So much good in this model
Wow, this model pretty much has it all. It passed most of my testing. I used the Kobold (Godlike) preset and have the temperature set to 0.8 for mine/ My results:
-Creative responses (check)
-Taking an initiative (check)
-Ability to roleplay side characters including its own (check)
-Avoiding common GPT phrases (check)
-Knowing known characters from franchises (Use instruction: Generate a conversation between {{user}} and character name from franchise.) Replace character name and franchise for what you want, of course. While it might not capture the character's personality, it at least knows who they are but a good character card helps a lot.
Love that I can use 8k context size with this but used Bakanayatsu for the larger size as recommended.
The minor downside is that it sometimes ends the response with a bit of a summary for what happens next despite my prompt for it to not summarize but the fix is simple. If this happens to you, put "Never Summarize" in the instruction.
Thank you for the feedback! I'm personally pretty happy with this model as it seemed to keep the best aspects of all the models merged. But it did have that side effect of it taking the story so far ahead it starts summarizing, a higher temperature seemed to help with it but i think its a okayish tradeoff. someone with more skill could probably fix it. But i lack the know-how nor do i have the compute power.
No problem! If I knew anything about fixing models I'd help with the summarizing, but it still does pretty well as it is and it's exactly the type of roleplay model that I wanted. You certainly chose the right models to merge.