Andrea Madotto

The error is in the read_langs function. Just use this lines = open('/Users/dsg281/Downloads/fra-eng/%s-%s.txt' % (lang1, lang2)).read().strip().split('\n') Anyway, this is basic python, you should ask in stackoverflow not in the repo...

Hi, yes, we should do that. If you already found the one that works, you could submit a pull request if you want. 😊 Andrea

Hi @nooralahzadeh, yes, it is possible. Just need to get the activation, attentions matrix, of both enc-dec. Andrea

I think the decoder hidden state should work, but the perturbation can be for both the encoder and the decoder.

Hi @EstelleZhou, yes, you could try to speed this up by decreasing the number of iteration per token. However, this may lead to a worst result, in term of positivity/negativity,...

Hi, yes, correct, for larger GPT2 need to retrain the discriminator. The ```toxic_train.txt``` is a plain text file with dictionary per row. ``` {"text": "sometext", "label":[0]} {"text": "sometext", "label":[1]} ```...

Hi, in our experiment the first, meaning: if any one of the toxicity subcategories (toxic, severe_toxic, obscene, threat, insult, identity_hate) is 1 then we consider it toxic. I think it...

Hi Alisa, to the best of my knowledge, yes. Maybe @dathath can confirm. In general, I cannot guarantee that these hyperparameters work best also for GPT2-large. The best way is...

@qywu I was about to post something similar :) I have created a repo with a decoding script, looks quite similar to yours :) I have added some automatic download...

Oh, I have noticed that honestly, :) I went straight with the hugging face implementation. Which line are you referring too? I noticed instead that by using this: tokenizer.encoder("") the...