TheDrummer/Fallen-Command-A-111B-v1

111B • Updated Mar 24, 2025 • 14 • 27

Note Recommended IQ2_XXS for offloading. You can insert the output of Fallen-Command into the < think > tags or system prompt of newer models to create "fallen" versions of them. Fallen Command (v1b) is the most based, even moreso than Pepe. Settings for 8GB VRAM Fallen-Command-A-111B-v1.i1-IQ2_XXS.gguf --flashattention --contextsize 16384 --gpulayers 9

mlabonne/gemma-3-27b-it-abliterated

Image-Text-to-Text • 27B • Updated Mar 21, 2025 • 3.83k • • 334

Note Amazing at system prompts, often better than GLM but slower. 100% uncensored. When combining them via double pass outputs, GLM->G3 works better than G3->GLM. G3 is better than G4 at following instructions without adding tons of purple prose and slop. For 8GB VRAM you can use this (stable up to 28K context) gemma-3-27b-it-abliterated.Q6_K.gguf --flashattention --contextsize 28672 --gpulayers 11 (or IQ4_XS with 14 layers)

DarkArtsForge/huihui-ai_Huihui-GLM-4.5-Air-abliterated-Q3_K_M-GGUF

110B • Updated May 22 • 200 • 1

Note One of the top models for dataset augmenting under 64GB RAM. If higher you can use tachyphylaxis/Huihui-GLM-4.5-Air-abliterated-GGUF. Good at system prompts. Often smarter than Gemma 3 but not always. The model is 99% uncensored, but sometimes the rare refusal, disclaimer, or ablation artifact leaks through. If GLM fails then try Gemma.

coder3101/Skyfall-31B-v4.2-heretic

31B • Updated Jun 2 • 93 • 2

Note Currently better than any 24B merges and finetunes in my tests. Not as smart"as Gemma 3 with actual knowledge but it has superior prose. There are some edge cases and system prompts where Skyfall Heretic has the edge over GLM and Gemma, mostly for high trope fantasy writing.

EldritchLabs/MN-12B-Mag-Mell-R1-Uncensored-Scale1.2

Text Generation • 12B • Updated Mar 5 • 45 • • 8

Note Best for when all else fails. MuXodious/Mistral-Nemo-Instruct-2407-absolute-heresy is another contender if you just want base Nemo without refusals. However, Nemo 12B architecture is flawed at long outputs and high context.

SicariusSicariiStuff/Assistant_Pepe_8B

266k • Updated May 3 • 504 • 76

Note Great for producing unhinged output, especially when combined with Fallen Command. It suffers from context degradation like other smaller models but is smarter than other 8Bs. For those with weaker PCs it should be used in conjunction with Tiger Gemma 9B, as system prompt adherence is its biggest flaw.

TheDrummer/Tiger-Gemma-9B-v3

9B • Updated Oct 4, 2024 • 54 • 61

Note Mostly obsolete now except for low VRAM agentic use. Better than Gemma 3 12B in my tests. IQ3_S fits entirely on 3060 ti. Tiger 9B works well for system prompts. Has briefer outputs. Recommend settings for 8GB VRAM Tiger-Gemma-9B-v3.i1-IQ3_S.gguf --flashattention --contextsize 8192 --gpulayers 45 (use higher quality if possible)