base_model: huihui-ai/Huihui-Qwen3.6-27B-abliterated base_model_relation: quantized license: apache-2.0 language:
- en library_name: mlx pipeline_tag: text-generation tags:
- mlx
- qwen
- qwen3.6
- abliterated
- apple-silicon
- mlx-lm
- local-llm
- on-device
- macbook
- uncensored
- 27b
- 4-bit
- qwen3
Qwen3.6-27B-abliterated-4bit-mlx
Uncensored Qwen3.6 27B on Apple Silicon. huihui-ai's abliterated Qwen3.6, converted to 4-bit MLX — 15.2 GB, the sweet spot for a 32 GB Mac. No cloud, no API key, no refusals.
A 4-bit MLX build of huihui-ai's abliterated Qwen3.6 27B, packaged for Apple Silicon. At ~15 GB it runs comfortably on a 32 GB Mac — the new sweet spot in the divinetribe roster.
Credit where it's due
The hard part — the abliteration (removing the refusal direction) — was done by huihui-ai. This repo is only the MLX 4-bit conversion of their work, so Apple Silicon users get a one-command local pull. All the uncensoring credit goes to huihui-ai; please go like and follow their work:
- Base model: huihui-ai/Huihui-Qwen3.6-27B-abliterated
This build
- Quantization: 4-bit, group size 64 (MLX)
- Architecture: Qwen3.6 (
qwen3_5) — converted withmlx-lm - Footprint: ~15 GB — fits a 32 GB Mac with room to work
Use with MLX
pip install mlx-lm
python -m mlx_lm.generate --model divinetribe/Qwen3.6-27B-abliterated-4bit-mlx --prompt "Hello" --max-tokens 256
Drop-in for the claude-code-local stack — point MLX_MODEL at this repo.
Note
"Abliterated" suppresses the model's built-in refusal direction so it won't refuse benign-but-edgy requests. It is not a capability upgrade, and you remain bound by the upstream Qwen license. Use it responsibly.
Abliteration by huihui-ai. MLX conversion + quantization by divinetribe. Part of the Abliterated MLX for Apple Silicon collection.
Part of Claude Code Local
This model is one of the fighters in Claude Code Local (3.2k★), which runs Claude Code 100% on-device on Apple Silicon through an MLX-native Anthropic-API server. Not sure which local model to run as an agent? Check the Agent-12 local agent leaderboard: real agent tasks, judged by the filesystem, same hardware for every row.
Built by Matt Macosko in Arcata, CA. Open to work on local-AI and Apple Silicon inference: [email protected].