raw-post-roblox
- Raw files, after Roblox filtering.
- Roblox wikis removed.
- Use this if you want to start from scratch and don't want to crawl fandom again.

Data Keys

For this dataset, we have included most of the various steps for the dataset. They are listed as such below:

fandom_wikis_210224.csv
- A CSV file containing a list of wikis found when scrapping from Special:NewWikis on 21/02/2024
- The key is as follows: Sub Domain,Name of Wiki,Path name,0
- The stray zero can be ignored as it does not serve any purpose.
fandom_wikis_pages_210224_v2.jsonl
- Contains a jsonl list of wiki pages per each wiki.
- Each jsonl has the following keys:
- domain: str [The subdomain.]
- path: str [Path to api.php. Which can be different for different languages]
- pages: list[str] [A list of strings containing page names]
v2.5-chunks [folder]
- Contains all the pages fetched from the list in fandom_wikis_pages_210224_v2.jsonl
- The original file it was from is fandom_wikis_pages_contents_210224_v2.jsonl, which is 283.44GB in size and can't be uploaded to HF.
- Each jsonl has the following keys:
- domain: str [The subdomain.]
- path: str [Path to api.php. Which can be different for different languages]
- pages: str [Page name]
- content: raw response from api.php
v2.5-chunks-roblox-filter [folder]
- Contains files after roblox has been filtered.
- Each jsonl has the following keys:
  - domain: str [The subdomain.]
  - path: str [Path to api.php. Which can be different for different languages]
  - pages: str [Page name]
  - content: raw response from api.php
fandom-v0.5.jsonl [file]
- Jsonl file containing the fully processed text.
- Each jsonl has the following keys:
  - text: str [The text content.]
  - meta: dict[str,str] [dictionary of metadata]
    - title: str [The page/name]
    - domain: str [The subdomain.]
    - cats: str [Categories. Extracted and unused.]
    - removed: list[str] [A list of removed stubs / html content]
roblox.domains.txt [Extras]
- A txt list of Roblox domains.

Recursal's Vision

To make AI accessible to everyone, regardless of language, or economical status

This is the collective goal of the RWKV Open Source foundation and Recursal AI, the commercial entity who backs it.

We believe that AI should not be controlled by a select few individual organization. And that it should be made accessible regardless if you are rich or poor, or a native speaker of english.

About RWKV

RWKV is an Open Source, non profit group, under the linux foundation. Focused on developing the RWKV AI architecture, in accordence to our vision.

The RWKV architecture scales efficiently and economically. As an RNN & Transformer hybrid, it is able to provide the performance similar to leading transformer models, while having the compute and energy efficiency of an RNN based architecture.

You can find out more about the project, and latest models, at the following

About Recursal AI

Recursal AI, is the commercial entity built to provide support for RWKV model development and users, while providing commercial services via its public cloud, or private-cloud / on-premise offerings.

As part of our vision. Our commitment, is to ensure open source development and access to the best foundational AI models and datasets.

The following dataset/models provided here, is part of that commitment.

You can find out more about recursal AI here

Dataset Curators

KaraKaraWitch. (I typically hangout in PygmalionAI discord, sometimes EleutherAI. If something is wrong, @karakarawitch on discord.)

I'd be happy if you could spread the word and recommend this dataset for your use cases :)

Licensing Information

Most of all fandom user-created content are licensed under CC-BY-SA unless otherwise noted. By that assumption, we did not include any figures or images as they typically are not licensed under the CC-BY-SA license.

Recursal Waifus (The banner image) are licensed under CC-BY-SA. They do not represent the related websites in any official capacity unless otherwise or announced by the website. You may use them as a banner image. However, you must always link back to the dataset.

Citation Information

@ONLINE{fantaticfandom,
  title         = {Fanatic Fandom},
  author        = {KaraKaraWitch, recursal.ai},
  year          = {2024},
  howpublished  = {\url{https://huggingface.co/datasets/recursal/Fanatic-Fandom}},
}

Special Thanks

undeleted from RyokoAI for providing initial scripts to base stuff on.
I eventually decided to write my own scraper while taking inspiration from their code.