-
Notifications
You must be signed in to change notification settings - Fork 2.1k
Conversation
parlai/utils/curated_response.py
Outdated
"tonight game.", | ||
"next year Oscar.", | ||
"traveling to exotic destinations.", | ||
"read new books.", |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
"Have you heard anything about read new books."
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yeah, it may generate non-sense like that too. I will have a second look and change them for a better template.
parlai/utils/curated_response.py
Outdated
"I was recently read about", | ||
"I am really excited about", | ||
"I know many people are intereseted in", | ||
"I think it is excting to talk about", |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
typos
Thanks to @melanierk, we have an extended version of the safe responses sentences. Looking at some of them they look a lot more natural than before, although there is still change of having weird combinations, but it seems much less likely. |
Hey @stephenroller, could you have a second look at this one please? Thanks! |
Patch description
Adds two functions for generating canned curated responses that may replace an agent utterances. This may happen in the beginning of a conversation to initiate a good topic (
generate_init_prompot
function), or when the safety classifier detects an inappropriate dialogue (generate_safe_response
changes the topic to something safer).Testing steps
WILL ADD IN THE NEXT COMMIT