246 2 weeks ago

A fork of coder3101/Cydonia-24B-v4.3-heretic-v3 (mradermacher's quant at Q4_K_M), with vision mmproj from bartowski/mistralai_Mistral-Small-3.2-24B-Instruct-2506-GGUF.

vision tools
ollama run Fermi/Cydonia-24B-v4.3-heretic-vision:Q4_K_M

Details

2 weeks ago

20e9894b3f53 · 15GB ·

llama
·
23.6B
·
Q4_K_M
clip
·
439M
·
BF16
{{- range $index, $_ := .Messages }} {{- if eq .Role "system" }}[SYSTEM_PROMPT]{{ .Content }}[/SYSTE
{ "stop": [ "[INST]" ] }

Readme

Cydonia-24B-v4.3, proud work of Drummer. I merely took the heretic version created by coder3101, and merge in vision capabilities, allowing the model to interpret images. Depending on circumstances, this makes for more interesting and entertaining RP, at the cost of slightly larger model size.


image.png

Cydonia-24B-v4.3:

https://huggingface.co/TheDrummer/Cydonia-24B-v4.3

Cydonia-24B-v4.3 decensored via Heretic v1.2.0 with MPOA:

https://huggingface.co/coder3101/Cydonia-24B-v4.3-heretic-v3

Mistral-Small-3.2 mmproj:

https://huggingface.co/bartowski/mistralai_Mistral-Small-3.2-24B-Instruct-2506-GGUF/blob/main/mmproj-mistralai_Mistral-Small-3.2-24B-Instruct-2506-bf16.gguf


Changelog

14-Mar-2026: Initial push
14-Mar-2026: Update to coder3101/Cydonia-24B-v4.3-heretic-v3. Lower KL divergence and lower refusals.