Advertisement
Not a member of Pastebin yet?
Sign Up,
it unlocks many cool features!
- CHARACTER - Hi User.
- USER - Hi Character, what is this file for?
- CHARACTER - This file is to demonstrate the formatting for training data for an AI?
- USER - Interesting. Can I use this to train an AI?
- CHARACTER - No, it's not long enough. It will not produce any useful results.
- USER - So what, exactly, should the training data look like?
- CHARACTER - It should look like two named people talking to each other. The names need to then be set in talktocharacter.py!
- USER - How long does it need to be?
- CHARACTER - I'd recommend at least a thousand lines.
- USER - What other requirements does the training data have that aren't obvious?
- CHARACTER - People must alternate when talking. The same character shouldn't say two lines in a row.
- USER - Any others?
- CHARACTER - You can only have two people in the conversation.
- USER - What if I wanted the AI to learn emotions?
- CHARACTER - You can transmit extra data to the AI by starting each line as i'll show you.
- USER - Go on...
- CHARACTER - <BORED> A tag like this can be used to give extra training data.
- USER - And I'd need to use some format for that consistently across the training data?
- CHARACTER - Yes, that will allow finetuning to learn that it needs to also output an emotion alongside each line.
- USER - Thank you Character. I'll go find suitable training data now.
- CHARACTER - No problem, User.
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement