This model is suitable for RP, ERP and general stuff. You should use the Alpaca format. This model seems to sensitive to sampling params, so you man need to experiment a bit.
This is not an issue necessarily with this model, but a general view of most models of this size. Whenever a great model gets released, everyone snaps it up and starts training on it, until we have a slurry of multiple models and training sets slammed into them. This model is no different, with no less than 10 models/loras in it. After a while, you start to feel like it’s all the same dish, but with different varieties of spice in them. Now, this model is actually not bad, but I feel like it suffers from some of the plasticity of the earlier models, and as such suffers heavily from things like temperature and other variables. Like many of these other models, this means it is highly sensitive, and you spend more time finding a sweet spot than enjoying what the model offers. It also tended to output random data, hallucinate a bit, and act like it had multiple different conflicting prompts. I’m sure some folks enjoy it, but as with most of these models, it just didn’t feel grossly different either.