#DreamGen/Opus-V1.2

7 messages · Page 1 of 1 (latest)

distant owl
#

Specialist NSFW roleplay and story writing model.
https://huggingface.co/dreamgen/opus-v1-34b-awq

This model is interesting because it is fine tuned on detailed scenario prompts and story steering. Exactly what a novelai user wants. May not play nicely with novelai out of the box needing a lot of config.

DreamGen (afaik 1 guy) is writing their own roleplay and story platform that the finetune is designed to work with. However, as they are hosting the LLM anyway, maybe they could plug it into router.

#

I had some luck getting the awq above running on a $0.70/h runpod node, and it was pretty fun but I couldn't get it to behave as well as the hosted version. Different temperature settings or something wrong with my prompt maybe.

#

Personally I would really like a more general purpose steerable story finetune if anyone finds one. This finetune knows about prompts and is really good at turning a story into a structure or steering the story mid way through as you add special prompts to its response. I'd like to see that on something smarter and less coombrained.

distant owl
#

I think I loaded the gguf rather than the awq on an A40 on runpod. Because at the time the awq weights were broken.
I used the official the bloke llm template and loaded with the transformer runtime. Used the notepad tab to interact.
Not sure if this size vm was necessary, but I wasn't confident it would fit in the smaller ones.
I like runpod pretty good for playing with new models but it's worse value for money for me than open router.

distant owl