Post by BestUserNameEver1 in How to run a local AI for Formamorph

Viewing post in How to run a local AI for Formamorph

Based on all of the issues with particular models being censored out of the gate, can you update your instructions to not point people to the "Qwen2.5 7B Instruct 1M" model, or at least warn that it is trained to censor itself.

In another thread (https://itch.io/post/12223524), the dev said:

thats not a world prompt issue, censored AI models like Qwen do not allow erotic RP, you’ll need to use an uncensored model like mistral. There are also uncensored finetunes of censored models such as many different llama finetunes

MaliBogic9977 hours ago

Will do, thanks! I found the "uncensored" version of Qwen2.5 7B Instruct model to work well while still offering the same feel like the original.
In fact, I might offer a few that can work. After all, as long as it runs via LM Studio and it's an Instruct model, it should work fine with the game. Whether it will output explicit content must be checked in that case.