Size unmatched when running EAP-IG notebook

[Miamm0v0]

Hello!
Thank you very much for your outstanding work. I have encountered a question when I am trying to run EAP-IG notebook. I was running knowledge_eap.ipynb using gpt2-medium and found that
Activation Hook Error hook_embed torch.Size([1, 18, 1024]) torch.Size([1, 17, 1024]) 0 and
RuntimeError: The size of tensor a (18) must match the size of tensor b (17) at non-singleton dimension 1
I tried to solve the problem following #7 in EAP-IG but it didn't work.

[littlefive5]

Can you provide your code and running examples?

[Miamm0v0]

g = Graph.from_model(model)
start_time = time.time()
# Attribute using the model, graph, clean / corrupted data and labels, as well as a metric
attribute(model, g, data, partial(logit_diff, loss=True, mean=True), method='EAP-IG-case', ig_steps=100)
# attribute(model, g, data, partial(direct_logit, loss=True, mean=True), method='EAP-IG-case', ig_steps=30)
# attribute(model, g, dataloader, partial(logit_diff, loss=True, mean=True), method='EAP-IG', ig_steps=30)
g.apply_topn(5000, absolute=True)
g.prune_dead_nodes()

g.to_json('graph.json')

gz = g.to_graphviz()
gz.draw(f'graph.png', prog='dot')

end_time = time.time()
execution_time = end_time - start_time
print(f"程序执行时间：{execution_time}秒")

I was running the knowledge_eap.ipynb and all went well until I got to the block I mention above. and I got the output of

[littlefive5]

Hello, sorry for being late.
This problem is due to the different lengths of the tokenized corrupted and clean subject tokens.
The tokenizers of GPT2 and LLAMA2 are different, making the case I used here unmatched. You can use different corrupted subjects to make sure the length is the same as Eiffel Tower such as the Santa Monica Beach.
I also enhance the assert here.

[zxlzr]

hi buddy, do you have any further questions?

[Miamm0v0]

Hi!
Thank you very much for your assistance. I think I have no more questions so far and I'm gonna close this issue.