Closed chand1012 closed 1 year ago
This seems to be somewhat remedied when using ChatGPT as your backend thanks to these commits: 611f0e758a4420c3611e8142e61521328d273760 & 09a0e1f529b726f1931dd5fad87f5cac1ac56015
I need to do some more testing on FastChat, as I opted to use completion so I didn't sacrifice any input tokens, but there's a bug with FastChat-T5 where it doesn't work since it is not properly handling token counts. For now I'm going to continue on other work and come back to this once the bug is fixed with FastChat.
Steve says the token responses seem a bit lacking. Try to find a way to increase response size effectively.