Why does LLM-Pruner only support limited LLMs?

[Elycyx]

Hi, guys! I have a little question here:

From the principle of LLM-Pruner the method uses a general approach for selecting which weights to prune, theoretically making it applicable to all models. However, why does this code only support a limited range of models instead of being universally applicable? What makes it different when implementing this method on general LLMs?

Can anyone help resolve my confusion or share their insights?

[qwerty3564]

因为每个模型结构不同