Questions about prompting
Sorry if this is obvious, but just wanted to ask incase.
For the prompt should we replace <SECOND PERSONA>, <FIRST PERSONA>, <SCENARIO> with our own text? or do we put it above?
is "%{having a conversation|in conversation|in a roleplay chat}" just a way to have more random variance and it will choose a random one?
Where should we put the chatlog under ?
Enter roleplay mode. You are currently %{having a conversation|in conversation|in a roleplay chat} with <SECOND>, whose %{traits are|persona is|characteristics are}:
<SECOND PERSONA>
%{You are|Play the role of|Take the role of} <FIRST> with the following %{persona|definitions|character sheet|traits}:
<FIRST PERSONA>
%{In addition|Additionally|Also}, %{keep the following scenario in mind|remember this scenario|pay attention to this scenario}:
<SCENARIO>
Ahhh I don't think this was explained properly. So, the experiment used the character names themselves as the User and Character roles. Here's a sample prompt using the system prompt template you sent:
SYSTEM: Enter roleplay mode. You are currently having a conversation with User, whose traits are:
<SECOND PERSONA>
You are Character with the following persona:
<FIRST PERSONA>
Also, pay attention to this scenario:
<SCENARIO>
User: <Add the first user message here>
Character:
Stuff enclosed in %{} was a random choice between given options. For example, %{Red|Blue} will either choose Red or Blue when creating the dataset from the python script linked in the model card.
And yes, everything enclosed in angle brackets should be replaced with your own text. Another thing to note is that the system prompt changes depending on what lora was used (ex. prompt-a onechar), so be sure to look at those respective system prompts and adapt accordingly.
I use this lora without instruct mode in SillyTavern, so I am not too familiar with instruct and system prompts, but this should work.
Thank you very much. I tried it according to how you said and i'm getting much better responses.