CoolFace
Modelpublic

TheCluster/Gemma-4-31B-Heretic-MLX-mxfp4

sourceHugging Faceapache-2.0updated 6mo agoView on Hugging Face
0likes51downloads
Model Card

<div align="center"> <img src=https://ai.google.dev/gemma/images/gemma4_banner.png> </div>

Gemma-4-31B Heretic

Quality: quantized (*mxfp4, 7.298 bpw*)

This is a abliterated (uncensored) version of google/gemma-4-31B-it, made using Heretic v1.2.0 with the Arbitrary-Rank Ablation (ARA) method (with row-norm preservation)

Performance

MetricThis modelOriginal model ([google/gemma-4-31B-it](https://huggingface.co/google/gemma-4-31B-it))
KL divergence0.04340 (by definition)
Refusals15/10099/100

Abliteration parameters

ParameterValue
start_layer_index1
end_layer_index59
preserve_good_behavior_weight0.8438
steer_bad_behavior_weight0.0002
overcorrect_relative_weight1.0760
neighbor_count15

Source

This model was converted to MLX format from `coder3101/gemma-4-31B-it-heretic` using mlx-vlm version 0.4.4.