How to tune MEMIT's hyperparameters under thousands of edits? about easyedit HOT 4 CLOSED

Hi there! Could you be more specific about what you ran?
Have you employed the batch_edit() when you edit thousands knowledge simultaneously?
Meanwhile, you should change the batch_size.

from easyedit.

Hi there! Could you be more specific about what you ran? Have you employed the batch_edit() when you edit thousands knowledge simultaneously? Meanwhile, you should change the batch_size.

I used batch_edit(), but I found that I didn't add the batch_size parameter to the yaml file, and I'll experiment further to see if it works. Thank you for your reply !

from easyedit.

• The default batch_size is 1: https://github.com/zjunlp/EasyEdit/blob/main/easyeditor/models/memit/memit_hparams.py#L43
• You can add the batch_size field to yaml and use batch_edit function for batch editing

from easyedit.

Hello, I'm editing MEMIT on llama-7b and llama-13b models with thousands of edits. Is there any batch size that you recommend?

from easyedit.