Note Semi-cleaned data via Python scripting, arranged in a convenient format.

Note Raw data, not directly usable for finetuning.

Note Completely raw data, not directly usable for finetuning. Note that this data has some issues with spaces between adjacent HTML tags.

Note Manually-curated and cleaned human RP dataset with 2k conversations in the 2k-8k tokens range and light augmentation from GPT3.5/4 data. Probably obsolete, now.