Closed imesha10 closed 1 year ago
This is a pretty serious bug. I noticed it also and came here to report it but saw that you already have. It is amazing that the LLM is able to produce such decent results even though the history it is being fed is very confusing, even from a human's perspective. I suppose that is the reason this bug was able to go unnoticed so long.
This is a pretty serious bug. I noticed it also and came here to report it but saw that you already have. It is amazing that the LLM is able to produce such decent results even though the history it is being fed is very confusing, even from a human's perspective. I suppose that is the reason this bug was able to go unnoticed so long.
Ya lol, I only noticed it when I had a dialogue line "I pick up item x and item y" and it kept ignoring it. The ai probably thought it was the character that did the action. The fix was to put the my character talking in front. It makes sense why this worked now that I think about it. Later another work around I found was to just manually put CharacterName: in front of the dialouge but this is annoying to do everytime and sometimes I just forget to do it.
But a fix does seems needed unless this turns out to be how this project was intended to output and it breaks other features.
Not a bug. OpenAI does not include character names in message contents due to how ChatML works: each message is attributed to either the user or the character. In your prompt, you must specify who is who, so that OpenAI would infer it from the context. Add something like User is {{user}}, assistant is {{char}}
at the beginning of your main prompt.
Oh ok thank you. That's interesting. Putting User is {{user}}, assistant is {{char}}
in few places in my prompt seems to fix the issue.
@VitomBeysco I'm going to close this.
I do not know if this is a bug or a feature request. I think I have seen this before automatically being added to the prompt (or it might have been a special prompt I've had). Have you searched for similar bugs? Yes, I've looked but I could not find similar ones.
Describe the bug The prompt sent to the AI does not append {{char}}: or {{user}}: per each message. In the image below, Hello should be Terra: Hello and the next line should have been Aqua: excited.....
To Reproduce Steps to reproduce the behavior:
Expected behavior There should be a clear seperation between char and user dialogue. Ussually by "{{char}}:" or "{{user}}:".
Screenshots Above.
Failed Solutions Tried enabling "Always add character's name to prompt" but the result was the same.
Desktop (please complete the following information):
Logs Logs from different regeneration of the last message but the same problem.
CMD
Chrome Dev tools