Requirements: A python installation with huggingface-hub module to use CLI.
Licensing: The license for the provided quantized models is derived from the original model (License detected: apache-2.0). For additional information see the original model's page above, or, if unavailable, the files and the page backups below.
I won't spam too much here, as most of the stuff written in the Impish_Bloodmoon_12B model card is relevant here too. Mostly. So go ahead and read it, if you want.
Angelic_Eclipse_12B got an exceptionally realistic common sense— a character would respond almost exactly like a real character would, no anime unhingedness, unless it got into such a situation 'organically'. It's wholesome, emotionally intelligent, but it can also... go all the way, spectacularly so. If you want a realistic model that won't break your balls, never rushes you, or you want to pull the brakes on an adventure that started with its impish sister, this is definitely the model for you! 👼
This model knows A LOT, as it was trained on a similar massive dataset (well over 1B tokens trained) of its nasty counterpart. It knows all there is to know about love and intimate relations, but... this does not mean it would nudge you towards this aggresively (Impish_Bloodmoon_12B 100% would!). Even in a scenario where X likes Y, it would shyly and gently pursue Y.
This model was architectured specifically for four things:
Reactions realism. It meant to reflect real-life behaviour accurately
Slow burn
Safety tuning
Powerful 'vanilla assistant'
This means that the model has a 'normal' vibe and personality, and would behave mostly like a typical assistant. It's presentable, with production-ready safety, but still extensively tuned for creative tasks. It would steer stuff towards SFW, and ah, 'take it slow'.
This model was meticulously tested by a friend, by hand. The example is fully available here. Feel free to check it out.
(The example was made using the included roleplay card, with these Silly Tavern settings, option #1.)
No editing was done to any of the replies. For the entire run, only up to 5 regens were allowed, start to finish. (in other words, what you see is what you'll get).
Almost all the data of Impish_Bloodmoon_12B, but completely grounded to reality.
VERY creative, high swipe diversity
Regarding the format:
It is HIGHLY RECOMMENDED to use the Roleplay \ Adventure format the model was trained on, see the examples below for syntax. It allows for a very fast and easy writing of character cards with minimal amount of tokens. It's a modification of an old-skool CAI style format I call SICAtxt (Simple, Inexpensive Character Attributes plain-text):
SICAtxt for roleplay:
X's Persona: X is a .....
Traits:
Likes:
Dislikes:
Quirks:
Goals:
Dialogue example
Silly Tavern Settings #1 - Higher temperature while still being coherent
Silly Tavern Settings #2 - Dynamic paragraphs, XTC, other stuff
Chat Examples:
Roleplay Examples (click below to expand)
Bonding with Nikki Rivera while trying to figure out how to survive along the way:
Roleplay_Nikki Rivera_Chat_Example
Model instruction template: ChatML
<|im_start|>system
You are an angelic AI, your job is to fulfill thy will of thy user.<|im_end|>
<|im_start|>User request
{prompt}<|im_end|>
<|im_start|>AI answer
