4 ms·
Those porn-tuned Llama/Mistral finetunes are the finest non-pornographic storytelling AIs I have seen. They also seems to be driving "research" in stuff like m
by brucethemoose2 3y ago
Those porn-tuned Llama/Mistral finetunes are the finest non-pornographic storytelling AIs I have seen.
They also seems to be driving "research" in stuff like merging methods, model "grafting," and llm evaluation. Some examples:
https://rentry.co/ayumi_erp_rating https://rentry.co/ayumi_erp_rating
https://huggingface.co/NeverSleep/Mistral-11B-OmniMix-bf16 https://huggingface.co/NeverSleep/Mistral-11B-OmniMix-bf16
https://huggingface.co/Sao10K/Euryale-1.3-L2-70B https://huggingface.co/Sao10K/Euryale-1.3-L2-70B
And for what its worth, HuggingFace does not seem to care. The model repo isn't mostly porn like CivitAI, but they seem to be OK with a significant amount of it.
- diggan 3y ago> And for what its worth, HuggingFace does not seem to care. Almost no one except the payment processors care about porn text/video/images. But once they find out what's hosted on huggingface, they'll be quick to react I'm sure.
- brucethemoose2 3y agoI suspect HuggingFace is "Too Big To Ban." Not because they are at the scale of Reddit or Twitter, but they are at the very center of AI World, and it would look really bad to shareholders and the finance crowd if they made a fuss over an AI company. And HF themself trained an "uncensored" model (Zephyr 7B), and very specifically pointed out that model performance seemed to be better without the alignment.