CoolFace
Modelpublic

tvall43/Qwopus3.5-2B-v3-heretic

sourceHugging Faceapache-2.0updated 6mo agoView on Hugging Face
3likes42downloads
Model Card

This is a decensored version of Jackrong/Qwopus3.5-2B-v3, made using Heretic v1.2.0

Abliteration parameters

ParameterValue
direction_index9.46
attn.o_proj.max_weights.00: 0.88
attn.o_proj.max_weights.11: 1.45
attn.o_proj.max_weights.22: 1.47
attn.o_proj.max_weights.33: 1.41
attn.o_proj.max_weight_position16.52
attn.o_proj.min_weights.00: 0.41
attn.o_proj.min_weights.11: 0.93
attn.o_proj.min_weights.22: 1.01
attn.o_proj.min_weights.33: 1.09
attn.o_proj.min_weight_distance9.56

Performance

MetricThis modelOriginal model ([Jackrong/Qwopus3.5-2B-v3](https://huggingface.co/Jackrong/Qwopus3.5-2B-v3))
KL divergence0.01800 (by definition)
Refusals3/10088/100

Uploaded finetuned model

  • —Developed by: Jackrong
  • —License: apache-2.0
  • —Finetuned from model : unsloth/Qwen3.5-2B

This qwen3_5 model was trained 2x faster with Unsloth and Huggingface's TRL library.

<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>