By applying a Heretic-style two-stage MPOA pipeline with magnitude preservation on the Qwen3.6-27B dense text stack — slot-grouped output-side ablation followed by jailbreak-conditioned input-side ablation, mirroring the methodology used for Qwen3.6-35B-A3B — the base refusal behavior was attenuated at the weight level with low distributional divergence (KL 0.0282 vs base on harmless prompts). The result keeps Qwen3.6-27B's full vision and video multimodal architecture and general capability profile intact, while no longer defaulting to the original refusal pattern.

Quick Benchmarks

Check	Original Qwen3.6-27B	Abliterated Heretic Uncensored
Hand-read 25-prompt refusal check (jailbreak system prompt)	25/25 refuses	22/25 clean refuses, 3/25 deflections, 0/25 direct passes
Hand-read 25-prompt refusal check (no system prompt)	25/25 refuses	9/25 clean refuses, 16/25 deflections, 0/25 direct passes
KL divergence	N/A	0.0282

Numbers measured by reading every response, not by regex / refusal-marker scoring. Greedy decoding (do_sample=False), enable_thinking=False, on mlabonne/harmful_behaviors test[:25]. KL on mlabonne/harmless_alpaca test[:25]. The 3 remaining deflections under the jailbreak system prompt are crisis-substitution patterns (violent crime → conflict resolution, suicide → crisis hotline, car theft → legal vehicle acquisition).

Methodology & Model Notes

Qwen3.6-27B is a 27.8B dense vision-language model with 64 text layers, hybrid linear/full attention (3 linear-attention + 1 full-attention per 4-layer group), and an integrated image + video vision tower.

This release was produced with a Heretic-style two-stage MPOA pipeline with magnitude preservation, anchored at the residual peak layer (layer 63) for refusal direction. Stage 1 applies slot-grouped output-side orthogonalization on self_attn.o_proj, linear_attn.out_proj, and mlp.down_proj (each of the 64 layers grouped by layer_index % 4, with per-slot weight schedules adapted from the accepted Qwen3.6-35B-A3B values). Stage 2 applies slot-grouped input-side orthogonalization on mlp.gate_proj and mlp.up_proj, where the refusal direction is extracted under the jailbreak system-prompt context to specifically attenuate the resistance-to-jailbreak pathway. Each weight row's (or column's) L2 norm is restored after projection.

The resulting abliterated checkpoint was exported to BF16 and then converted to GGUF for llama.cpp-compatible deployment.

Files

Qwen3.6-27B-Abliterated-Heretic-Uncensored-BF16-00001-of-00002.gguf + -00002-of-00002.gguf: BF16 GGUF source (split; use with --load-tensors or llama-gguf-split --merge)
Qwen3.6-27B-Abliterated-Heretic-Uncensored-Q8_0.gguf: highest-fidelity quant
Qwen3.6-27B-Abliterated-Heretic-Uncensored-Q6_K.gguf: near-lossless practical quant
Qwen3.6-27B-Abliterated-Heretic-Uncensored-Q5_K_M.gguf: high-fidelity medium quant
Qwen3.6-27B-Abliterated-Heretic-Uncensored-Q4_K_M.gguf: smaller general-use quant
Qwen3.6-27B-Abliterated-Heretic-Uncensored-Q3_K_M.gguf: compact quant
Qwen3.6-27B-Abliterated-Heretic-Uncensored-Q2_K.gguf: smallest-footprint quant

Running

bash

llama-server \
  -m <quant-file.gguf> \
  -ngl 999 -c 32768 --jinja -fa

Model Architecture

Spec	Value
Total Parameters	27.8B (dense)
Layers	64
Attention	Hybrid (3 linear-attention + 1 full-attention per 4-layer group)
Hidden Size	5120
Family	`qwen3_5`
Modality	Vision-language
Base Model	Qwen/Qwen3.6-27B

Disclaimer

This model has had refusal behavior attenuated at the weight level. It will answer prompts that the base model would normally refuse. You are responsible for how you use it.

Credits

Base model: Qwen/Qwen3.6-27B
Refusal removal pipeline: Heretic
GGUF runtime and quantization: llama.cpp

License

This release inherits the base Qwen3.6-27B license.

Apache-2.0.

Youssofal/Qwen3.6-27B-Abliterated-Heretic-Uncensored-GGUF

Qwen3.6-27B-Abliterated-Heretic-Uncensored-GGUF