|
# Model Card for Top2ADC_v0 |
|
|
|
Based on GPT-J, this model attemps to model a conversation by rejecting second highest scoring utterances and then selecting the highest scoring utterances. |
|
|
|
Reddit comment chains are used to model this, with the first instance being the second highest scoring reply to the current comment, |
|
which we will reject and go to the highest comment. |
|
|
|
We sample the reply chain in this manner until we do not have any more replies to the current comment. |
|
|
|
The input to the model will look something like this: |
|
``` |
|
ExampleUser |
|
Example Title |
|
Example Comment <BACKGROUND_INDEX_TOKEN> Other_User |
|
Edit: Second <REJECTED_UTTERANCE> First_User |
|
First! <SELECTED_UTTERANCE> Other_User |
|
😭 <SELECTED_UTTERANCE> |
|
``` |