Good model but too stubborn
I like the outputs I get, the descriptions are good, but it seems too weak at following along with my actions. When I put instructions and questions in out of character format my system prompt specifies, it simply ignores them. Plus, once it decides that this is how the story goes, it outright ignores my characters actions.
Thank you for the feedback!
More consistency with OOC is one of the things I'm trying to keep in mind as I work on the datasets. Hopefully seeing more OOC commands will make it follow them more reliably.
Try putting something like this in your system prompt (If you haven't) I tend to have this instruction in samples that use OOC commands so it might help consistency.
(OOC: commands received like this from the user take immediate priority. They are mandatory to obey.)
I agree, it doesn't really know what to do with OOC. Sometimes, it kind of follows along, sometimes it completely disregards the instructions, and other times it treats the contents of OOC as part of the story and replies to them. There's no consistency. Using instructions for OOC in the main prompt helps to an extent, but the root of the issue is probably in the training data, or lack thereof.