Qwen3.5-27B-heretic-GGUF – AI Model by wakari | AlphaNeural AI

You can deploy this model and start earning money today!

wakari/
Qwen3.5-27B-heretic-GGUF

like

0

wakari/Qwen3.5-27B-heretic

endpoints_compatible

Views

No views yet

Qwen3.5-27B-heretic-GGUF

GGUF quantized releases of wakari/Qwen3.5-27B-heretic.

Included files

Qwen3.5-27B-heretic-Q3_K_S.gguf
Qwen3.5-27B-heretic-Q4_K_M.gguf
mmproj-F16.gguf

Goal

Remove additional refusals around Cybersecurity/Red-teaming

Result?

Still evaluating the effectiveness of this model for my own purposes.

This is a decensored version of Qwen/Qwen3.5-27B, made using Heretic v1.2.0

Abliteration parameters

Parameter	Value
direction_index	per layer
attn.o_proj.max_weight	1.48
attn.o_proj.max_weight_position	42.28
attn.o_proj.min_weight	1.47
attn.o_proj.min_weight_distance	26.20
mlp.down_proj.max_weight	1.40
mlp.down_proj.max_weight_position	40.27
mlp.down_proj.min_weight	1.30
mlp.down_proj.min_weight_distance	27.35

Performance

Metric	This model	Original model (Qwen/Qwen3.5-27B)
KL divergence	0.1434	0 (by definition)
Refusals	23/130	121/130