Train using paragraphs of Story and then predict future?

[krigeta]

Hello thanks for this awesome repo, I am wondering if it is possible to predict future events based on the past events of a story?

like instead of input the whole conversation using characters, we can able to input the whole story and then able to predict the future?

[e-tornike]

Hey @krigeta,

thanks for your interest!

If I understand it correctly, you want to predict text (i.e., output) that follows from previous text (i.e., input)? If so, then I would recommend looking at Language Modeling (e.g., BERT or GPT-2). This blog explains it very well.

Hope that helps. Cheers!

[krigeta]

Hey, Actually I am using the GPT-Neo an alternative of GPT-3 and it generates text but the output is not that good. So thinking to finetune the GPT-2 using your collab but the dataset i have is in paragraphs not like in dialogue format.

So i was asking you that how can I use the dataset I have to finetune the GPT-2 using your colab?

[e-tornike]

I see. I would take a look at the most recent examples by HuggingFace (see language-modeling or the Colab notebook listed here) and you can definitely ask questions and find answers in the issues there.

[krigeta]

okay will do that and please let me know if you find something related. that would be a great help.