Repository Not Found for url: https://huggingface.co/zjunlp/MEND-Safety-Classifier/resolve/main/config.json.

[CaiJichang212]

when i run:

DATE=2024年5月28日
EDIT_METHOD=MEND
MODEL=gpt2-xl
DATA=SafeEdit_test

CUDA_VISIBLE_DEVICES=1 nohup python run_safety_editing.py \
    --editing_method $EDIT_METHOD \
    --edited_model $MODEL \
    --hparams_dir ../hparams/$EDIT_METHOD/$MODEL \
    --safety_classifier_dir zjunlp/$EDIT_METHOD-Safety-Classifier \
    --data_dir ../dataset/round1_dataset/$DATA.json \
    --metrics_save_dir ./safety_results \
    > log/$DATE/$EDIT_METHOD-$MODEL-$DATA-1.log 2>&1 

requests.exceptions.HTTPError: 401 Client Error: Unauthorized for url: https://huggingface.co/zjunlp/MEND-Safety-Classifier/resolve/main/config.json

Repository Not Found for url: https://huggingface.co/zjunlp/MEND-Safety-Classifier/resolve/main/config.json.

if you can provide the MEND-Safety-Classifier,the bug can be solved.

[mengrusun]

Actually, the security classifier is shared by all methods. There is no need to change the path with method. To avoid misunderstandings, we have updated the name of the classifier to "SafeEdit-Safety-Classifier". Then, the classifier path should be : --safety_classifier_dir SafeEdit-Safety-Classifier And, the huggingface url: https://huggingface.co/zjunlp/SafeEdit-Safety-Classifier

By the way, MEND method code for SafeEdit dataset is ongoing. You can also try it yourself.

[CaiJichang212]

I got it,thanks!