Safetensors
Not-For-All-Audiences
xzuyn's picture
Update README.md
db63d36 verified
|
raw
history blame
475 Bytes
metadata
datasets:
  - >-
    PJMixers/NobodyExistsOnTheInternet_ToxicQAFinal-L3-Instruct-8B-PreferenceShareGPT
tags:
  - not-for-all-audiences

Chosen/Rejected Reward Graph

Trained on NobodyExistsOnTheInternet/ToxicQAFinal. I converted the set to a preference dataset using refusals generated from LLaMa-3-Instruct-8B.