You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I want to train the dialogue generation with the library simctg. But the dataclass changes between the original source code and the example of "training_tutorial_on_wikitext103".
I'm using GTP2 and I want to reproduce DailyDialog results. Did I need to get the original dataclass for dialogue generation ? Or the new dataclass in "training_tutorial_on_wikitext103" works for all examples and English and Chinese GPT2 ? Specially for the tokens ?
Did I need to change something to the "training_tutorial_on_wikitext103" example for english dialogue generation ?
Thanks for help.
The text was updated successfully, but these errors were encountered:
You can directly use the codes provided in "training_tutorial_on_wikitext103" to reproduce the results on dialogue generation. The only thing you need to take care of is making sure your file format is consistent with our wikitext-103 files provided in the training example.
Hello,
Very good job for your work. It's amazing !
I want to train the dialogue generation with the library simctg. But the dataclass changes between the original source code and the example of "training_tutorial_on_wikitext103".
I'm using GTP2 and I want to reproduce DailyDialog results. Did I need to get the original dataclass for dialogue generation ? Or the new dataclass in "training_tutorial_on_wikitext103" works for all examples and English and Chinese GPT2 ? Specially for the tokens ?
Did I need to change something to the "training_tutorial_on_wikitext103" example for english dialogue generation ?
Thanks for help.
The text was updated successfully, but these errors were encountered: