Datasets:

Tags:
Not-For-All-Audiences
Libraries:
Datasets
License:
grimulkan's picture
Update README.md
f88f4cd verified
|
raw
history blame
874 Bytes
---
license: apache-2.0
---
This is a version of [bluemoon-fandom-1-1-rp-cleaned](https://huggingface.co/datasets/Squish42/bluemoon-fandom-1-1-rp-cleaned) further cleaned up using [Karen_TheEditor 13B](https://huggingface.co/FPHam/Karen_theEditor_13b_HF), in Fastchat format.
I tried to fix as many of the grammatical issues as possible and didn't drop any conversations, but there are still issues since Karen is not perfect (and only 13B).
If I detected any large deviations from the original text in the corrections, I fell back to a standard spell-checker, excluding estimated proper nouns from the spell-checker (which is also not perfect).
I intended this as a first pass and mainly wanted to test an automated way to clean language data sets, but I never got around to doing a better job. Still, it is probably better than most other versions out there (for now).