license: mit
task_categories:
- conversational
- text-generation
size_categories:
- 100K<n<1M
janitorai-cards
This dataset contains 190k cards that I received from janitorai, from a source that wished to remain anonymous.
My addition to this data is conversion of cards to v2 character card format, and a local webpage that can be used to explore the dataset.
Webpage
Ther webpage lets you browse cards, search by text, fitler by tags and order by date/name/popularity.
To use the webpage, put index.html into a directory, and download and extract archives into same directory: 0123.zip, 4567.zip, 89ab.zip, cdef.zip, and html.zip.
After that, just open index.html in the browser.
The directory structure should look like this:
π
β£ββ π index.html
β£ββ π cards
β β£ββ π 0
β β£ββ π 1
β β ...
β βββ π f
βββ π html
β£ββ π allcards.js
β£ββ π cards.js
βββ π cardsmeta.js
For performance reasons, the webpage only loads 10000 most popular cards when you open it. To view all, click the "Load all" button in the top row.
Caveat: instead of downloading the card, it opens it in a new pageβyou have to save it yourself. I can't figure out how to get the download to work.
Files
- 0123.zip, 4567.zip, 89ab.zip, cdef.zip - archives with v2 character cards, tested to work with SillyTavern.
- cards-js.7z - all v2 character cards in json format, without images, tested to work with SillyTavern.
- index.html - webpage for browsing cards.
- html.zip - files with information about cards - it's needed for the webpage to function.
- orig.7z - original json files with cards from janitorai - not compatible with any software.