Skip to main content

Indie game storeFree gamesFun gamesHorror games
Game developmentAssetsComics
SalesBundles
Jobs
TagsGame Engines
(1 edit)

I was originally using the devs AI which worked pretty well beyond eventually just forgetting everything and ruining scenarios so I decided to host my own after I found this post. However, I am having issues where the AI will not be very descriptive and only writes like 3-4 lines. It also will not add the dialog my character is saying into its responses.

The other main issue I have as after a bit of time the AI seems to get certain data stuck in it as it keeps adding the same lines into every response, sometimes multiple times in a response. 

Lastly, I am unable to get the AI to respond to questions or commands that it will not try to use inside the story. With the devs, I was able to ask it enough times by calling it AI and asking it how I can do certain things. Eventually it told me that if I use  "!info," (w/o quotes) the anything after it would consider a command or question that it would not use in the story. I was able to get it to spawn enemies, alter behaviors, and much more. But the ones I am using now just ignore it no matter how many times I ask. Any ideas how to force or train it to respond outside the story?

I have tried two different models which are:

qwen2.5-7b-instruct-uncensored

Mistral 7b instruct uncensored ( not sure of the proper name as I am at work and don't remember exactly but I will update when I get home. )

Are my issues just due to them being newer and not as experienced as the devs? Also I have tried adding the prompts from the guide and the giant list of ones from ArlostRVN. But these didnt seem to help. I made sure I restarted the world to make sure the prompts take effect though I am not sure if you have to or if they're used right away. 

Please let me know if you need anymore information or if you have ideas on how to help.

So I ended up getting fixing my issues for the most part though the data getting stuck still happens but doesn't seem to just be an issue for me from what I have heard from others.


The non descriptive responses from the AI was with single-paragraph mode turned on and turning that off helped but also I moved to a higher quant version of my AI and that helped a lot. Along with helping the data not get stuck as often though it still does happen, I think once the context fills up too much it starts happening.


In regard to getting the AI to respond outside the story, it just took some coaxing and rules for it to start replying.


Moved from Mistral-7B-Instruct-v0.3-abliterated.Q4_K_S to the same model Q5_K_M which still fit in my VRAM.