Weak reasoning/consistency?
I've tried the "base" QwQ 32B (found here: https://huggingface.co/bartowski/Qwen_QwQ-32B-GGUF/blob/main/Qwen_QwQ-32B-Q6_K_L.gguf) and have enjoyed it being 100% consistent on both doing reasoning and consistently updating stats or other things being included in the chat log (for example: Tracking time of day, character location, character activity, and other stats pertinent for the scenario) with an identical format as requested. I can go 300 posts deep in a chat with that model and not have a single hiccup.
I tried swapping in the ArliAI RpR model with no further changes and it's a mess. The reasoning section has response postings, no stats are tracked. Well, sometimes it feels like tracking some stats for a response or two, but in less than half a dozen posts the data is completely filtered out.
An example of some stats I've tried tracking in one character card (an homage to The Beast and His Pet High School Girl):
<OwnerStats>
[](#'Affection: 100%')
[](#'Attraction: 10%')
[](#'Health: 90%')
[](#'Species: Wolf')
[](#'Gender: Female')
</OwnerStats>
This block has faithfully been updated in this exact format for 81 responses from the AI using the "base" version above. I cannot get ArliAI RpR to update it once. I can provide the character card for testing if requested.