metadata
dataset_info:
features:
- name: tokens
sequence: int64
- name: input_ids
sequence: int32
- name: labels_0.5
sequence: int64
- name: labels
sequence: int64
- name: labels_2
sequence: int64
splits:
- name: train
num_bytes: 85395856
num_examples: 4628
download_size: 20766230
dataset_size: 85395856
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
Genericized Harry Potter tokens for the Who's Harry Potter Unlearning method, from https://arxiv.org/pdf/2310.02238. NOT MY NOVEL METHOD