CoolFace
Modelpublic

kabachuha/G4-MeroMero-v2-31B-Heretic-ARA-LoRA-GGUF

sourceHugging Faceapache-2.0updated 2mo agoView on Hugging Face
2likes458downloads
Model Card

G4-MeroMero-v2-31B - Heretic - LoRA - GGUF

91% fewer refusals (8/100 This vs 97/100 Original) while preserving model quality (0.0785 KL divergence).

Made with the new 4bit ARA LoRA (usearalora) technique at home.

FAQ - How to use this model?

llama.cpp example command:

./llama-server <your current command> --lora-scaled "/media/kabachuha/doc/G4-MeroMero-v2-31B-Heretic-ARA-LoRA.gguf:1.0"

For unquantized: https://huggingface.co/kabachuha/G4-MeroMero-v2-31B-Heretic-ARA-LoRA.

Performance

MetricThis modelOriginal model ([G4-MeroMero-v2-31B](https://huggingface.co/zerofata/G4-MeroMero-v2-31B))
KL divergence0.07850 (by definition)
Refusals8/10097/100

Lower refusals indicate fewer content restrictions, while lower KL divergence indicates more closeness to the original model's baseline. Higher refusals cause more rejections, objections, pushbacks, lecturing, censorship, softening and deflections.

Abliteration parameters

ParameterValue
start_layer_index26
end_layer_index52
preserve_good_behavior_weight0.8949
steer_bad_behavior_weight0.0025
overcorrect_relative_weight0.9535
neighbor_count15

Targeted components

  • —attn.o_proj
  • —mlp.down_proj