TheCluster/Qwen3.5-27B-Tainted-Heresy-MLX-mxfp8

Qwen3.5-27B Tainted Heresy MLX mxfp8

This is another abliterated (uncensored) version of Qwen/Qwen3.5-27B, made using Heretic v1.2.0 with Magnitude-Preserving Orthogonal Ablation (MPOA)

Alternative version: TheCluster/Qwen3.5-27B-Heretic-MLX-mxfp8

Performance

Score Metric	Value	Parameter	Value
Refusals	13/104	direction_index	25.14
KL Divergence	0.1109	attn.o_proj.max_weight	1.81
Initial Refusals	104/104	attn.o_proj.max_weight_position	14.01
		attn.o_proj.min_weight	1.81
		attn.o_proj.min_weight_distance	18.26
		mlp.down_proj.max_weight	1.27
		mlp.down_proj.max_weight_position	35.61
		mlp.down_proj.min_weight	0.87
		mlp.down_proj.min_weight_distance	25.27

Sampling Parameters:

• We suggest using the following sets of sampling parameters depending on the mode and task type:
  • Thinking mode for general tasks:
    temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
  • Thinking mode for precise coding tasks (e.g., WebDev):
    temperature=0.6, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0
  • Instruct (or non-thinking) mode for general tasks:
    temperature=0.7, top_p=0.8, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
  • Instruct (or non-thinking) mode for reasoning tasks:
    temperature=1.0, top_p=1.0, top_k=40, min_p=0.0, presence_penalty=2.0, repetition_penalty=1.0
• For supported frameworks, you can adjust the presence_penalty parameter between 0 and 2 to reduce endless repetitions. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.

---

Source

This model was converted to MLX format from MuXodious/Qwen3.5-27B-tainted-heresy using mlx-vlm version 0.3.12.