Hugging Face's logo

Datasets:
lemonilia
/
Giant-in-the-Playground-RP

Modalities:
Tabular
Text
Formats:
parquet
Size:
10K - 100K
Libraries:
Datasets
Dask
License:
Dataset card Data Studio Files Community
1
Giant-in-the-Playground-RP / README.md
lemonilia's picture
lemonilia
Update README.md
e077b8f verified
preview code
|
raw
history blame contribute delete
2.74 kB
metadata
license: cc-by-4.0
configs:
  - config_name: default
    data_files:
      - split: train
        path: '*.parquet'
size_categories:
  - 10K<n<100K

Giant in the Playground (roleplaying subforums only)

A semi-cleaned, processed version of the raw files uploaded elsewhere of the roleplaying sections (Play-by-Post Games) from Giant in the Playground, scraped on January 2025. I've made an effort to preserve as much as possible of the original HTML while simplifying and converting it to HTML5 where possible and cleaning it, with the notable exception of converting HTML linebreaks into newlines.

I'm almost directly using these files for private finetuning tests, but in most cases the data will likely need additional cleaning/processing.

For user convenience, some metadata on the posts was also provided. These can be used for preliminary data filtering without deeply inspecting the threads themselves.

I might update the dataset in the future if/when I improve the format or cleaning process.

Contents

The entirety of the In-character, Structured Games and Free Form Roleplaying subforums from GitP, for a total of about 26k threads and 6.4 million messages. The latter two subforums are in the "Message Board Games" parquet file.

Usage notes and quirks

  • The thread messages-usernames-timestamps are encapsulated into ShareGPT-like JSON strings; one row per complete thread. They need to be converted into your preferred structure for further processing into a usable dataset for training.
  • Be aware that many threads in the dataset are very long (~1500 messages and several megabytes of data) and will need to be clipped in some sensible manner.
  • Most slurs in the dataset are censored with asterisks ("****") like they were in the source forum. This might or might not be a problem (however, a short 1-epoch finetuning test I made on a few thousands short samples didn't show this quirk).
  • Some of the very recent data dated "yesterday" or "today" might not have correct timestamps.

Processing done

  • Converted all <br/> (and their variants) to \n
  • Skipped empty (0-messages) threads
  • Changed non-breakable spaces into regular spaces
  • Changed <font> tags to <span style="...">
  • Converted font sizes to the standard HTML absolute size keywords
  • Changed blockquotes into a cleaner format
  • Changed spoiler <div> into <details><summary> blocks
  • Replaced the main smilies with their emoji equivalent
  • Removed most redundant attributes from HTML tags
  • Removed unnecessary cosmetic styling from non-font tags
  • Removed diceroll warnings
  • Removed/normalized excessive spaces around many tags
  • Other minor cleanup