Pretraining KoLD Dataset with pretrained "koelectra-v3" model.
dataset : https://github.com/boychaboy/KOLD
pretrained_model : https://huggingface.co/monologg/koelectra-base-v3-discriminator
So you should use tokenizer with "koelectra-base-v3-discriminator".
label maps are like
{0: "not_hate_speech", 1: "hate_speech"}