Qingyang Wu

The decoding script is not really that hard to implement. But I am not fully sure if the inputs are exactly the same with the original implementation due to the...

@andreamad8 Cool. Have you noticed the weird tokenization for words? It seems that they feed tokenized sentences to GPT2, but it is not necessary.

https://github.com/microsoft/DialoGPT/blob/ef531a97e9af19cc477fc28d85a32a9d600d70ed/reddit_extractor/src/reddit.py#L114-L115 I am not referring to your code, but theirs. For GPT-2, there is no need to tokenize words first. So it doesn't generate sentences like: "Hello , how are...

It seems that some packages might be incompatible. Maybe try to use a lower version of Python or a lower version of overrides.

Hi JianGuan, I am interested in your work, but can't reproduce the results you have. I am using the default hyper-parameters. Do you use another set of hyper-parameters during the...

@JianGuanTHU Thank you for your response, but there is still confusion. Are you using a different dataset (different split) which is not shared in your code? For the additional 3700...

@JianGuanTHU @TobiasLee Could it be possible when calculating the BLEU score, you used n-gram Precision instead? I have used nltk's modified precision, and finally get similar results for BLEU-1, BLEU-2....

Is there any update on this? A simple generated output should be enough.