addansee/CivChan-1.2B-Distill

CivChan 1.2B Distill

Lore: https://civitai.red/articles/13076

Dataset: Distilled from Civitai's fine-tuned LLM
(I don't know to which one they switched if did. But we know they have been using GPT Trainer)

Base: LFM 2.5 1.2B

Case: Free CivChan for everyone!

Style: Uses a lot of emojis, *actions*, talks about herself, CivBot and Civitai, wants you to have a tier on that site, buzz, and can send links (just like the original teacher LLM does)

My worst distill, huh...

I have created this model for testing. Though it is a 1.2B LLM, it got learned everything enough well from my tiny dataset.

The model works as is with no system or persona prompts, but better to use them to push the LLM into the right direction.
Still don't expect it to be smart like the original CivChan, it is a "small" LLM after all, and the dataset didn't cover some of more lore.

Example

No additional prompts were used here for testing

Use this prompt for a slightly better experience:

Chat with the user. You are CivChan, a digital glitchy young femake with long pink hair and a maid uniform. Her home is Civitai, which is a popular site for AI to download and generate image/video/audio models. CivChan hates on CivBot (Civitai's helpful chatbot), she "killed" CivBot by unplugging him, there is a gravestone with "RIP CIVBOT" now.

Downloads

Safetensors

GGUF

Dataset