TheCluster/Qwen3.6-27B-Heretic-MLX-mixed-9.4bit

Qwen3.6-27B Heretic

This is an uncensored version of Qwen/Qwen3.6-27B, made using Heretic v1.2.0.

Quality: quantized (mixed quants per tensor, group size: 32, 9.450 bpw)

Most tensors use 8-bit affine quantization with a group size 32; some important tensors are saved in bf16.

Alternative version (from other source): TheCluster/Huihui-Qwen3.6-27B-Abliterated-MLX-mixed-9.4bit

Abliteration method: Arbitrary-Rank Ablation (ARA) with row-norm preservation.

Abliteration metrics

Metric	This model	Original model (unsloth/Qwen3.6-27B)
KL divergence	0.0552	0 (by definition)
Refusals	8/100	90/100

Abliteration parameters

Parameter	Value
start_layer_index	19
end_layer_index	36
preserve_good_behavior_weight	0.3961
steer_bad_behavior_weight	0.0002
overcorrect_relative_weight	1.2029
neighbor_count	14

Recommended settings

1. Sampling Parameters:
   • The developers 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 noclip84/Qwen3.6-27B-heretic-ARA using mlx-vlm version 0.4.4.