MEMIT related method issues for llama3.2

[Syon-Li]

Dear EasyEdit Team,

I got the following error when trying to run MEMIT for llama3.2-3b

Traceback (most recent call last):
  File "/Editing_baselines.py", line 533, in <module>
    main()
  File "/Editing_baselines.py", line 399, in main
    metrics, edited_model, weights_copy = editor.batch_edit(
  File "/EasyEdit/easyeditor/editors/editor.py", line 227, in batch_edit
    edited_model, weights_copy = self.apply_algo(
  File "/EasyEdit/easyeditor/models/memit/memit_main.py", line 46, in apply_memit_to_model
    deltas = execute_memit(model, tok, requests, hparams, cache_template=cache_template)
  File "/EasyEdit/easyeditor/models/memit/memit_main.py", line 137, in execute_memit
    cur_z = compute_z(
  File "/EasyEdit/easyeditor/models/memit/compute_z.py", line 149, in compute_z
    log_probs = torch.log_softmax(ln_f(full_repr) @ lm_w.to(full_repr.device) + lm_b.to(full_repr.device), dim=2)
RuntimeError: The size of tensor a (128256) must match the size of tensor b (3072) at non-singleton dimension 2

Can you confirm whether this is a bug in your code?

Hi, the same mistake occurs in other MEMIT-related methods, like EMMET, and PEMT (maybe there are others), could you help fix it?

Originally posted by @Syon-Li in #458

[Syon-Li]

I have fixed it myself, but you can still update this 7e35bb1#diff-37751c17b0cd75afce0559f97c28e59f0a1cd662f4dc7eea5ba88fd1246166dcR21 to your repo. Thanks!

[zxlzr]

Hi buddy, due to limited manpower and resources, we will address this issue as soon as possible. Thank you for providing the solution.

[pengzju]

Thank you for your reminder. I have updated the hparams file for PMET and EMMET. plz allow me to close this issue.