Exceptional
I'm not sure if this is worthy of a whole community entry - but in my local testing - using OoobaBooga and SillyTavern (with Alpaca-ish formatting) - this model works rather well! It has some of the same issues as many similar "low context & 13b" models, but when using the chat instruct mode without ST - it's WAY above the (public & free) ChatGPT responses, and rivals Bing - all without any internet access. It's also not shy when expressing opinions.
It's rather smart, sticks to complex orders well and tries its best, in general, to fight the limited reasoning. Its so good, in fact, it prompted me to respond here =)
ERP and RP work well - but common phrases are still to be found, and are mostly rather similar. It has it's moments of clarity - and uses them very impressively so. Thanks for the hard work, DaringMaid is certainly my new all-rounder for locally hosted, 4k context, Roleplay-Focussed models!
Happy Holidays, btw.
Edit: I've used the "gptq-4bit-32g-actorder_True" permutation as usual. VRAM budget was 12GB ideally, running on Windows 10, loaded by ExLlama2_HF.