Nous Hermes is poor at changing its appearance and persona. (Resolved, not model's fault)
Unlike GPT4-x-Vicuna-13b, Nous Hermes is giving lots of headaches in regards to changing persona and acts obnoxious when asked more uncensored questions.
Unlike GPT4-x-Vicuna-13b, Nous Hermes is giving lots of headaches in regards to changing persona and acts obnoxious when asked more uncensored questions.
What is the prompt format being used
Both versions of prompts were used:
### Instruction:
%1
### Response:
And the one provided by GPT4ALL by default.
### Human:
%1
### Assistant:
Both versions of prompts were used:
### Instruction: %1 ### Response:And the one provided by GPT4ALL by default.
### Human: %1 ### Assistant:
Can you show me what the actual prompt looks like when in like step 3 of the conversation?
Alright, for the first time, after restarting the GPT4ALL app twice, the result is positive. This might not be related to the Nous Hermes model itself.
### Instruction:
%1
### Response:
For my roleplaying discord bot, I create a "chat history log" of the past messages in the instruction field i.e.
Instruction:
Chat History:
<botname>: message
<username>: message
<botname>: message
Yes, I agree!